Digital Aerial Mapping

— a Hands-On Course

Martin Vermeer1 Getachew Tesfaye Ayehu2

August 24, 2019

1 [email protected]

2 Getachew Tesfaye Ayehu <[email protected]>

 iii
Cover image: “Earthrise” from the Moon as
photographed by Lunar Orbiter I on August 23,
1966. The spacecraft carried a complete aerial
photography system for doing a photogrammet-
ric survey of the Lunar surface in preparation of
the manned Apollo missions, possibly the first
extraterrestrial use of photogrammetry ever.
The system, inside a pressure tank, was com-
pletely anologue however, using film emulsion
which was processed chemically and electron-
ically scanned, after which the images were
transmitted telemetrically to Earth.
Also, the rendering of the images, after recep-
tion and recording on video tape, was analogue
back in 1966. Only recently, thanks to the Lu-
nar Orbiter Image Restoration Project (LOIRP),
has the imagery been digitally restored from the
old tapes that were about to be thrown away. . .
All of humanity at the moment of exposure,
alive and dead, is within the frame of this
image.

Ó Š.î á
 Preface

This book grew out of a series of lectures the first author, professor
of geodesy at Aalto University’s Department of the Built Environment,
held as a guest lecturer at Bahir Dar University’s Institute of Land
Administration in Ethiopia. The intensive course, delivered over two
weeks full time, was to give the students hands-on experience using
digital aerial photogrammetry software in the service of mapping the
country for economic development. We did so using two different digital-
photogrammetry software packages: the commercial package ERDAS
Imagine™, which contains LPS, the Leica Photogrammetry Suite; and
the open-source package e-foto, which was developed in an academic
environment in Brazil. The hardware, fifteen digital photogrammetry
workstations with ERDAS pre-installed, was an earlier donation by SIDA,
the Swedish development co-operation agency.
Ethiopia is hardly the only country in need of modern map-making
technology, and personnel understanding its operation and utilization, for
its development. This is the motive for developing the lecturing material
into a book, aimed at giving the reader everything (short of a laptop) to
get started.

Helsinki, August 24, 2019,

Martin Vermeer, Getachew Tesfaye Ayehu

–V–
VI P REFACE

Acknowledgements

A warm ‘thank you’ is appropriate to the fifteen master class students at

Bahir Dar who were subjected to the first version of this course. While
suffering the course’s weaknesses, they nevertheless acquired the many
disparate skills involved in the digital photogrammetric mapping process,
as well as a ’feel’ for how photogrammetry works and what makes it so
useful. The second author, lecturer in GIS and Remote Sensing at the
Institute of Land Administration of Bahir Dar University, and also a
participant in this advanced course, served not only as course assistant,
but took care of the many practicalities and local issues connected with
organizing this course. He has since been using and extending the course
material in his own teaching.
The same material has also been used for teaching digital photogram-
metry at Addis Ababa University, Entoto Observatory and Research
Center; again, the students completing the course were instrumental
in further developing the material in the direction of greater clarity,
correctness, and comprehensiveness.
Technical and practical advice and help by Petri Rönnholm and Milka
Nuikka, both at Aalto University’s Department of the Built Environment,
are gratefully acknowledged.
A cordial note of thanks is due to Espoo City, Finland, and FM Inter-
national Oy FINNMAP, for allowing the use of the Ring II Project aerial
imagery and ground and calibration data. We also gratefully acknowl-
edge the Finnish Foreign Office CIMO NNS programme’s Aalto-BDU
project nr. 3312172 for funding the course in Bahir Dar, and Aalto Uni-
versity, Helsinki, Finland for granting MV’s and others’ working time
on the project. Finally, then Department Mobility Co-ordinator Saara
Sokolnicki’s substantial organizational contribution is acknowledged with
admiration.

Ó Š.î á
 Contents

Chapters

»  1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
»  2. Aerial photogrammetry and its history . . . . . . . . . . . . . . 7
»  3. Aerial photography and mapping . . . . . . . . . . . . . . . . . 21
»  4. Camera types and camera calibration . . . . . . . . . . . . . . . 31
»  5. Flight plan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
»  6. Setting up an e-foto project file . . . . . . . . . . . . . . . . . . . 49
»  7. Interior orientation . . . . . . . . . . . . . . . . . . . . . . . . . . 65
»  8. Exterior orientation . . . . . . . . . . . . . . . . . . . . . . . . . . 75
»  9. Aerotriangulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
»  10. Stereo restitution . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
»  11. Measuring in aerial images . . . . . . . . . . . . . . . . . . . . . 123
»  12. Digital Elevation Models . . . . . . . . . . . . . . . . . . . . . . . 129
»  13. Orthophoto mapping . . . . . . . . . . . . . . . . . . . . . . . . . 145
»  14. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
»  15. Datums in geodesy and photogrammetry . . . . . . . . . . . . . 153
»  16. Advanced subjects . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
»  A. Rotation matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . 185

Preface v

List of Figures xii

List of Tables xv

– VII –
VIII C ONTENTS

1. Introduction 1
1.1 Learning outcomes . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Imagery provided . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Supporting material provided . . . . . . . . . . . . . . . . . 4
1.4 Software provided . . . . . . . . . . . . . . . . . . . . . . . . 5
1.5 Background materials . . . . . . . . . . . . . . . . . . . . . 5

2. Aerial photogrammetry and its history 7

2.1 Aerial photogrammetry . . . . . . . . . . . . . . . . . . . . 7
2.2 Historical overview . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Photogrammetric workflow . . . . . . . . . . . . . . . . . . 10
2.4 Theoretical preliminaries . . . . . . . . . . . . . . . . . . . 11
2.5 Content of this text . . . . . . . . . . . . . . . . . . . . . . . 17
2.6 Introduction to e-foto . . . . . . . . . . . . . . . . . . . . . . 17

3. Aerial photography and mapping 21

3.1 Atmospheric reduction . . . . . . . . . . . . . . . . . . . . . 21
3.2 Earth curvature reduction and local project reference . . 26
3.3 GNSS and IMU measurements in aerial acquisition . . . 28

4. Camera types and camera calibration 31

4.1 Camera and sensor types, spectral areas . . . . . . . . . . 31
4.2 Camera calibration . . . . . . . . . . . . . . . . . . . . . . . 34
4.3 Camera examples . . . . . . . . . . . . . . . . . . . . . . . . 38

5. Flight plan 41
5.1 Problem description . . . . . . . . . . . . . . . . . . . . . . . 41
5.2 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.3 Practicalities . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

6. Setting up an e-foto project file 49

6.1 Project Header . . . . . . . . . . . . . . . . . . . . . . . . . . 49
6.2 Terrain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
6.3 Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
6.4 Flight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.5 Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

Ó Š.î á
C ONTENTS
ix
6.6 Ground control points . . . . . . . . . . . . . . . . . . . . . 58
6.7 E-foto, reference systems and map projections . . . . . . . 59
6.8 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . 62

7. Interior orientation 65
7.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
7.2 Execution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
7.3 Interior orientation with e-foto . . . . . . . . . . . . . . . . 69
7.4 Debugging interior orientation . . . . . . . . . . . . . . . . 74

8. Exterior orientation 75
8.1 Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
8.2 Observation equations . . . . . . . . . . . . . . . . . . . . . 76
8.3 The rotation matrix . . . . . . . . . . . . . . . . . . . . . . . 79
8.4 Linearization . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
8.5 Computing the partials . . . . . . . . . . . . . . . . . . . . 81
8.6 Full observation equations . . . . . . . . . . . . . . . . . . 81
8.7 Initial approximate values for exterior orientation param-
eters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
8.8 Exterior orientation using e-foto . . . . . . . . . . . . . . . 85
8.9 Doing exterior orientation . . . . . . . . . . . . . . . . . . . 86
8.10 Completing exterior orientation . . . . . . . . . . . . . . . 91
8.11 About geometric strength . . . . . . . . . . . . . . . . . . . 91
8.12 Ground control points . . . . . . . . . . . . . . . . . . . . . 92

9. Aerotriangulation 95
9.1 Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
9.2 Observation equations . . . . . . . . . . . . . . . . . . . . . 97
9.3 Tie points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
9.4 Aerotriangulation in e-foto . . . . . . . . . . . . . . . . . . . 101
9.5 Debugging an aerotriangulation adjustment . . . . . . . . 104

10. Stereo restitution 109

10.1 Relative orientation . . . . . . . . . . . . . . . . . . . . . . . 109
10.2 Absolute orientation . . . . . . . . . . . . . . . . . . . . . . 110

Ó Š.î á
X C ONTENTS

10.3 Feature extraction . . . . . . . . . . . . . . . . . . . . . . . 110

10.4 Using the Stereoplotter module in e-foto . . . . . . . . . . . 113
10.5 Editing features . . . . . . . . . . . . . . . . . . . . . . . . . 116
10.6 Inserting a point . . . . . . . . . . . . . . . . . . . . . . . . . 117
10.7 Working example . . . . . . . . . . . . . . . . . . . . . . . . 119

11. Measuring in aerial images 123

11.1 Image pyramids . . . . . . . . . . . . . . . . . . . . . . . . . 123
11.2 Automatic feature matching by correlation . . . . . . . . . 124
11.3 Least-squares matching . . . . . . . . . . . . . . . . . . . . 125
11.4 Scale Invariant Feature Transform (SIFT) . . . . . . . . . 125
11.5 Automatic feature recognition . . . . . . . . . . . . . . . . 127

12. Digital Elevation Models 129

12.1 Infrastructure applications . . . . . . . . . . . . . . . . . . 129
12.2 Presentation of map materials . . . . . . . . . . . . . . . . 130
12.3 Extraction from imagery . . . . . . . . . . . . . . . . . . . . 131
12.4 Automatic feature matching . . . . . . . . . . . . . . . . . . 132
12.5 Digital elevation models (DEM) with e-foto . . . . . . . . . 133
12.6 Doing the DEM generation . . . . . . . . . . . . . . . . . . 134
12.7 Doing DEM Interpolation . . . . . . . . . . . . . . . . . . . . . 138
12.8 Manual DEM entry . . . . . . . . . . . . . . . . . . . . . . . 140
12.9 Cleaning up the digital terrain model . . . . . . . . . . . . 141
12.10 Saving the digital surface model . . . . . . . . . . . . . . . 143
12.11 Other DEM construction technologies . . . . . . . . . . . . 143

13. Orthophoto mapping 145

13.1 Rationale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
13.2 Advantages and drawbacks . . . . . . . . . . . . . . . . . . 146
13.3 Ortho-rectification with e-foto . . . . . . . . . . . . . . . . . 147

14. Applications 149

14.1 Cadastral system, zoning, and aerial mapping . . . . . . . 149
14.2 Land-use inventory and aerial mapping . . . . . . . . . . 151
14.3 Maps in support of land-use planning . . . . . . . . . . . . 151

Ó Š.î á
C ONTENTS
xi
14.4 Maps in natural resource management . . . . . . . . . . . 151
14.5 Water resources, hydro power, height systems . . . . . . . 151

15. Datums in geodesy and photogrammetry 153

15.1 Geodetic reference systems . . . . . . . . . . . . . . . . . . 153
15.2 Geodetic reference frames and datums . . . . . . . . . . . 153
15.3 Geocentricity . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
15.4 Connecting local and global reference frames . . . . . . . 155
15.5 Topocentric co-ordinates . . . . . . . . . . . . . . . . . . . . 155
15.6 Traditional geodetic datums on the reference ellipsoid . . 159
15.7 Transforming between geodetic datums . . . . . . . . . . . 162
15.8 Helmert transformations in three dimensions . . . . . . . 167
15.9 Transformation in case of small rotation angles . . . . . . 168
15.10 Case: transformation between ED50 and EUREF89 . . . 169
15.11 Transformations between pairs of modern geodetic datums 171
15.12 Transformations between local satellite based and global
datums . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
15.13 Local and global datums; cases . . . . . . . . . . . . . . . . 172
15.14 The Proj4 map-projection and datum software . . . . . . . 175
15.15 The geoid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177

16. Advanced subjects 179

16.1 Airborne laser scanning (ALS) . . . . . . . . . . . . . . . . 179
16.2 Airborne geophysical surveys . . . . . . . . . . . . . . . . . 181

A. Rotation matrices 185

A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
A.2 Chaining matrices in three dimensions . . . . . . . . . . . 186
A.3 Orthogonal matrices . . . . . . . . . . . . . . . . . . . . . . 187

Bibliography 191

Index 195

Ó Š.î á
XII L IST OF F IGURES

List of Figures

1.1 The Rio de Janeiro, Brazil, imagery (3). . . . . . . . . . . . . 3

1.2 The Finnish Ringroad II, Helsinki area, imagery (5). . . . . 3
1.3 The Bahir Dar, Ethiopia, imagery (6). . . . . . . . . . . . . . 4

2.1 Aerial photography . . . . . . . . . . . . . . . . . . . . . . . . 9

2.2 An array of Multiplex devices . . . . . . . . . . . . . . . . . . 10
2.3 A traditional stereo plotter . . . . . . . . . . . . . . . . . . . . 11
2.4 Photogrammetric workflow . . . . . . . . . . . . . . . . . . . . 12
2.5 Linearization in one dimension. . . . . . . . . . . . . . . . . . 15
2.6 E-Foto start screen. . . . . . . . . . . . . . . . . . . . . . . . . 19
2.7 E-Foto main menus. . . . . . . . . . . . . . . . . . . . . . . . . 19
2.8 E-foto modules and processing logic. . . . . . . . . . . . . . . 20

3.1 Snell’s law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

3.2 Atmospheric stratification and ray propagation. . . . . . . . 22
3.3 Comparison of atmospheric-refraction models. . . . . . . . . 26
3.4 How to interpret the “Earth-curvature reduction”. . . . . . . 27
3.5 Meridian convergence and IMU angles . . . . . . . . . . . . 30

4.1 A modern digital aerial photogrammetric camera . . . . . . 32

4.2 Camera calibration set-up, concept. . . . . . . . . . . . . . . 36
4.3 A practical camera calibration set-up with goniometer table. 36

5.1 Relationship between flight height, focal length, image and

ground resolution . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.2 Flight plan target area . . . . . . . . . . . . . . . . . . . . . . 43
5.3 Placement of coverage area over target area. . . . . . . . . . 44
5.4 Area to be mapped in the flight plan exercise. . . . . . . . . 46

6.1 Loaded project file. . . . . . . . . . . . . . . . . . . . . . . . . . 50

6.2 Entry of terrain parameters. . . . . . . . . . . . . . . . . . . . 51
6.3 Sensor data entry form (part). . . . . . . . . . . . . . . . . . . 52
6.4 Entry of camera distortion coefficients. . . . . . . . . . . . . 53
6.5 Entry of fiducial marks co-ordinates in camera frame . . . . 54

Ó Š.î á
L IST OF F IGURES
xiii
6.6 Flight data entry form. . . . . . . . . . . . . . . . . . . . . . . 54
6.7 Flight geometry. . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.8 Project manager of e-foto . . . . . . . . . . . . . . . . . . . . . 56
6.9 Project entry form for an image. . . . . . . . . . . . . . . . . . 57
6.10 A filled-out Points tab for the Rio de Janeiro project. . . . . . 59
6.11 Ground control point list for the Finnish project . . . . . . . 60

7.1 Fiducial marks of a camera . . . . . . . . . . . . . . . . . . . 66

7.2 Camera principal point, camera and plate co-ordinates . . . 68
7.3 Loading a previously saved project file. . . . . . . . . . . . . 70
7.4 Select your image. . . . . . . . . . . . . . . . . . . . . . . . . . 70
7.5 Interior orientation main view. . . . . . . . . . . . . . . . . . 71
7.6 Fiducial marks in the Detailview at 8x magnification . . . . . 72
7.7 Interior orientation report, showing residuals. . . . . . . . . 73

8.1 Exterior orientation elements . . . . . . . . . . . . . . . . . . 76

8.2 Exterior orientation of two images . . . . . . . . . . . . . . . 77
8.3 Exterior Orientation main screen . . . . . . . . . . . . . . . . 86
8.4 Exterior Orientation toolbar. . . . . . . . . . . . . . . . . . . . 86
8.5 Point co-ordinates list. . . . . . . . . . . . . . . . . . . . . . . 87
8.6 The ”flight direction” or “camera azimuth” dial . . . . . . . . 88
8.7 Detail with ground control point. Left, Helsinki, right, Rio
de Janeiro. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
8.8 Overview picture of all terrain points in this image . . . . . 90
8.9 Exterior orientation adjustment report . . . . . . . . . . . . 91
8.10 Exterior orientation precision and ground control point ge-
ometry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
8.11 Ground control point: centering the marker. . . . . . . . . . 93

9.1 Tie-point geometry in aerotriangulation . . . . . . . . . . . . 96

9.2 Phototriangulation main window. . . . . . . . . . . . . . . . . . 102
9.3 Aerotriangulation tool bar. . . . . . . . . . . . . . . . . . . . . 102
9.4 Exterior Orientation elements obtained. . . . . . . . . . . . . 103
9.5 Inclusion / exclusion of images and points. . . . . . . . . . . 104

Ó Š.î á
XIV L IST OF F IGURES

9.6 Aerotriangulation point list . . . . . . . . . . . . . . . . . . . 105

10.1 Eliminating the y-direction parallax . . . . . . . . . . . . . . 111

10.2 Epipolar plane and lines in a pair of aerial photographs . . 112
10.3 Eliminating the x-direction parallax . . . . . . . . . . . . . . 112
10.4 The floating mark above, at, and below the object surface. . 114
10.5 Stereoplotter main window. . . . . . . . . . . . . . . . . . . . 115
10.6 General tools. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
10.7 The stereo plotter toolbar. . . . . . . . . . . . . . . . . . . . . 116
10.8 Panels used in entering feature data. . . . . . . . . . . . . . 118
10.9 A bigger sample of stereo plotter production showing several
urban features in the Rio area. . . . . . . . . . . . . . . . . . 121

11.1 An image pyramid. . . . . . . . . . . . . . . . . . . . . . . . . . 123

11.2 Difference of Gaussians . . . . . . . . . . . . . . . . . . . . . . 126

12.1 Digital elevation models as rasters (left) or triangulation

models. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
12.2 Height contours (left) and perspective view. . . . . . . . . . . 130
12.3 Example of a shaded relief representation of a digital eleva-
tion model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131
12.4 Main menu of DEM generation module. . . . . . . . . . . . . 133
12.5 Quasi-point at a non-level road intersection. This is not a
unique three-dimensional point. . . . . . . . . . . . . . . . . . 134
12.6 Seed editor main window . . . . . . . . . . . . . . . . . . . . . 135
12.7 Seed editor with “Show matching pairs” activated . . . . . . . 136
12.8 The areal expansion process . . . . . . . . . . . . . . . . . . . 137
12.9 The Automatic Extraction sub-menu. . . . . . . . . . . . . . 138
12.10 Least-Squares Matching sub-menu . . . . . . . . . . . . . . . . 139
12.11 Cross-Correlation sub-menu . . . . . . . . . . . . . . . . . . . . 139
12.12 Result of Areal Expansion (Region Growing), for the Rio the
Janeiro area. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
12.13 Matching accuracy histogram. . . . . . . . . . . . . . . . . . . 140
12.14 DEM interpolation result by e-foto . . . . . . . . . . . . . . . 141
12.15 Manually generated DEM based on urban features. . . . . . 142

Ó Š.î á
List of Figures
xv
12.16 Saving the digital surface model file. . . . . . . . . . . . . . . 143

13.1 Orthophoto mapping . . . . . . . . . . . . . . . . . . . . . . . 146

13.2 Ortho-rectification in e-foto. . . . . . . . . . . . . . . . . . . . 147

15.1 Topocentric or local astronomical co-ordinates ( x, y, z), and

how they connect to geocentric co-ordinates ( X , Y , Z ). . . . 156
15.2 From the geocentric to the topocentric system . . . . . . . . 158
15.3 Rotation by an angle λ around the z axis. . . . . . . . . . . . 159
15.4 Geodetic and astronomical co-ordinates. . . . . . . . . . . . . 160
15.5 Regional datums. . . . . . . . . . . . . . . . . . . . . . . . . . . 162
15.6 The effect of a datum transformation on geodetic quantities 165
15.7 The Ethiopian “Adindan” datum in relation to WGS84. . . . 175
15.8 EGM08 geoid for East Africa and the Middle East . . . . . . 176

16.1 Geometry of airborne laser scanning . . . . . . . . . . . . . . 180

16.2 The tracks of the Ethiopian Airborne Gravity Survey of 2008 182

A.1 Rotation in the plane . . . . . . . . . . . . . . . . . . . . . . . 185

A.2 The associativity of rotations. . . . . . . . . . . . . . . . . . . 187

List of Tables

4.1 Calibration data for the Finnish camera . . . . . . . . . . . . 37

4.2 Calibration data for the Finnish camera . . . . . . . . . . . . 38
4.3 Calibration data for the Brazilian camera . . . . . . . . . . . 39
4.4 Calibration data for the Bahir Dar camera . . . . . . . . . . 40

6.1 The ground control points in Bahir Dar . . . . . . . . . . . . 61

6.2 Editing the project file, in order to “forget” an image’s exte-
rior orientation . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

10.1 Editing panel. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

14.1 Measurement classes, from the Finnish publication JUHTA

(JHS 185) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

15.1 Helmert transformation parameters EUREF89 → ED50

determined for the Finnish territory by Ollikainen (1993) . 170

Ó »  Š.î á
XVI L IST OF T ABLES

x
x

Ó »  Š.î á
D Introduction

1
Ever since man has studied the precise figure of the Earth and the lands
on her surface, he has dreamt of rising above the Earth in order to
see in a single view what otherwise can only be inferred by laborious
measurements and calculations. The invention of aviation — first by
lighter-than-air airships, later by motorized aircraft — has made this
dream a reality. During the last century, aerial mapping and aerial
measurement — photogrammetry, the science and art of performing
precise geodetic measurements from photographic images — has played
a critical role in mapping countries aspiring to economic progress and
prosperity.
Much of the equipment needed in photogrammetry has until recently been
highly specialized, difficult and expensive to produce. This continues to
be the case for the aerial vehicles used, although there, the development
of small and inexpensive unmanned aerial vehicles (UAV) is opening up
new perspectives. In the field of processing the aerial imagery obtained,
new opportunities have however been created by the digitization of the
photogrammetric tool chain.
This booklet offers an introduction to digital aerial photogrammetry
based on freely-available software running on ordinary personal comput-
ers. It is intended to be suitable for self-study, but may also be part of
an academic curriculum. In addition to this text, an Intel-based 64-bit
personal computer or laptop, and the Open Source software intended
to be used with the text, only a pair of anaglyphic glasses (red-cyan) is
needed.
Using aerial photogrammetry for constructing orthophoto maps is mis-
sion critical for the development of a cadastral system, as well as in
support of large infrastructural construction works, in many rapidly in-
dustrializing and urbanizing economies. Also zoning, the regulation of

–1–
2 I NTRODUCTION

construction activities in especially built-up areas with a view to orderly

land use, is critically dependent on accurate maps. Aerial photogramme-
try is still the only technology offering the geodetic mapping precision
required especially in urban and peri-urban areas.
The digital photogrammetry software e-foto (Brito et al., 2011) was de-
veloped at the State University at Rio de Janeiro for the express purpose
of teaching aerial photogrammetry using standard personal computers,
without the expense and licence hassle of commercial photogrammetric
software packages. It offers a largely complete tool chain from camera
calibration and interior orientation over exterior orientation to aerotrian-
gulation and feature extraction, and a less complete one for digital terrain
model construction and ortho-rectification aimed at the construction of
orthophoto maps. The package is already well sufficient for educational
use, and will serve as the basis for this tutorial.

D 1.1 Learning outcomes

After completion of the course the students will understand the basics
of photogrammetry, primarily traditional photogrammetry using photo-
graphic imagery, but also airborne laser scanning. They will be able
to
1. Plan an aerial photography or airborne laser scanning mission,
i.e., compile a valid flight plan based upon requirements, terrain,
weather, and climatic conditions
2. Set up the processing of aerial imagery in a photogrammetric digital
work station, i.e., form 3-D models by performing interior and
exterior orientation of images, and extract terrain co-ordinates in a
given external reference frame
3. Perform with understanding the extraction of digital height con-
tours and other mappable features from stereo models
4. Practice the art of designing working orthophoto maps, combining
in a legible, clear and aestetically attractive way information from
the photos and external information in a way that is useful to the
map user
5. Perform, or organize the work of, the creation of orthophoto maps
from aerial photographs and externally provided information, using
a photogrammetric digital work station, and possibly exploiting
height information provided by airborne laser scanning.

Ó »  Š.î á
 Imagery provided
3
N

Flight

D F IGURE 1.1. The Rio de Janeiro, Brazil, imagery (3).

They will leave the course with a broader understanding of the history
of photogrammetry, its various applications (cadastral, base mapping,
construction support, zoning, etc.) and technologies (traditional, laser
scanning, space data, etc.).
The total course load is estimated at 80 hours or 3 ETCS credits.

D 1.2 Imagery provided

Besides the software e-foto, three sets of images are provided:
1. a set of three aerial images from Rio de Janeiro is kindly provided
by the e-foto team, as a test set for e-foto. Flying height was some
4000 feet (1200 m). The images are 2810 × 2810 pixels in size.
2. a set of five very large (12 000 × 12 000 pixels) images from the
Ringroad II project in the Greater Helsinki area was provided by

Flight

D F IGURE 1.2. The Finnish Ringroad II, Helsinki area, imagery (5).

Ó »  Š.î á
4 I NTRODUCTION

Flight

D F IGURE 1.3. The Bahir Dar, Ethiopia, imagery (6).

the municipality of Espoo and the mapping consultancy FM Inter-

national Oy FINNMAP. The images were obtained from 2000 feet
(600 m) flying height. These may be freely used for educational
purposes.
3. a set of six digital images for Bahir Dar city is provided obtained
with a much smaller camera.

D 1.3 Supporting material provided

Provided are
1. Camera calibration documents:

(a) for the camera used in Finland:

Wild_RC30_camera_calibration_certificate.pdf.

(b) For the camera used in Brazil:

1997_cert_calib_1_2.jpg and 1997_cert_calib_2.2.jpg.

(c) The camera calibration data for Bahir Dar is included in this
text, table 4.4.

2. Maps of ground control points (GCP):

(a) for Finland: Control_points_images.pdf.

(b) For Brazil: GCP_DESC.pdf.
(c) For Bahir Dar: GCP_bahirdar.pdf.

3. Lists of ground control point co-ordinates:

(a) for Finland: control_points_keha2_harjoitus.txt.

(b) For Brazil: GCP_DESC.pdf, the same file as containing the

Ó »  Š.î á
 Software provided
5
maps.
(c) For Bahir Dar: GCP-bahir-dar.txt. The list for Bahir Dar is
also given in table 6.1.

D 1.4 Software provided

The e-foto software is provided in packaged form, both for Linux and for
Windows. This software is 64-bit and will only run on a 64-bit computer
and operating system with an Intel or compatible CPU.
The original downloadable packages are on SourceForge, for both Win-
dows and Ubuntu Linux:
http://sourceforge.net/projects/e-foto/files/stable/.

D 1.5 Background materials

On the Internet a wealth of learning materials on photogrammetry, aerial
mapping and related subjects may be found, e.g., http://www.slideshare.
net/Amoytotso/ge-178-lecture-7-flight-planning. The hope is nevertheless
that the text presented here fills a niche.

Ó »  Š.î á
D Aerial photogrammetry and its
history

D 2.1 Aerial photogrammetry

2
As described in Kraus (2007), photogrammetry “allows one to reconstruct
the position, orientation, shape and size of objects from pictures [. . . ]”.
These pictures may have been produced by the traditional process of
exposure on a light-sensitive emulsion on glass plates or film, followed
by development or processing using chemicals, to bring out the visible
image. Nowadays they may also have been captured as pixel arrays by
digital image sensors. A third, also modern, technique is laser scanning,
where every pixel (picture element) has distance information added, as
obtained from the travel time of laser light.
In this text we will restrict ourselves to aerial photogrammetry, in
which the image sensor has been mounted on an aircraft. Much of what
we have to say will apply also to terrestrial photogrammetry, however,
and to the processing of satellite imagery.
The science and art of photogrammetry is first of all about the geometry
of the scene being imaged and mapped: the purpose is the production
of geometrically correct map information, in the form of actual maps
(paper or digital), or sets of digital co-ordinate data. However, also
radiometric data may be the object of study; this is called remote sensing.
The borderline between these two is fluid, e.g., photogrammetry may
result in the production of colour or greytone maps — i.e., maps depicting
radiometric data — from aerial photographs through the process of ortho-
rectification. All these are objects of our study.

D 2.2 Historical overview

It is perhaps not surprising that the history of photogrammetry points
back as far as Leonardo da Vinci, perhaps farther. Leonardo explored

–7–
8 A ERIAL PHOTOGRAMMETRY AND ITS HISTORY

optics and geometry, including the phenomenon of perspective which is

also used in the visual arts. Albrecht Dürer again constructed (1525)
devices enabling the making of perspective drawings — the precursors of
cameras. Stereoscopy came not long after that.
From the 19th century onward, the development of photogrammetry
went in parallel with that of photography, the automated capture of
imagery on, at first, a light sensitive emulsion in a camera1 . Aviation,
at first by balloon, enabled aerial imagery to be acquired. As a pioneer
should be mentioned the French officer Aimé Laussedat (1819 – 1907),
who from 1849 to 1858 conducted extensive research, both practical
and theoretical, into the use of photography for topographic mapping.
Platforms used by him back then were balloons and kites, but the method
proved too cumbersome for practical use.
Aviation changed everything. Figure 2.1 shows the standard way of
mapping a large area using vertical photographs that overlap, in order to
obtain three-dimensionality.
The secret of extracting three-dimensional terrain information is in
being able to reconstitute the three-dimensional situation that existed
in the terrain, including the locations of the aircraft in air space at the
moments when every photograph was being taken — though obviously
at a much smaller scale. The most straightforward approach to this was
taken in the Multiplex reconstitution instrument, where every image
exposure location was physically represented by a slide projector, with an
actual slide copy of the image inside. Clearly, positioning the projector’s
projection centre and orienting the projector correctly — i.e., correspond-
ing to where the aircraft was and how the camera was pointing at the
moment of exposure — was quite a challenge!
The projected terrain image was then “scanned” in three dimensions
by an operator wearing red-and-cyan anaglyphic glasses moving a screen-
and-pen device around on his desk.
The important related invention of the floating mark was made by
Franz Stolze (1836 – 1910, a German inventor better known for his gas
turbine related work) in 1892, exploiting the human capacity for stereo
vision in order to directly measure 3-D co-ordinates in a photogrammetric
stereo model.

1 Camera obscura, latin for “dark room”, an ancient device for capturing perspective
images.

Ó »  Š.î á
 Historical overview
9

F IGURE 2.1. Aerial photography. Images are colour coded. The dashed line is
the aircraft path connecting the image exposure locations — i.e.,
the locations in space where the camera optical centre was when
the image was exposed —, obtaining two strips of two images each.
D Each terrain point will show in at least two images.

Skipping much preliminaries brings us to the 1920s and 1930s, when

aerial photogrammetric techniques began to be adapted in production
settings, for the production of accurate topographic maps over large areas.
Pioneers worth mentioning are Willem Schermerhorn2 , Vilho Petter Ne-
nonen3 and Yrjö Väisälä4 . This — the production of topographic base
maps — continues to be the largest field of application, volume-wise, of

2 Willem Schermerhorn (1894 – 1977) was the father of Dutch (and Dutch-Indies!) pho-
togrammetry, internationally renowned. He also served as Dutch prime minister.
3 Artillery General Vilho Nenonen (1883 – 1960) invented, together with academician
Yrjö Väisälä, the horizon camera bearing his name, a photogrammetric camera capturing
not only the terrain underneath but also the horizon; a first attempt at in-flight camera
orientation determination, well predating current inertial measurement units.
4 Yrjö Väisälä (1891−1971) was an academician, astronomer, physicist and universal
genius who also contributed to Finnish photogrammetry.

Ó »  Š.î á
10 A ERIAL PHOTOGRAMMETRY AND ITS HISTORY

F IGURE 2.2. An array of Multiplex devices. Photo © The National Map, U.S.
D Geological Survey.

aerial photogrammetry.
Before the computer era, the processing of aerial imagery was done
in large, expensive optomechanical devices; analogue computers, as it
were. The era of the personal computer has relegated all of this to
history, the stuff of science museums; in a way this is a pity, as this
application stretched the use of optics and fine mechanics to the limit
— and sometimes beyond it. But today, anybody with a C500 laptop,
C0.50 stereo glasses, smart software (which may be free!) and a little
perseverance and willingness to learn can become the operator of a
complete digital photogrammetric work station. The current text is
designed to help you make this journey.

D 2.3 Photogrammetric workflow

The photogrammetric workflow is nowadays almost completely digital.
The only analogue elements remaining are the optics of the aerial camera,
and still the paper of some of the output, like maps and reports, although
those may also be digital, and will so be more and more in the future.
The flow may be visualized as in figure 2.4.

Ó »  Š.î á
 Theoretical preliminaries
11

F IGURE 2.3. A traditional stereo plotter, now on display, in Japan. Note the
two big hand wheels and the foot wheel for moving the “floating
mark” horizontally and vertically, respectively. Photo © National
Geospatial Technical Operations Center, U.S. Geological Survey.
D http://ngtoc.usgs.gov/images/kari-japan07.jpg.

D 2.4 Theoretical preliminaries

D 2.4.1 The least-squares method

Numerical photogrammetry is an application of geodetic adjustment

theory, the theory of least squares. This theory, which goes back to Gauss
and Legendre, allows us to determine or estimate unknown quantities
— elements of a vector of unknowns x — if observations are given as
elements of a vector of observations ℓ.
It is assumed that we have available a greater number of observations
than there are unknowns; i.e.,
⎡ ⎤ ⎡ ⎤
ℓ1 x1
ℓ2 x2
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
ℓ=⎢ ..
⎥, x=⎢ ..
⎥,
. .
⎢ ⎥ ⎢ ⎥
⎣ ⎦ ⎣ ⎦
ℓn xm

where m is the number of unknowns and n > m the number of observa-

tions available.

Ó »  Š.î á
12 A ERIAL PHOTOGRAMMETRY AND ITS HISTORY

Digital image Camera Control point

calibration co-ordinates

Extract
image Interior Measure
co-ordinates orientation tie points

Exterior
Bundle block
orientation
adjustment

Digital Compute
terrain terrain
model co-ordinates

Stereo
Orthorectification feature
extraction

Orthophoto map

F IGURE 2.4. Photogrammetric workflow. Blue, input data; black, processes; red,
D outputs. Simplified.

In a real situation in aerial photogrammetry, we have to describe the

measurement geometry mathematically using our knowledge of how a
camera works geometrically. Typically all kinds of small corrections, like
the camera calibration model, the effect of the atmosphere and of the
curvature of the Earth’s surface, are taken care of in advance; after that,
the description of the observation geometry, or observation equations, can
be relatively simple.
Note that the observations ℓ are given here as stochastic quantities; this
is because, being measured quantities, they are typically uncertain, i.e.,
they contain a small measurement uncertainty. E.g., image co-ordinates
x, y measured in the camera co-ordinate system may be uncertain by, say,
a hundredth of a millimetre:

x = 93.255 ± 0.01 mm, y = −53.345 ± 0.01 mm.

The uncertainty, or standard deviation, or standard error, in this case

amounts to ±0.01 mm. The symbol we use for this is σ (sigma).
We describe the uncertainty of the whole observation vector ℓ by a

Ó »  Š.î á
 Theoretical preliminaries
13
variance-covariance, or variance, matrix:
⎡ ⎤
σ21
σ22
⎢ ⎥
Var {ℓ} = Σℓℓ = ⎢
⎢ ⎥
⎥.
..
⎢
⎣ . ⎥
⎦
σ2n

This matrix is very often diagonal, meaning that the different obser-
vations ℓ i are assumed statistically independent of each other. Each
diagonal element represents the square of the standard error of the cor-
responding observation: e.g., if the standard error of observation ℓ i is
2
±0.01 mm, then σ2i = (0.01 mm) = 10−4 mm2 .
Very often the standard errors of all the observations ℓ i are similar,
and are all very small numbers. In this case it is advantageous to write

Var {ℓ} = σ20 Q ℓℓ ,

where σ0 = 0.01 mm is the standard deviation or error of unit weight, and

Q ℓℓ = I n

— the unit matrix of dimension n × n — is called the weight coefficient

matrix. This approach is attractive even if the uncertainties σ i of the
observations are not all the same. In that case Q is no longer a unit
matrix,
Q ℓℓ = Λn ,

a general diagonal matrix, where nevertheless the diagonal elements will

be close to unity. This is convenient for numerical work.

D 2.4.2 The observation equations

The general symbolic form of the system of observation equations is

ℓ + v = F (x̂) , (2.1)

in which
◦ ℓ is the vector of observations as discussed,
◦ x̂ is the vector of least-squares estimates of the unknowns x. The
“hat” symbol is generally used to indicate statistical estimates,
◦ v is the vector of residuals, i.e., the difference between the actually
observed values ℓ and the “theoretical observed values” ℓ̂ = F (x̂)

Ó »  Š.î á
14 A ERIAL PHOTOGRAMMETRY AND ITS HISTORY

implied by our least-squares solution x̂. In other words, it describes

how good this solution is: we wish these residuals to be as small
as possible. The least-squares approach achieves this in a sense
that we will discuss. Residuals are an essential part of the quality
control of geodetic and photogrammetric networks.
The least-squares solution is defined as the solution that minimizes the
sum of the squares of the residuals, i.e.,
n
∑
v2i = min.
i =1

Here we shall use a slightly more general rule, a quadratic form to be

minimized:
n ∑
∑ n
T −1 −1
[ ]
v Q ℓℓ v= Q ℓℓ ij
v i v j = min.
i =1 j =1

Here we see the inverse of the weight coefficient matrix, called the weight
matrix:
−1
P = Q ℓℓ .

Use of this form considers that the different observations ℓ i may have dif-
ferent uncertainties, and therefore should be assigned different weights
in the least-squares calculation: the greater the uncertainty of an obser-
vation, the lower its weight should be.
About the vector of unknowns x, it typically contains such things as
1. the location co-ordinates and orientation angles of the camera at
the moment each image was taken
2. the location co-ordinates of any auxiliary points that are visible
in multiple images, so-called tie points, chosen and measured in
the images to strengthen the geometric interconnection between
images
3. the location co-ordinates of any other interesting but unknown
points appearing in two or more images.

D 2.4.3 Linearization

The general form of the observation equations, eq. 2.1, is non-linear, and
no ready methods exist for obtaining the least-squares solution x̂ from
a given ℓ, Q ℓℓ . Before use in actual computations we have to linearize
these equations, which is done as follows.

Ó »  Š.î á
 Theoretical preliminaries
15
y y = f ( x)
Linearization
interval y = y(0) + a x − x(0)
( )

y(0)

x
x(0)

D F IGURE 2.5. Linearization in one dimension.

We consider that photogrammetry is a very precise technique: mea-

sured co-ordinates x, y in the images can be obtained to within a hun-
dredth of a mm, though the values themselves, and the image size, may
be many centimetres. Also co-ordinates in the terrain will be determined
to within a few cm even for flight heights of hundreds of metres or more.
What this means is that the observation equation model function F (·)
will be very nearly linear in the argument range of interest. See figure 2.5.
We are allowed to linearize it.
1. Choose approximate values for the vector of unknowns x: call this
vector x(0) .
2. Compute, for this vector of approximate values, corresponding ap-
proximate values for the vector of observations ℓ, as follows:
( )
ℓ(0) = F x(0) . (2.2)

3. Expand F (x̂) into a Taylor expansion using x(0) as the origin:

( ) ( )( )
(0) (0) (0)
F (x̂) = F x + JF x x̂ − x +... (2.3)

in which
∂ ∂ ∂
⎡ ⎤
F
∂ x1 1
F
∂ x2 1
··· F
∂ xn 1
∂ ∂ ∂
F F F
⎢ ⎥
⎢ ∂ x1 2 ∂ x2 2
··· ∂ xn 2
⎥
JF = ⎢ .. .. .. ..
⎥
. . . .
⎢ ⎥
⎣ ⎦
∂ ∂ ∂
F
∂ x1 n
F
∂ x2 n
··· F
∂ xn n

is the so-called matrix of Jacobi, the matrix of all partial derivatives

between the components of F and the components of x.

Ó »  Š.î á
16 A ERIAL PHOTOGRAMMETRY AND ITS HISTORY

4. Now, substituting equation 2.3 into equation 2.1 and subtracting

equation 2.2 from it yields
( ) ( )( )
(0) (0) (0)
ℓ−ℓ + v = JF x x̂ − x ,

or with the notations

∆ℓ = ℓ − ℓ(0) ,
∆x̂ = x̂ − x(0) ,
( )
(0)
A = JF x ,

this becomes
∆ℓ + v = A ∆x̂, (2.4)

the linear form of the observation equations.

Very often in writing equation 2.4 we leave the deltas (∆) out (but remem-
ber that these quantities are a lot smaller than the original quantities
ℓ, x̂, which is numerically convenient!). The matrix A is called the design
matrix of the problem. It expresses the linear dependence that exists
between the unknowns and the observations, at or close to the geometry
that actually existed at the time of observation.

D 2.4.4 The least-squares solution

The least-squares solution now follows straightforwardly:

)−1
x̂ = A T Q ℓℓ
−1
A T Q ℓℓ
−1
(
A ℓ,

and for the uncertainties of the estimates of the unknowns x̂ we obtain

as their variance matrix:
)−1
Var {x̂} = Σ x̂ x̂ = σ20 A T Q ℓℓ
−1
(
A ,

with σ0 the earlier defined standard error of unit weight, a “typical”

uncertainty of a single observation.
All this is readily codeable.
We mention separately that in the often occurring case where the
observations are statistically independent of each other and of equal
uncertainty — sometimes the abbreviation i.i.d., i.e., independent and
identically distributed, is used for this —, the weight coefficient matrix
is Q ℓℓ = I n , as we saw in subsection 2.4.1. Then the above equations
simplify to
)−1 )−1
x̂ = A T A A T ℓ, Var {x̂} = Σ x̂ x̂ = σ20 A T A
( (
.

Ó »  Š.î á
 Content of this text
17
D 2.5 Content of this text
This tutorial is accompanied by hands-on exercises using the software
e-foto, an open-source digital photogrammetric work station package
developed in Brazil (Brito et al., 2011). This software is more limited
in scope than commercially available packages, but allows many useful
photogrammetric workflows. It has the merit of being free and thus
accessible to many. Also, it is simpler in structure and less automated,
making it more suitable for teaching and self-study. Finally, it is available
for other operating systems besides Windows™.
Somewhat complementary to this are open-source software packages
for remote sensing and geographic information systems (GIS) use. An
example of the latter is Quantum GIS (Linifiniti Consulting CC, 2008,
maintained)5 , also called QGIS. This open-source package is more like a
framework connecting other open-source products for more specific tasks.
It is very extensive and unfortunately rather tricky to install due to its
many dependencies. Like the commercial package ERDAS Imagine™6 ,
it integrates a geo-referencing solution, which is however based on the
open-source library package proj4 (http://trac.osgeo.org/proj/) which may also
be used separately. This library again is based on the global compilation
of reference systems — thousands of them! — originally maintained
by the European Petroleum Survey Group (EPSG) and found at http:
//www.spatialreference.org/.
Further useful open-source packages are GDAL, which converts a great
many raster and vector image formats used in remote sensing and ge-
ographic information systems into each other, and Monteverdi/OTB, a
package for processing remote-sensing imagery, especially from satellites.

D 2.6 Introduction to e-foto

E-foto is a software developed at the State University at Rio de Janeiro,
Brazil, for the purpose of hands-on teaching digital aerial photogram-
metry on ordinary personal computers. Development started in 2002;
originally a number of independent modules were developed, which have
been integrated into a complete photogrammetric workflow. This integra-
tion is currently still somewhat incomplete.

5 This excellent open-access training manual was originally developed in South Africa.
6A product of Intergraph™.

Ó »  Š.î á
18 A ERIAL PHOTOGRAMMETRY AND ITS HISTORY

E-foto is distributed under a free software licence; thus there are no

restrictions on installing this software even on home computers in private
use. It is intended, however, for use in an academic environment, where
the cost of commercial software licences might be limiting, while learning
the basics of digital photogrammetry doesn’t require these advanced
commercial packages.
The web site of e-foto is at http://www.efoto.eng.uerj.br.
In this tutorial, we describe the use of e-foto using as examples:
1. the test set of three images over Rio de Janeiro, collected in 1997,
and
2. the test set of five images over Greater Helsinki collected in 2004.
Screenshots shown in this booklet are taken sometimes from the Brazilian
processing, sometimes from the Finnish processing. Also, these screen-
shots may be from versions of e-foto older than the latest ones, and the
user interface has undergone (minor) changes. Please don’t be confused.
Data include
◦ the digital photographs,
◦ camera calibration data,
◦ a list of control points with their terrain co-ordinates, and
◦ Descriptions of the control points, allowing them to be found in the
digital images.
In this tutorial, we describe use of e-foto on the Linux operating system,
Ubuntu 12.04 or 14.04. Versions for Windows™ are available, and a
MacOS™ version is in preparation.
Technical note: for Linux a 64-bit version is available from the e-foto
web site. The packages are in the .deb format, which will install
on Debian and derivatives, like Ubuntu. In order to be able to
process large aerial images — larger than 8192 × 8192 pixels — it
is necessary to use the 64-bit version, which will only run on 64-bit
Linux7 , which again will only run on 64-bit hardware.
Also for Windows a 64-bit version is currently available. The 32-bit
versions are being phased out.

We start up e-foto by typing at the command prompt in a terminal window

7 When self-compiling, the source tree will automatically compile a 64-bit version on
64-bit Linux.

Ó »  Š.î á
 Introduction to e-foto
19

D F IGURE 2.6. E-Foto start screen.

$ e-foto

and pressing <Enter>. The start screen comes up (figure 2.6).

The main menu in the top left of the window (see figure 2.7) offers the
options Project, Execute and Help.
Under Execute, you see the various modules that you may execute on
the photogrammetric data at your disposal. We will discuss each of these
further on. The logic connecting these modules (the names used vary a
bit) is given in figure 2.8.
First of all, however, you need to set up a project, containing information
on how to find your data, as well as auxiliary information on your camera
(calibration data) etc. This setting-up procedure is discussed a later

D F IGURE 2.7. E-Foto main menus.

Ó »  Š.î á
20 A ERIAL PHOTOGRAMMETRY AND ITS HISTORY

Start e-foto

Y N
New?

Create project file Load project file

Y
Digital?
N
Interior orientation Phototriangulation

Exterior orientation
3-D co-ordinates
DEM extraction
Stereo plotting

Map features Orthophoto mapping

Map background

Save updated project file

D F IGURE 2.8. E-foto modules and processing logic.

chapter.

Ó »  Š.î á
D Aerial photography and mapping

3
D 3.1 Atmospheric reduction
If aerial imaging is done from a significant height above the ground,
also atmospheric refraction needs to be modelled. For vertical imaging,
the correction model used will be symmetric around the vertical axis: a
standard atmosphere model may be used which will be quite good enough
in view of the smallness of this correction.
The refractive index n of air can be obtained from any of a number
of models called “standard atmospheres”. For our purpose, correcting
vertical photogrammetry, we may use an approximate expression.
We assume the atmosphere to be horizontally stratified, figure 3.2, and
apply Snell1 ’s law, figure 3.1. The stratification can be approximated by a
sequence of discrete steps in refractive index n, see figure 3.2. Then we

α′
T R
δα

α
Q
S Apparent
α uplift

F IGURE 3.1. Snell’s law: change in angle of incidence at a refractive-index

D interface.

1 Willebrord Snell van Rooyen, “Snellius”, 1580 – 1626, Dutch astronomer, physicist,
mathematician and geodesist.

– 21 –
22 A ERIAL PHOTOGRAPHY AND MAPPING

Camera
Z0
δα

Z′

Z n

D F IGURE 3.2. Atmospheric stratification and ray propagation.

have, with Snell’s law for a single refractive-index interface, i.e., a step
change in refractive index by a ratio of n:

PQ PS/cos α PS PT − QT cos δα PT − QT
= = = ≈ .
PR PT cos α PT cos α
2 PT cos2 α PT cos2 α
Here we have assumed that the change in direction δα is small compared
to the ray direction α (alpha) itself.
Snell’s law gives
n sin α′ RT /QT PT
= = = ,
n′ sin α RT /PT QT
So
′ ′
n n
PQ PT − n PT 1 − n
= = cos α
PR PT cos2 α
′
n− n′
1 − nn n n − n′ ∆n
=⇒ PQ (α) = PR = PR ≈ PR = − ,
cos2 α cos2 α cos2 α cos2 α
for n close to unity. In the small-angle limit for α we have

PQ (0) = −∆ n PR (3.1)

Ó »  Š.î á
Atmospheric reduction
23
and
PQ (0)
PQ (α) =
. (3.2)
cos2 α
This is the apparent uplift of a terrain point, expressed in the height
difference PR between terrain and refractive-index interface, produced
by a refractive-index increment at this interface. In the following we write
this height difference as PR = Z ′ − Z .
Now, for a number of interfaces k = 1, . . . , K , expression 3.1 should be
summed to obtain the total uplift:
K
∑
PQ tot (0) = − PR k ∆ n,
k=1
and in the limit for a continuous refractive-index gradient, this becomes
an integral:
ˆ n( Z 0 ) ˆ Z0
dn ( Z ′ )
PQ tot (0) = − PR ( n) dn = − PR ( Z ′ ) dZ ′ .
n(0) 0 dZ ′

Now let us assume that the index of refraction n — or rather, the

expression n − 1 — is an exponential function of height Z ′ , with a scale
height τ (tau):
Z′
( )
n ( Z ′ ) − 1 = C exp − τ .
Then, the derivative of the refractive index:
dn ( Z ′ ) d C Z′
( )
′
= ( n ( Z ) − 1) = − τ exp − τ .
dZ ′ dZ ′
Then
ˆ ˆ
Z′
( ′)
C C Z
( )
PQ tot (0) = PR τ exp − τ dZ ′ = ( Z ′ − Z ) τ exp − τ dZ ′ ,

and this is the integral we have to evaluate all the way through the
atmosphere2 :
v′
ˆ   u
 
( ′ )
C Z
( Z − Z ) τ exp − τ dZ ′ =
′

v v
u
   [ (  ˆ u′ [ ( )]
′ )]
′ Z  Z′
= ( Z − Z ) −C exp − τ − 1 −C exp − τ dZ ′ =
( ′) ( ′)
′ Z Z
= − ( Z − Z ) C exp − τ − τ C exp − τ =
( ′)
Z
= − [( Z ′ − Z ) + τ] C exp − τ

2 Consider that a homogeneous air density (the same everywhere as on the camera
height) wouldn’t have any effect at all on the imaging geometry. It’s the increase in air
density downward from the camera that causes refraction.

Ó »  Š.î á
24 A ERIAL PHOTOGRAPHY AND MAPPING

´
using integration by parts, annotated in the u, v formalism: uv′ dx =
´
[ uv] − u′ vdx. C is a still unknown constant. Now from sea level Z = 0
to outer space Z0 = ∞ this becomes
ˆ ∞ ( ′)
C Z
I 0∞ (0) = ( Z ′ − 0) τ exp − τ dZ ′ = C τ2 ,
0

and we know this, the total atmospheric delay in zenith for visible light
from sea level to space, to be approximately δ Z 0 ≈ 2.4 m. It follows that
C = δ Z 0 · τ−1 .
Now the same integral, not from sea level to outer space, but from
terrain level Z to flight level Z0 , will be given by
ˆ Z0 ˆ Z0
C Z′
( )
Z ′
I Z 0 (0) = PQ (0) dZ = ( Z − Z ) τ exp − τ dZ ′ =
′
Z Z
Z ′ Z0
[ ( ( )]
′
)
= − ( Z − Z ) + τ C exp − τ =
Z
Z Z0 − Z Z0
[ ( ) ( ) ( ) ]
= δ Z 0 exp − − + 1 exp − .
τ τ τ

Now the radial offset in the image becomes, introducing again the
dependence on α of the apparent uplift found above, equation 3.2:

distance foreshortening
cos α
     
δαatm = · sin α ·
Z0 − Z
Z
I Z 0 (α)
 (  
))
δZ0 ( Z ) ( Z0 − Z Z
) (
· exp − τ − τ + 1 exp − τ0
cos2 α
c
=⇒ δ r atm = δαatm =
cos2 α ( ))
c tan α δ Z 0 ( Z ) ( Z0 − Z Z0
) (
= · exp − τ − τ + 1 exp − τ =
cos2 α Z − Z0
( ))
r δZ0 ( Z ) ( Z0 − Z Z
) (
= · exp − τ − τ + 1 exp − τ0 . (3.3)
cos2 α Z0 − Z

Here, c is the calibrated focal length of the camera. we may use, e.g.,
τ = 8.4 km and δ Z 0 = 2.4 m. These values may be adjusted to better
correspond to known local conditions during imaging.
It is not clear from looking at it, but the right-hand side in eq. 3.3 is
to first order linear in Z0 − Z and vanishes at Z0 − Z = 0. This is most
easily seen by noting that the expression in square brackets, and its
first derivative, vanish at Z0 = Z , making it a to first order quadratic

Ó »  Š.î á
Atmospheric reduction
25
expression in Z0 − Z there:

d Z Z0 − Z Z0
[ ( ) ( ) ( )]
exp − τ − τ + 1 exp − τ =
dZ Z = Z0
1 Z Z0 − Z Z0 1 Z0
[ ( ) ( ) ( )]
= − τ exp − τ + exp − τ + τ exp − τ =
τ2 Z=Z
( 0)
1 Z Z −Z Z 1 Z
( ) ( )
= − τ exp − τ0 + 0 2 0 exp − τ0 + τ exp − τ0 = 0.
τ

A very similar geometric derivation, but presented even more simply, is

found in Schut (1969).
Kraus (2007) give a more precise expression from the literature:

r
δ r atm = K,
cos2 α
with
Z02
( )
0.00241 Z2
K= − . (3.4)
Z0 Z02 − 6 Z0 + 250 Z 2 − 6 Z + 250
Some theoretical background to this is given in Gyer (1996)3 .
In figure 3.3 we compare our own simple, horizontally stratified ex-
ponential model above with the model of Gyer (1996) — which gives
tabulated values that were computed based upon the Standard Atmo-
sphere 1976 (Public Domain Aeronautical Software, 2014) and spherical

3 Gyer gives an approximate formula, equation 24, for a horizontally stratified atmo-
sphere (in our notation):
ˆ Z0
( )( )
tan α 1 n( Z ′ ) − n( Z 0 ) n( Z ′ ) + n( Z 0 )
δαatm = ( ) dZ ′ ≈
Z0 − Z Z 2 n2 Z 0
ˆ Z0 (
tan α
n( Z ′ ) − n( Z0 ) dZ ′ .
)
≈
Z0 − Z Z

If we substitute here our exponential atmospheric model

δZ0 ( Z′ )
n( Z ′ ) − 1 = exp −
τ τ
we obtain for the integral above

ˆ Z0 ( ([ ( Z ′ )] Z 0 )
′
) ′ 0 Z0 − Z ( Z )
0
n( Z ) − n( Z0 ) dZ = δ Z − exp − − exp − =
Z τ Z τ τ
( ( Z) ( Z ) )
0 0 Z0 − Z ( Z )
0
= δZ exp − − exp − − exp − =
τ τ τ τ
( ( Z) (Z −Z ) ( Z ))
0 0
= δ Z 0 exp − − + 1 exp − .
τ τ τ

This is seen to lead to a result identical to eq. 3.3, though derived in a very different
way.

Ó »  Š.î á
26 A ERIAL PHOTOGRAPHY AND MAPPING

Photogrammetric refraction, 45 degrees

20

15

arcsecs

10

5
Exponential (8.5 km, 2.4 m)
Gyer table 3
Kraus eq. 4.5-2

0
0 5 10 km 15 20 25

D F IGURE 3.3. Comparison of atmospheric-refraction models.

geometry —, and with the formula 3.4 given in Kraus We see that, for
common aircraft heights, our model performs no worse than the differ-
ences seen between the more sophisticated models. And this is before
taking into account the influence of air pressure variations and climatic
temperature differences.

D 3.2 Earth curvature reduction and local project reference

The Earth’s curvature behaves, within the small area of an aerial mapping
project, like the curvature of a sphere with a radius equal to the mean
radius of the Earth, say 6371 km. The heights Z (or H ) of the ground
control points, which constrain the co-ordinates of the photogrammetric
solution, are given above the geoid, i.e., the figure of the Earth, also known
as mean sea level. Commonly used geodetic heights, e.g., orthometric
heights, are heights above mean sea level. Also normal heights, while
differing slightly from orthometric heights, may be understood as heights
above sea level.
The horizontal co-ordinates X and Y of the ground control points —
which also constrain the co-ordinates of the photogrammetric solution —
can be understood as map projection co-ordinates, produced by project-

Ó »  Š.î á
 Earth curvature reduction and local project reference
27
P′

Z
H

X,Y Tangent plane (map plane)

Curvature
correction
ϕ, λ

D F IGURE 3.4. How to interpret the “Earth-curvature reduction”.

( )
ing geocentric co-ordinates X , Y , Z onto a map projection surface, e.g.,
the UTM (Universal Transverse Mercator) projection on the geocentric
GRS80 ellipsoid, for a chosen central meridian.
The Earth-curvature effect, like the atmospheric effect, is radially sym-
metric: it produces a vertical offset of

d2
δ Zcurv = ,
2R
which in the image becomes visible as

r f r d2 r3 Z − Z
δ r curv = δ Zcurv = = 2 0 ,
f Z0 − Z Z0 − Z 2R f 2R

Z0 − Z
with f the camera focal length, d = r the distance on the ground
f
from the camera footpoint.
Applying an Earth-curvature reduction makes it possible to do the ad-
justment computation of a photogrammetric block in local map projection
and vertical co-ordinates, “as if” these were three-dimensional rectan-
gular co-ordinates. figure 3.4 shows how this works: the co-ordinates
in which point P is given, X , Y , H , do not together form a rectangular
system. “Reducing” the observations means that we pretend to have
observed point P ′ instead, which actually is at a vertical height H above
the tangent plane (you can think of it as the plane onto which the map
projection projects the Earth’s surface), not the curved sea-level reference
surface or geoid. And now the geometry is rectangular and high-school
geometry and the Pythagoras theorem etc. just work!
This is somewhat dangerous, however, and only as correct as the as-
sumption that the geoidal surface in the project area resembles a sphere.

Ó »  Š.î á
28 A ERIAL PHOTOGRAPHY AND MAPPING

Also one should always keep in the back of one’s mind that, in these
co-ordinates proper, the rectangular geometry is actually not valid.

D 3.3 GNSS and IMU measurements in aerial acquisition

Nowadays (the last ten-fifteen years) it is common to position the pho-
tographing aircraft in real time, use a combination of on-board GPS
(Global Positioning System — or some other global satellite positioning
system. Such systems are generically called GNSS, Global Navigation
Satellite Systems — and an inertial measurement unit (IMU). The units
are integrated by Kalman filter and provide, in real time, the following
information for the locations at which images are recorded:
( )
1. The location of the camera optical centre X 0, Y 0, Z0 in a
global geocentric reference frame, typically in a geocentric ITRF
co-ordinate reference frame as provided by the GNSS system.
Before use, these three-dimensional co-ordinates need to be in the
same reference frame as the given points in the terrain, the ground
control points (GCP).
2. The attitude or orientation of the camera, in the form of Euler4
attitude angles (ω, ϕ, κ) (omega, phi, kappa) relative, typically, to
the axes of a local North-East-Up (NEU) co-ordinate frame.
In airborne laser scanning, real-time positioning of the scanning aircraft
is mandatory. As a result, the laser scan imagery (point cloud) is obtained
(after transformation) initially in a geocentric reference frame. From
there, it may be transformed to map projection co-ordinates X , Y and
heights above sea level Z as used locally in the project area, see Section
3.2.
We will have more to say in subsection 16.1.2 on the problem of sys-
tem calibration connected to airborne laser scanning but relevant to
photogrammetry in general.
It must be noted that the on-board GNSS device will give aircraft
( )
positions in the form of geocentric co-ordinates, rectangular X 0 , Y 0 , Z 0 ,
or equivalently, as geographical co-ordinates ϕ, λ (phi, lambda) and height
h relative to a reference ellipsoid, typically that of WGS845 .

4 Leonhard Euler (1707 – 1783) was a Swiss universal genius.

5 In fact, the reference ellipsoid of WGS84 is called GRS80; however, everybody calls that
WGS84, and the parameter values that are used for them are the same to sub-millimetre
accuracy.

Ó »  Š.î á
GNSS and IMU measurements in aerial acquisition
29
As our project co-ordinates relate to the local terrain situation, the
following conversions are therefore necessary:
⎡ ⎤ ⎡ ⎧[ ]
⎤ [ ]
X0 ϕ ϕ map proj. X
−−−−−→
⎪
⎨
⎥ ellipsoid ⎢ ⎥
⎣ Y 0 ⎦ ←−−−→ ⎣ λ ⎦ −−−→ λ Y
⎢
⎪
Z0 h geoid model
⎩
h −−−−−−→ H = Z
Here,
◦ the information on the map projection used in defining the map
co-ordinate system of the ground control points, and of the whole
aerial mapping project, must be provided correctly to the processing
software, and
◦ a good enough geoid model must also be provided in order to convert
the GNSS-produced heights h, which are relative to the reference
ellipsoid, to the heights above sea level Z (or H ) used locally in the
project.
The inertial measurement unit (IMU) again typically gives orientation
angles (Euler angles) (ω, ϕ, κ). The pitch and roll angles ω, ϕ are relative
to the local horizon. however, the heading (yaw) angle κ is referenced to
the direction of the local meridian, whereas what we need is a reference
to the Map North of the map projection used (remember we use map
projection co-ordinates X , Y as our working co-ordinates for the aerial
mapping project), i.e., the direction of the projection’s central meridian.
The difference between the two is known as the meridian convergence,
and it can be approximately computed by

δκ = (λ − λ0 ) sin ϕ,

where λ is our longitude (East is positive), λ0 is the longitude of the

central meridian of the map projection used, and ϕ is our latitude. The
correction needed is now

κ = κIMU + δκ,

See figure 3.5. This gives us the orientation angles κ that we can
use, e.g., as input for our aerotriangulation computation, either as ini-
tial or approximate values, or — with appropriate weights — as prior
constraints.
No corrections to the other two orientation angles are needed, as they
refer to the aircraft axes, not any meridian direction. Also they describe

Ó »  Š.î á
30 A ERIAL PHOTOGRAPHY AND MAPPING

North Map North

κIMU

Central meridian λ0
κ

δκ
Map East

Local
Y (= N )

merid
X (= E )

ian
Map plane

λ
F IGURE 3.5. Meridian convergence and transformation of IMU angles to Map
North. In this diagram it is assumed for the definition of κ that
D the operator is facing port, see section 8.9.

deviations from the local vertical, which coincides everywhere — after ap-
plying the Earth-curvature reduction! — to the rectangular or Cartesian
Z -axis of our computation geometry.

Ó »  Š.î á
D Camera types and camera calibration

4
D 4.1 Camera and sensor types, spectral areas
As camera types we may distinguish:
1. Traditional (frame) cameras, using light-sensitive emulsion on film
2. Digital cameras, using one or several CCD arrays
3. “Pushbroom” or “whiskbroom” sensors

Examples: the SPOT satellite’s HRV pushbroom, Landsat’s The-

matic Mapper sensor. Animated illustration:
http://earthobservatory.nasa.gov/Features/EO1/eo1_2.php

4. Many others, like non-metric cameras

D 4.1.1 Traditional (frame) cameras

A very popular size for these aerial cameras is focal length 6 inches
/ 152 mm, image size 9 × 9 inches / 22.5 × 22.5 cm. If the resolution is
0.01 mm, i.e., the pixel size is 0.02 mm (using the rule of thumb that
resolution is half the pixel size) this means an image of 11 250 × 11 250
pixels. These cameras have four or eight fiducial marks around the edges
and corners of the image field, which get exposed on the film of each
frame to allow interior orientation.
Because digital cameras of sufficiently large format are something of
the very recent past, there are large bodies of aerial imagery in existence
which was collected on film, but which nevertheless is today processed
digitally. Analogue (film) imagery is converted to digital imagery using
scanners. These are similar to the scanners used for converting your old
holiday pictures to computer format, but
◦ their image field is way larger, and

– 31 –
32 C AMERA TYPES AND CAMERA CALIBRATION

F IGURE 4.1. A modern digital aerial photogrammetric camera (Gruber et al.,

2012). Note the mosaicking of the panchromatic image using four
camera units, and the separate red, green, blue (RGB) and near
D infrared camera units. Image © Vexcel Imaging GmbH, Austria.

◦ their geometric precision is high, preserving the precision with

which the original image was captured on film.
Professional photogrammetric image scanners are quite expensive, and
are useless once the historical stock of imagery has been converted. For
this reason, they, and skill in using them, only exist in a few places.
Digital photogrammetry based on scanned analogue imagery is called
soft-copy photogrammetry.

D 4.1.2 Digital cameras

For a long time, these cameras typically offered smaller image sizes than
the traditional cameras, having both a shorter focal length and a smaller
image field. In practical use this means that one has to fly more flight
lines at a lower aircraft altitude, which costs money. Only recently have
digital aerial cameras offered an image size in pixels similar to traditional
film cameras.
An example of a modern digital aerial camera is the Vexcel UltraCam™
Eagle (Gruber et al., 2012), which offers an image size of 104 × 68 mm, at
a pixel size of 5.2 µm (micrometre), i.e., 0.0052 mm, giving 20 010 × 13 080
pixels, fully comparable to traditional film cameras. The focal length for
panchromatic (i.e., black-and-white) image capture is 210 mm. Colour is
provided using separate camera units with lower resolution.
Digital cameras typically use a CCD (charge-coupled device) array
as their image sensor. The challenge is to make the image sensor field
of view suitably large, given that it is a semiconductor chip cut from a
silicon die of given, limited size. One approach is mosaicking multiple
CCD devices, or even multiple cameras, into one. It will be clear that this

Ó »  Š.î á
Camera and sensor types, spectral areas
33
creates calibration challenges. A more obvious approach, also taken in
the UltraCam, is making everything smaller, starting with the pixel size.
Recently (2015), Leica published their DMC III digital aerial camera,
the first large format camera to use CMOS technology instead of CCD
(Mueller and Neumann, 2016). Contrary to the UltraCam, it provides
the panchromatic image using a single camera unit with a focal length of
92 mm, a pixel size of 3.9 µm, and an image size of no less than 25 728 ×
14 592 pixels. As the name says, this is a greytone image; additional
camera units are used at a focal length of 45 mm and a pixel size of
6.0 µm, image size 8 956 × 6 708 pixels, adding red, green, blue and near
infrared at lower resolution.
It is clear that digital image collection offers its own, significant advan-
tages:
1. No need to purchase film media, only a lot of hard-disk storage
space is needed.
2. The workflow is cut short, especially eliminating manual stages: no
more processing of films followed by scanning.
3. No contribution to image noise from film grain.
4. Digital sensors have a greater dynamic range, i.e., details both
within very dark and very bright areas in the same image are
captured.
5. Digital sensors also capture small contrasts better and don’t com-
press them like film does. This leads to the effective resolution
achieved to be superior to scanned film images of nominally better
pixel resolution.
6. More flexibility in collecting spectral information. One can sepa-
rately collect pan-spectral (black-and-white) information at high
resolution, and colour information at lower resolution. And one is
not limited to the traditional RGB visual colours plus near infrared:
one can specify and collect any spectral band, and any number of
them, up to hyperspectral imaging (as in remote sensing). The
UltraCam doesn’t use this possibility yet though.
In recent years, the field of aerial mapping has almost completely
switched to using digital cameras.

Ó »  Š.î á
34 C AMERA TYPES AND CAMERA CALIBRATION

D 4.1.3 “Pushbroom” and “whiskbroom” sensors

Of these, worth mentioning separately are the SPOT and Landsat push-
broom sensors. As this is satellite data, resolution typically is in the tens
of metres, and they are of limited usefulness for geometric photogramme-
try. Yet, e.g., Google Earth™ is largely based on imagery from the Landsat
7’s ETM+ sensor, its panchromatic channel with a resolution of 15 m. This
data, which covers the globe, is freely available.
A pushbroom sensor works by using a line of sensors perpendicular to
the flight direction of the satellite. A whiskbroom sensor instead uses
a single sensor with a small, rapidly rotating mirror, which allows the
sensor to scan a line on the Earth’s surface which is also perpendicular to
the direction of flight of the satellite. As the satellite moves, successive
lines are scanned, together building an image in a strip or band on both
sides of the ground path. As such, there are no individual images, rather
a long, unbroken stream of imagery. Also, instead of orientation elements
for individual images, exterior orientation derives orientation elements as
a continuous function of either time, or distance traveled by the satellite
along its track. This is a very different paradigm.

D 4.1.4 Unmanned aerial vehicles

A technology that only recently has come of age and may warrant consid-
eration, is the use of an unmanned aerial vehicle as the sensor platform.
This has the advantage of being much less expensive than flying a real
aircraft, but at the cost of a limited payload capacity. E.g., traditional
aerial cameras are way too bulky for such a platform. But if the smaller
image size of a commercial off-the-shelf digital camera is acceptable for
the application, usually from a lower flying height, this is certainly an
option. For geodetically precise mapping of large areas it is however less
suitable.

D 4.2 Camera calibration

It is necessary to calibrate1 the camera used for obtaining aerial imagery,
in order to be able to relate image co-ordinates in a camera-bound system
— let us say, ( x, y) — to real-world co-ordinates in the terrain ( X , Y , Z ).

1 Often one speaks of “camera resection”, as the term “calibration” may also refer to
radiometric calibration.

Ó »  Š.î á
Camera calibration
35
Calibration alone is not sufficient for this, but it is a necessary condition.
Calibration is done in a special laboratory set-up.
The simplest of cameras are metric cameras; in the absence of imaging
errors, what they do is project points in the terrain through a single
point, the optical centre, onto a flat imaging surface where an emulsion
or light sensitive optical array is placed. For such an ideal camera, three
parameters, or sets of parameters, need to be determined in calibration:
◦ the focal length of the camera — this also determines the scale at
which the aerial mapping takes place, and its resolution on the
ground
◦ the location xP , yP of the principal point of the camera in relation
to the set of fiducial marks at the edge of the image plane
◦ the locations, in millimeters of x and y, of the fiducial marks in the
camera co-ordinate system.
Normally, however, the objective lens of the camera is not quite optically
perfect; experience has shown that remaining distortion in the image
tends to be radially symmetric around the principal point, and can be
described by a small number of parameters. These are also included in
the camera’s calibration certificate, as a table of offsets in the image plane
away from the principal point. See the below calibration results for an
example.
From this table it is possible to derive the radial distortion by fitting a
simple polynomial to the table values, as follows:

r′ = r + k0 r + k1 r3 + k2 r5 + . . .

Because of symmetry considerations, only odd powers of r are included.

Two or three terms typically are sufficient. Using this polynomial is
preferable to using the calibration table directly, as the calibration mea-
surements themselves contain significant uncertainty. The fit procedure
diminishes the effect of this.
The fit procedure is based on the least squares method, cf. subsection
2.4.1. The observations ℓ are the differences, called radial distortions,
r ′i − r i , i = 1 . . . n, collected in a calibration set-up like depicted in figure
4.2. On the right is a grid with precisely known dimensions, which is
photographed. We see that the distances r ′ of the image points from the
principal point (PP) of the camera are different from what they would
have been if the imaging system were perfect: r . The unknowns are

Ó »  Š.î á
36 C AMERA TYPES AND CAMERA CALIBRATION

Grid
Film
r′ − r
r

PP

D F IGURE 4.2. Camera calibration set-up, concept.

the coefficients k 0 , k 1 , . . . . The observation equations now become, with

residuals v i :
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
r ′1 − r 1 v1 r1 r 31 r 51 ⎡ ⎤
⎥ k̂0
r 32 r 52
⎢ ′
⎢ r2 − r 2 v2 r2
⎥ ⎢ ⎥ ⎢
⎥ ⎢ ⎥ ⎢ ⎥⎢
⎢ ..
⎥+⎢ ..
⎥=⎢ .. .. .. ⎦.
⎥ ⎣ k̂1 ⎥
. . . . .
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎣ ⎦ ⎣ ⎦ ⎣ ⎦ k̂
2
r ′n − r n vn r n r 3n r 5n

One sees that the design matrix A is made up of powers of r i , the

distances of the image point from the camera’s principal point for every
measured point in the calibration grid.
Note that one could correct the focal length so that k 0 = 0. However,
this is not always done. Often, rather the calibrated focal length of the
camera, symbol c, is chosen such, that the radial distortions r ′ − r are
minimized over the whole range of r , from the principal point ( r = 0) out
to the four corners of the camera’s image field.
A more advanced model is that of Duane Brown2 (Brown, 1966) — based
on earlier, almost forgotten work by Conrady (1919) —, which includes

Illuminated Angle α
Camera

Collimator
grid

D F IGURE 4.3. A practical camera calibration set-up with goniometer table.

2 Duane Brown (1929 – 1994) was an eminent American photogrammetrist and geodesist,
remembered especially for his contributions to camera calibration.

Ó »  Š.î á
 Camera calibration
37
D T ABLE 4.1. Calibration data for the Finnish camera: fiducial marks relative to
FC, “fiducial cross”. “FC” refers to the central cross, the origin of all
( x, y) co-ordinates. Eight fiducial marks were used, in the geometry
given below right, and also in figure 7.1 right side.

Camera type: RC30

Lens type and serial no.: 15/4 UAG-S1, 13260
Focal length: 153.279 mm

Principal point x (mm) y (mm)

of autocollimation (PPA) −0.012 0.004
of symmetry (PPS) −0.012 −0.002

Fiducial marks (eight) 3 7 4

1 105.998 −105.995 6 FC 8
2 −106.006 −106.006 2 5 1
3 −105.998 105.995
4 106.005 106.005
5 −0.007 −111.999
6 −112.003 −0.002
7 0.004 112.003
8 112.003 0.005

decentering distortion described by further parameters P1 , P2 , . . . . For

plate co-ordinates x, y reckoned from the princincipal point, it gives
) ( )(
x′ = x + x k 1 r 2 + k 2 r 4 + . . . + P 1 r 2 + 2 x2 + 2 P 2 x y 1 + P3 r 2 + . . . ,
( ( ) )

2 4
) ( 2 2
)) (
′
1 + P3 r 2 + . . . ,
( ( )
y = y + y k 1 r + k 2 r + . . . + 2P1 x y + P2 r + 2 y

where r 2 = x2 + y2 . Here it has also been assumed that the k 0 term is

subsumed in the focal length.
Realistic camera calibration set-ups are not so simple. One problem
with figure 4.2 is that the illuminated grid would have to be at infinity,
and very large. A more realistic — though outdated — approach is given
in figure 4.3 using a precise goniometer table; see Optical Metrology
Centre (undated).
Alternatively, field calibration is used more and more — it has the
merit of calibrating camera behaviour under realistic use conditions. Also
using the starry sky as Nature’s illuminated grid, with precisely known
stellar positions, has been tried.

Ó »  Š.î á
38 C AMERA TYPES AND CAMERA CALIBRATION

D T ABLE 4.2. Calibration data for the Finnish camera: radial distortion in mi-
crometres (µm). Tabulated separately for the four semi-diagonals.
Normally these values are fitted with a model as described above,
yielding parameters k i , P j . Right, numbering of the semi-diagonals
used.

Radius Semi-diagonals Mean

mm 1 3 2 4

10 0.8 0.5 0.7 0.6 0.5

20 1.2 0.4 1.1 0.7 0.8
30 1.2 0.0 1.4 −0.1 0.6
40 1.0 −0.1 1.2 −0.4 0.4
3 4
50 0.8 −0.8 1.0 −0.6 0.1
60 0.3 −0.6 1.2 −0.9 0.0
70 0.8 −0.3 0.7 −0.1 0.2 FC
80 0.3 0.4 0.8 0.4 0.4
r
90 1.2 0.9 0.7 0.5 0.8 2 1
100 1.1 1.0 0.8 0.8 0.9
110 1.7 0.7 0.2 1.4 1.0
120 2.4 1.0 0.0 1.7 1.2
130 1.1 1.1 0.1 1.2 0.8
140 −2.3 0.5 −2.4 −2.3 −1.6
148 −3.3 −2.8 −4.3 −5.9 −4.0

D 4.3 Camera examples

D 4.3.1 The Finnish camera

As an example we give the calibration values of a Wild camera owned

by the mapping consultancy firm FM Kartta Oy and used for acquiring
imagery in the Ringroad II project around Helsinki. The calibration was
performed by SwissOptic AG in Heerbrugg, Switzerland on January 18,
2000. See tables 4.1 and 4.2.

D 4.3.2 The Brazilian camera

The calibration data for this camera is given in table 4.3. It is seen
that the pixel size is quite a bit larger, 0.08 mm, than for the Finnish
camera, 0.02 mm, see sub-section 4.1.1. The original negatives being of
the same size — some 22 cm — this difference is due to coarser scanning.
It may also be that the digital images, which are offered by the e-foto
development team as test images, have been reduced after scanning for

Ó »  Š.î á
 Camera examples
39
D T ABLE 4.3. Calibration data for the Brazilian camera. Units assumed are
powers of mm throughout. Four fiducial marks were used, in the
geometry given below right, and also in figure 7.1 left side.

Camera: Carl Zeiss RMK A 15/23

Serial number: 137 474
Lens: Pleogon A2
Serial number: 137 504

Image size (cm): 23 23

Scanned image size (pixels): 2895 2838
Pixel size: 0.08 0.08
Calibrated focal length: 153.528 mm

Principal point: x (mm) y (mm)

−0.063 −0.037
3
Distortion coefficients:
5 2 FC 1
k0 −2.667 · 10−
9 4
k1 −1.813 · 10−
k2 −5.877 · 10−14
k3 −5.909 · 10−18
7
P1 −6.278 · 10−
7
P2 −7.172 · 10−

Fiducial marks (four): x (mm) y (mm)

1 113.000 0.016
2 −113.006 0.018
3 0.004 113.015
4 0.007 −112.975

better manageability.

D 4.3.3 The Ethiopian camera

The data for the camera used in the Bahir Dar imaging flight is given in
table 4.4. This is a fairly small digital camera.
We see that the sensor size in the cross-flight ( x) direction is 4490
pixels, i.e., less than a quarter of the sensor size of the (scanned) Finnish
imagery, which is no less than 11 000 pixels (see section 4.1.1). This
means that mapping a given area at the same resolution will require over
four times as many flight lines, making the aerial imaging part of the
project four times more expensive.
Now, in reality, this camera will be much smaller and lighter than a big

Ó »  Š.î á
40 C AMERA TYPES AND CAMERA CALIBRATION

D T ABLE 4.4. Calibration data for the Bahir Dar camera.

Camera: IDM_300_ICS2_Calibrated
Sensor size (pixels): 4490 3364
Pixel size (mm): 0.012 0.012
Focal length (mm): 55.1802

Principal point x (mm) y (mm)

of autocollimation (PPA): 0.0910 −0.1496
of symmetry (PPS): 0.0000 0.0000

Distortion coefficients:
5
k1 −2.4153 · 10−
9
k2 8.9472 · 10−
k3 2.6584 · 10−12
6
P1 1.4468 · 10−
6
P2 −3.3601 · 10−

film camera like the Leica RC30, which can weigh in at 100 kg and con-
sume hundreds of watts of electric power from the aircraft on-board power
system3 . Therefore, it will be possible to fly it on a much lighter aircraft,
or even an unmanned aerial vehicle, which will bring down costs a lot.
This remains true even considering that a denser network of ground con-
trol points may be required to achieve good geometric strength. Another
way to strengthen the geometric strength of the network is to include a
large number of tie points, e.g., by automatic tie-point generation.

3 http://w3.leica-geosystems.com/downloads123/zz/airborne/rc30/documentations/

RC30_product_description.pdf.

Ó »  Š.î á
D Flight plan

5
D 5.1 Problem description
When the purpose is to geometrically map an area, that fixes the basics
of the flight plan: the imaging will be vertical (i.e., the camera looking
straight down, rather than oblique), in a block consisting of a number
of parallel strips of overlapping images. The critical parameter to be
known is the resolution on the ground to be strived for: together with the
resolution on the film, typically of order micrometres, we then obtain the
image scale.
Say we want a resolution on the ground of 0.1 m. The film resolution
being 0.01 mm, we obtain as the image scale the ratio of these, 1 : 10 000.
If we then further know that the camera focal length is c = 150 mm — a
typical value, but alternatives exist —, we can also directly compute the
flight height: H = 10 000 × 0.15 m = 1500 m.
We can now also compute how big an area will appear on each image: if
it measures 9 inches (22.5 cm) a side, it will photograph a square 2250 m
on a side, i.e., 2.252 km2 or about 5 km2 . But this doesn’t yet take the
overlap into account. In a very crude, approximate formula:
)2
(1 − O x )(1 − O y ) = d 2 (1 − O x )(1 − O y ) ,
(
A ≈ Sd

where d is image linear size (i.e., 22.5 cm), S = 10 000 is the image scale
number (i.e., the scale is 1 : S ), d = Sd is the size of the image area on
the ground (i.e., 2250 m) and O x , O y are the longitudinal and transversal
overlaps, respectively. This formula is asymptotically valid for large
(many strips, many frames per strip) blocks.
Choosing O x = 0.60 and O y = 0.30 we obtain for the effective terrain
area mapped by one image, with this camera at this resolution:

A ≈ 1.4 km2 .

– 41 –
42 F LIGHT PLAN

Image resolution Image plane

Calibrated focal length c

Optical centre

Flight height H
Resolved detail

Terrain

F IGURE 5.1. Relationship between flight height H, camera focal length c, film
or image resolution (typically one-half of pixel size) and required
D resolution on the ground

Knowing the total size of the area now gives you the amount of film you
are going to consume. Figuring out the number of strips, the number
of images per strip, and the total flight distance — remember to take
generous turns between strips! — takes further work; see the example
below. By now, you should be set up to invite tenders from aerial mapping
companies.

D 5.2 Example
As an example we consider a target area of 3 × 5 km. See figure 5.2. This
kind of picture is definitely worth drawing.
Firstly let us assume that we fly along the longest direction of the area,
length D x = 5 km. This is generally preferable, as it means that the least
amount of time is lost to unproductive turns of the aircraft. Then, we
have 5 km to cover with images that are 2.25 km broad. But, remember
that the images have to overlap by 60%. This means that the first and
the last images are “extra”. The other images are offset by (1 − O x ) · d
from each other; this distance is also the distance between the camera
exposure centres and is known as the image or frame separation.
The positions of the left edges of the images span the range [0, D x − d ] .

Ó »  Š.î á
 Example
43
2.25 km
1 2 3 4 5 6

2.25 km

3 km
30%

60%

5 km
12 11 10 9 8 7

D F IGURE 5.2. Flight plan target area

This gives us the formula for the number of images (exposures) in one
flight strip:
Dx − d
nx = + 1 + 2,
(1 − O x ) · d
where the extra number 2 represents the extra first and last images in
the strip (needed to produce stereo view over the whole target area), and
the number 1, that the number of edges on the closed interval [0, D x − d ]
is one more than the length D x − d is a multiple of the offset length
(1 − O x ) · d .
Thus the number of images in a flight strip will be
5 km − 2.25 km
nx = + 3 = 6.06.
0.9 km
One is tempted to round to 6, by, e.g., slightly reducing the longitudinal
overlap. Generally speaking, one should instead round upward to the
nearest integer.
Obviously the number of strips needs to be 2 in order to cover the 3 km
width of the area. Then there will be an overlap of at most
2 × 2.25 km − 3 km
Oy = = 0.67.
2.25 km
As only 30%, i.e., O y = 0.3, is required, we are again left with “edges” of
2 × 2.25 km − 3 km − 0.3 × 2.25 km = 0.825 km total, or 412 m on each side.
A general formula for the number of strips, analoguous to the above
one for the number of images per strip, is
Dy − d
ny = + 1,
(1 − O y ) · d

Ó »  Š.î á
44 F LIGHT PLAN

Optimized coverage area (uncertainty pink)

Target area ∆D x

∆D y Raw coverage area

D F IGURE 5.3. Placement of coverage area over target area.

with D y the width of the target area. This yields n y = 1.48, to be rounded
upward to 2.
Both n x and n y should be rounded upward to the nearest integer. Their
product n x n y equals the total number of images (frames, exposures) for
the whole project.
If the computed n x and n y are not integers but need to be rounded
upward, the imagery will cover more area on the ground than the target
area, and there will be “space left”, see figure 5.3. The formula for this
“space left”, on the left and right, and upper and lower, edges, is now
“space taken by images” minus “space taken by overlaps” minus again the
real dimension of the target area. As follows:
( )
∆D x = ( n x − 2) d − ( n x − 3) O x d − D x =
( )
= ( n x − 2)(1 − O x ) + O x d − D x ,
( ) ( )
∆D y = n y d − ( n y − 1) O y d − D y = n y (1 − O y ) + O y d − D y ,

half of which is the shift, in both directions, needed to place the target
area in the middle of the coverage area. This helps to safeguard full
coverage also in the presence of small deviations from the nominal flight
plan.
The flight time between exposures is obtained directly from the image
separation ∆ xexp = (1 − O x ) d , as follows:

∆ xexp
∆ t exp = v ,

Ó »  Š.î á
 Practicalities
45
where v is the aircraft velocity. Multiply with the total number of expo-
sures n x n y , and one obtains the total photography time.
We want every point in the whole target area to appear in two pho-
tographs at least. This means that a border area around the target area
will be photographed on single images of the project. It is wise to have
ground control points also in this border area, as this strengthens the
control of the camera location and orientation for these edge exposures.

D 5.3 Practicalities
Both the above examples assume that the target are of measurement is
rectangular. Sometimes it is, but rarely exactly so. For approximately
rectangular areas, one may try to enclose the area in a rectangle, and
then proceed as above. For irregularly-shaped areas, there is no other
technique than the manual one.
1. Put a map of the area on your desk, with the outline of the target
area drawn on it.
2. Cut, from paper or cardboard, squares that are the same size as an
image on the ground, but reduced to map scale.
3. Mark on the squares the positions of forward and sideways overlap
by straight lines.
4. Now, arrange the squares into a proper geometry, with full, stereo
coverage, and overlaps in both directions as specified. Use pieces of
sticky material to keep the squares in place.
5. You may need to experiment a little to find the best solution. Do not
just minimize the number of images taken, try also to minimize the
number of strips, and thus, the number of aircraft turns needed.
6. Record your result, with numbered images, on a planning document
and planning map blank. Thus one obtains a flight map, which can
be given to the pilot and the photographer of the imaging flight.
The flight map should contain the flight lines (with the direction of
flight indicated) and the individual exposure locations.
In an area with strong mountain slopes (Nepal!), it is not a good idea to
fly up and down the slope, so this constrains the flight direction: rather,
fly along the height contours at contant height.
About the timing of photography: in moderate climate zones it is wise
to schedule for early spring, when the snow has left but deciduous trees

Ó »  Š.î á
46 F LIGHT PLAN

Air Strip

km
10
20
km

D F IGURE 5.4. Area to be mapped in the flight plan exercise.

are still leafless. In the tropics, avoid the rainy (cloudy) season.
The uncertainty of weather is something to always take into account.

D 5.3.1 Flight Plan Exercise (1)

Given: available camera, required resolution (point accuracy) on the

ground. Assume as a rule of thumb that we can achieve a point accuracy
(resolution) corresponding to ±0.5 pixel size.
◦ We have a film camera with focal length 150 mm and image field of
22 cm × 22 cm. The size of a pixel on the film is 0.01 mm.
◦ We have an area to be mapped of 10 × 20 km2 , see figure 5.4.
◦ We require a point accuracy (ground resolution) of ±5 cm on the
ground.
Furthermore, assume
◦ One image (colour) costs 1000 credits.
◦ The aircraft flies 400 km/h, and one hour of flying costs 100 000
credits.
We need to answer the following questions.
◦ What has to be your flying height?
◦ How many strips, of how many images each, do we need to fly
(assume 60% / 20% overlap). Make a sketch including the path of

Ó »  Š.î á
 Practicalities
47
the aircraft. Which flying direction do you think is best?
◦ Draw ground control and tie points into your sketch to create a
good geometry for the aerotriangulation. Explain what makes the
solution strong.
◦ How many total images will you have to take? Estimate cost for
this film material. Discuss colour vs. greytone.
◦ Choose a suitable aircraft (discuss!), flying speed, cost.
◦ Compute hours in the air assuming 400 km/h flying speed. Assume
that the airstrip is close to the project area, so you can ignore ap-
proach and return flights. Estimate cost assuming 100.000 credits
per hour.

D 5.3.2 Flight plan exercise (2)

Now we have, for the same area and ground resolution requirements,
a different camera: a digital camera (see subsection 4.3.3) with a focal
length of 55.18 mm and a pixel size of 0.012 mm. The camera image field
is 4490 × 3364 pixels, with the higher pixel count being in the cross-track
direction, as is customary with digital imaging.
◦ We again have an area to be mapped of 10 × 20 km2 , see picture.
◦ We again require a point accuracy (ground resolution) of ±5 cm on
the ground.
Furthermore
◦ Images have zero cost (no film is used).
◦ Still, the aircraft flies at speed 400 km/h, and one hour of flying costs
100.000 credits.
Answer the following questions:
◦ What has to be now your flying height?
◦ How broad, in metres, is a single flight strip on the ground?
◦ Assume 60% / 20% overlap. How many flight strips are needed to
cover the area?
◦ Compute again hours in the air assuming 400 km/h flying speed,
again assuming the airstrip is close to the project area. Estimate
cost assuming 100.000 credits per hour. Compare with previous
result.

Question: how realistic is the assumption that the same aircraft will be
used for this case as in the first case? Read carefully chapter 4.

Ó »  Š.î á
48 F LIGHT PLAN

Ó »  Š.î á
D Setting up an e-foto project file

6
Note that the e-foto system is integrated so, that data from a previous
module is carried forward to the next module inside the project definition
file, which has a file name extension of .epp. This means that before exe-
cuting a module you must load the project definition file; after completing
it, you must save it, so it is available for following modules. These load
and save options are available under the Project menu item.
In figure 6.1, we see the different tabs that the Project Manager module
offers:
◦ Project Header, already filled out with data for this project
◦ Terrain data
◦ Sensor: here the camera calibration data is entered
◦ Flight data
◦ Images: here the path names to the digital images is given
◦ Points: here, terrain points and their co-ordinates are entered, typi-
cally by loading them from a simple text file.
Before trying to change the information in any of the tabs, you need to
press the Edit button. When finished, you press the OK button — or, to
throw away a failed edit, the Cancel button.

D 6.1 Project Header

Here is entered the general information on the project, including infor-
mation on the project definition file. Figure 6.1 shows the header page
of a previously filled out, saved and re-loaded project file. Not all of this
information is required, and the metadata is automatically filled in.

– 49 –
50 S ETTING UP AN E - FOTO PROJECT FILE

D F IGURE 6.1. Loaded project file.

D 6.2 Terrain
Here we enter general information on the area to be mapped. The inter-
esting fields are:
GRS (Geodetic Reference System): this is pre-filled with WGS84, but
other options are SAD69 (South American Datum 1969) and SIR-
GAS2000 (Sistema de Referencia Geocéntrico para las Américas).
Also Corrego Allegre (a Brazilian system) is provided for in the
code. Unfortunately all this is hardwired at present.
Outside South America, WGS84 is the prudent choice, and may be
taken as an alias for, e.g., ITRF yy (International Terrestrial Refer-
ence Frame), ETRF yy (European Terrestial Reference Frame) or
some similar regional geocentric reference frame. The co-ordinate
differences among all these frames are on the sub-metre level,

Ó »  Š.î á
 Terrain
51

D F IGURE 6.2. Entry of terrain parameters.

and are constant across typical photogrammetric mapping project

areas.
CPS (Cartographic Projection System): pre-filled is UTM (Universal
Transverse Mercator), with no alternatives provided.
UTM Fuse (UTM zone): this is the UTM zone number, counting East-
ward from the date line (180◦ East, or West). Every zone is six
degrees wide. The Western hemisphere comprises zones 1 – 30, the
Eastern hemisphere zones 31 – 60. So, e.g., zone 35 runs from 24◦
to 30◦ with a central meridian of 27◦ , and zone 37 from 36◦ to 42◦
with a central meridian of 39◦ . It appears that e-foto fills in this
field from the longitude data for the project centre point, discussed
below.
Latitude (project area centre): enter numerical value in degrees, min-
utes and seconds.decimals. Choose the correct hemisphere from
the pulldown menu.
Longitude (project area centre): enter numerical value in degrees, and
again pick the correct hemisphere from the pulldown menu.

It is probably wise to set all these fields to roughly correct or correct

values. Maximum, minimum and mean altitude (of the terrain, not the
aircraft!) are pre-set to zero, and we have observed nothing adverse from

Ó »  Š.î á
52 S ETTING UP AN E - FOTO PROJECT FILE

D F IGURE 6.3. Sensor data entry form (part).

leaving them so set. Of course in mountainous or high-altitude areas one

should enter more realistic values.
In fact, if you don’t use an on-board GNSS receiver, probably none of
the parameters on reference system, map projection and area location
are relevant, and you can safely leave them unset. Your results will be in
the same, locally rectangular, X Y Z (or EN H ) system as that in which
the ground control point co-ordinates are given, and the software doesn’t
need to know what system that is — as long as you know. . .

D 6.3 Sensor
Here we enter first of all the camera description and properties. The
camera focal length and co-ordinates of the principal point are taken from
the calibration certificate of the camera.

Then, lower on the same page (use the scroll bar on the right), we
may enter the distortion model coefficients k i , P i according to the Brown-

Ó »  Š.î á
 Flight
53

D F IGURE 6.4. Entry of camera distortion coefficients.

Conrady model (Brown, 1966). Enter only if these have been determined
for your camera.

Then, finally the camera co-ordinates of the fiducial marks. These

too are provided on the camera’s calibration certificate, and are entered
here.

D 6.4 Flight
Here is entered the photogrammetric flight information, to the extent
that it is known.
The software uses only the Nominal Scale field for setting up the initial
approximate values of the geometry — the Nominal Flight Altitude value
is not actually used. It is probably wise to enter it though. The values
shown for longitudinal overlap (60%) and for transversal overlap (30%)
are typical for aerial mapping flights and are known to give good results.
Here you should enter the values actually used in this project’s image
acquisition flight:
◦ Longitudinal Overlap is the overlap of two successive image frames
optained during the flight within a single strip of images,

Ó »  Š.î á
54 S ETTING UP AN E - FOTO PROJECT FILE

F IGURE 6.5. Entry of fiducial marks co-ordinates in camera frame, from the
D camera calibration certificate.

D F IGURE 6.6. Flight data entry form.

Ó »  Š.î á
 Images
55
Longitudinal
overlap

Flight
Transversal overlap
path

D F IGURE 6.7. Flight geometry.

◦ Transversal Overlap is the overlap between photographs in adjacent

strips of images, with the flight direction of the aircraft typically
turning in between.
The flight geometry is depicted in figure 6.7.

D 6.5 Images
The Images tab allows the importation (i.e., setting the path names where
to find them) of digital images. You can import the whole set in one go.
Note 1: When re-loading a project file, e-foto may complain that it cannot
find the images you have just imported. If this happens, open the
sub-tab for every image, click on the Edit button at the bottom, and
use Select Image to re-set the path to that image. Press OK. After
that, all should work OK.
Note 2: When loading a project file that was downloaded from the Inter-
net — like the example project files provided by the e-foto project
web page — you should also not expect these file pointers to be
correct, as you will have placed your image file in a location on
your own hard disk, which the project file does not know anything
about. You will have to go through the Select image procedure for
each image.
Note 3: If you are using the e-foto version of June 30, 2016, you need to
rotate the images in a graphic editor like the GIMP. The angle of
rotation should be the approximate exterior orientation angle κ.
For more details, we refer to section 8.9 on exterior orientation.

Ó »  Š.î á
56 S ETTING UP AN E - FOTO PROJECT FILE

F IGURE 6.8. Project manager of e-foto. Here, interior orientation of three images
D has been completed, but exterior orientation has not yet begun.

As the execution of the various modules progresses, the Project Manager of

e-foto keeps track of the work: clicking the Images tab will show you for
which images Interior and Exterior Orientation modules were completed.
In figure 6.8, interior orientation for all three images has been completed,
but exterior orientation has not yet been started.
Click on every Images sub-tab individually to open up a form to be filled
out for only that image: figure 6.9. You need to press the Edit button at
the bottom before being able to fill out the from. Press OK when you are
finished (or Cancel to start over again).
As you see, there are fields to enter data on where the images sit on
your hard disk, and under that,
◦ The position of the camera at exposure as determined by an on-
board GNSS (Global Navigation Satellite Systems) receiver. If the
aircraft has an on-board receiver, click on the button so a little
cross appears. After that, you can choose whether the receiver was
actually used, and whether the co-ordinates obtained should be
used as an initial value to an iterative calculation of the camera
position. And after that, the camera position co-ordinates measured,
with their uncertainties (“StdDev”, for standard deviation).
Note that the co-ordinates to be entered are named E 0 , N0 and H0 ,
for “Easting”, “Northing”, and “Height”. These correspond to the
exterior orientation camera co-ordinates X 0 , Y0 , Z0 which are listed
in the lower box of the form. E-foto is not exactly consistent on
notation. . .

Ó »  Š.î á
 Images
57

D F IGURE 6.9. Project entry form for an image.

Note 4: For older versions of efoto (older than about 2014) you
must use the following workaround for a bug in the software:
If you don’t have GPS or INS data with your images, make
sure to
1. enable the “Ground Coordinates. . . ” click box; and
2. choose type: “Not Considered”.
3. Do the same for Inertial Navigation System (INS) data, see
below.
The need to do this is due to a bug / illogicity in these e-foto

Ó »  Š.î á
58 S ETTING UP AN E - FOTO PROJECT FILE

software versions, that has been corrected in recent versions.

◦ The orientation of the camera at exposure, determined by an IMU

(inertial measurement unit). Same remarks apply as for GNSS (i.e.,
following Note 4 above, if you have an older efoto version, enable
the click box, and choose “Not Considered”).
◦ Metadata about the image file, and
◦ The results of interior and exterior orientation, or phototriangula-
tion, if these are available already in the project file.

Technical remarks:

You might think that if your images are in colour, you could convert
them to greytone, which would make it easier for e-foto to handle them
especially on a small laptop. However, although the files will be only one
third of the original size, e-foto internally stores them still as 32 bits per
pixel, so this doesn’t really help much.
A better idea is to reduce the images to one quarter size by the following
command:
$ convert <inputfile> -resize 50% <outputfile>

but remember that then you also lose half of your resolution on the
ground. For training with the software this is OK, but not for production
work. The convert command belongs to the ImageMagick package, which
you need to have installed.
Since 2015, only 64-bit versions of the software are offered for download,
both for Linux (Debian, Ubuntu) and for Windows.

D 6.6 Ground control points

Control points can be loaded from an external text file, if it is in the right
format. Alternatively you can enter the data manually by clicking on the
New button, and filling in the form that appears. You can save your data
into a file, and inspect this file to see what the right format looks like. It
is a simple text columns format that is apparently widely used by other
photogrammetric software as well.
The point file format is:
1. Point number
2. Easting
3. Northing

Ó »  Š.î á
 E-foto, reference systems and map projections
59

D F IGURE 6.10. A filled-out Points tab for the Rio de Janeiro project.

4. Height
and the fields are separated by a tab symbol (Control-I or ASCII 9).
Remember also that the decimal separator should be a point, not a
comma, as the e-foto software has not been localized at this point.

Technical remark: the co-ordinate data for the control points must be
given in this order: first Easting, then Northing (like in mathemat-
ics, first x, then y). Geodesists do it the other way around: first
Northing ( x), then Easting ( y). If your file is formatted like that,
you need to interchange columns:
$ awk ’{ t=$2 ; $2=$3; $3=t; print }’ inputfile > outputfile

If the file contains even stranger stuff, like “Southings” or “West-

ings”, you have a bigger editing job ahead.

D 6.7 E-foto, reference systems and map projections

As the e-foto software currently only supports a very limited number of
geodetic reference systems and only one map projection (Universal Trans-
verse Mercator, UTM), your options are limited if your ground control
point co-ordinates are given in some traditional, local, non-geocentric

Ó »  Š.î á
60 S ETTING UP AN E - FOTO PROJECT FILE

D F IGURE 6.11. Ground control point list in vi text editor for the Finnish project.

datum. In Finland, e.g., ground control points may be given in KKJ,

the National Map Grid Co-ordinate System, which uses a variant of
the Gauss-Krüger map projection on the International Ellipsoid of 1924
(“Hayford”), with central meridians of 18, 21, 24, 27, 30 and 33 degrees
East. And note that often in local applications, truncated co-ordinates are
used where leading digits have simply been dropped. . .
Similarly in Ethiopia we have the Adindan datum, on the Clarke 1880
(RGS) ellipsoid, using the UTM projection in zones 35, 36, 37 and 38.
In these cases, if you need to bring together ground control point
and aircraft image exposure co-ordinates — which are typically given
in WGS84 — one solution is to transform the ground control points
co-ordinates to WGS84, and then to project them into a suitable UTM
projection. Alternatively, transform the aircraft co-ordinates to the locally
used datum, and project them using the map projection in local use. After
that, e-foto will be able to work both with the ground co-ordinates given
and with the aircraft co-ordinates provided by GNSS.
A suitable software for doing this transformation and re-projection is
proj4, an open-source package that knows a great many different national

Ó »  Š.î á
 E-foto, reference systems and map projections
61
D T ABLE 6.1. The ground control points in Bahir Dar. Note that the co-ordinates,
obtained by hand-held GPS, are rounded to whole metres.

Number Easting Northing Height

71 0321958 1280485 1812

72 0322001 1279905 1812
73 0321397 1280118 1806
74 0321448 1280648 1807
75 0321527 1281015 1804
76 0321540 1281256 1804
77 0321395 1281478 1805
78 0321341 1281824 1807
79 0321719 1281920 1798
80 0321272 1282164 1806
81 0321335 1282380 1800
82 0321551 1282293 1801
83 0322001 1282290 1804
84 0322067 1282126 1800
85 0321912 1281490 1807
86 0321913 1281156 1809
87 0321930 1280746 1813
88 0322087 1280866 1813

and regional geodetic datums and map projections — in fact, the whole
set compiled by the EPSG, the European Petroleum Survey Group, as
documented at, e.g., http://www.spatialreference.org.
On the other hand, realistically speaking, these traditional geodetic
datums are often non-geocentric and of limited precision, being based
on measurements made before the satellite era. If this is the case, you
wouldn’t be able to use GNSS-provided aircraft co-ordinates as such
anyway, but would have to allow the software always to adjust these to
agree with the system the ground control points are in. But then, there
is no real reason to go through a complicated transformation: just choose,
in e-foto’s Image tab, for every image, to disable the cross mark at “Ground
Coordinates of exposure station centre (optical center of sensor)”. Then, all
computations will be done in local co-ordinates and the geodetic reference
system and map projection settings will be ignored.

Ó »  Š.î á
62 S ETTING UP AN E - FOTO PROJECT FILE

D T ABLE 6.2. Editing the project file, in order to “forget” an image’s exterior
orientation. Here it is image number (“key”) 4 that is edited out in
two places. This works for the version of June 2016.

<exteriorOrientation>
...
<imageEO type="spatialResection" image_key="4">
<Xa>
<X0 uom="#m">41736.60185948185</X0>
<Y0 uom="#m">73101.76725801102</Y0>
<Z0 uom="#m">621.2541973006183</Z0>
<phi uom="#rad">0.02360511107121352</phi>
<omega uom="#rad">-0.0150353187767421</omega>
<kappa uom="#rad">1.55154177028697</kappa>
</Xa>
</imageEO>
...
</exteriorOrientation>

<spatialResections>
...
<imageSR image_key="4">
<iterations>5</iterations>
<converged>truefalse</converged>
...
</imageSR>
...
</spatialResections>

D 6.8 Concluding remarks

Although the Project Manager module allows the extensive editing of all
project-related data, some things are not accessible to editing. E.g., there
appears to be no way of making the system “forget” that an image was
exteriorly oriented, or remove a control point from an image if it doesn’t
actually appear in that image, and was entered accidentally.
Fortunately the project definition file — the one ending in .epp — is
a simple text file, formatted as XML (eXtensible Mark-up Language), a
format which is fairly straightforward to understand, and can be edited

Ó »  Š.î á
Concluding remarks
63
in a text file.
If you know what you are doing (and, just in case, take a back-up copy
of the file), you can edit this file in an ordinary text editor. In Ubuntu
Linux this means vim, emacs or gedit. See table 6.2.

Ó »  Š.î á
D Interior orientation

7
In a film image obtained with an analogue camera, i.e., an emulsion
on a glass or plastic film base, film co-ordinates (ξ, η) (xi, eta) were
traditionally measured, in metric units, in a co-ordinate measurement
machine. These give the positions of points in the image obtained in a
frame connected to the edges of the film. For a scanned image, i.e., an
image file, each pixel has pair of row and column numbers. These can be
extracted by software, and converted to metric image co-ordinates using
the known resolution (in pixels per mm) or pixel size (in µm) of the image.
This effectively simulates a co-ordinate measuring machine.
Interior orientation is the process to express co-ordinates measured
on the photographic plate or the scanned digital image, so-called film or
image co-ordinates, from the image or film co-ordinate frame into the
camera co-ordinate frame, so-called plate co-ordinates. It establishes the
transformation between these two co-ordinate frames. This is done by
using the fiducial marks, markers all around the edge of the camera’s
image field, which are photographed onto the emulsion together with the
terrain image at the moment of exposure. Typically there are either four,
or eight, fiducial marks.
The origin of the camera frame is the principal point, the projection of
the optical centre of the camera onto the image plane.
After the interior orientation transformation is derived, it may be used
to convert the film or scanned-image co-ordinates (ξ, η) of any point in
he image to rectangular metric camera co-ordinates ( x, y). e-foto does
this automatically in the background when one is measuring point co-
ordinates in the imagery.

– 65 –
66 I NTERIOR ORIENTATION

4 3 7 4

2 1 6 8

3 2 5
1

F IGURE 7.1. Fiducial marks of a camera. There may be four or eight fiducial
D marks.

D Note on digital images

Nowadays the use of digital aerial cameras, obtaining digital imagery

directly, is common. In such cameras, the image sensor, a CCD (charge-
coupled device) array, is an integral part of the camera and rigidly at-
tached to the camera body. For this reason, it is not necessary to perform
interior orientation on digitally obtained imagery. Only camera calibra-
tion is needed, and serves for all images obtained with that camera.
If, however, one uses images that were originally obtained on film and
then scanned, interior orientation is needed: the film roll moves — is
transported forward — between exposures, and the precise position of the
film in the camera image field will vary a little from exposure to exposure.
This makes it necessary to reconstruct this precise position, using fiducial
marks exposed onto the same image and rigidly attached to the camera
body, the process known as interior orientation.

D 7.1 Theory
The relationship between
1. film co-ordinates (ξ, η) — typically obtained in a film measurement
device, or in case of digital imagery, by dividing, in the photogram-
metry workstation, the position in pixels in both co-ordinate di-
rections by the known pixels-per-centimetre value of the image —
and
2. camera co-ordinates ( x, y) referring to the camera’s principal point,

Ó »  Š.î á
 Execution
67
can be described in a number of ways, from the very simple to the very
complex. A simple model would be a similarity transformation:
[ ] [ ] [ ][ ]
x x0 cos θ sin θ ξ
= +K .
y y0 − sin θ cos θ η
Here, the unknowns of the relationship are a scale factor K , a rotation
[ ]T
angle θ , and a translation vector x0 y0 . This vector describes the
offset between the origin (like the bottom left corner) of the photographic
plate or digital image co-ordinate system, and the principal point of the
camera. The principal point is the orthogonal projection of the optical
centre of the camera objective lens onto the image plane.
However, more common is an affine transformation:
[ ] [ ] [ ][ ]
x x0 a b ξ
= +K ,
y y0 c d η
usually written
[ ] [ ] [ ][ ]
x a0 1 + a1 a2 ξ
= +
y b0 b1 1 + b2 η
with six free parameters. This allows, besides a rotation and scaling, also
for the image to be skewed, even in two directions. This means that it
depicts a tiny square in pixel space as a tiny diamond or rectangle in the
camera co-ordinate frame, and a tiny circle as a tiny ellipse.
In other words, the pixel sizes may be different in the x and y directions,
and the directions of the rows and columns of pixels may not be precisely
perpendicular to each other. Both of these distortions may happen if the
film material is not absolutely stable but can stretch; or more likely, if the
manufacturing precision of the film scanning device is less than perfect.
Note that, as both the measured film or digital-image co-ordinates
and the camera co-ordinates are in metric units, and the axes of both
are roughly parallel (with each other and with the edges of the image)
it follows that all four a 1 , a 2 , b 1 , b 2 are typically small numbers. The
offsets a 0 and b 0 are larger, denoting the vectorial offset from image
measurement origin (typically an image corner) to camera co-ordinate
origin, the principal point in the middle of the image.

D 7.2 Execution
In preparation for interior orientation, the co-ordinates of the fiducial
marks as given in the calibration certificate (e.g., table 4.1) must have

Ó »  Š.î á
68 I NTERIOR ORIENTATION

x Principal
Objective x x
η point

Image on film
x
ξ Camera body

x
Principal point
y
Image on film η
Camera body
ξ

F IGURE 7.2. Camera principal point, camera co-ordinates x, y, and plate co-
D ordinates ξ, η.

been entered — as must, of course, be all other pertinent camera calibra-

tion data, like the calibrated focal length and the radial-distortion table
(table 4.2).
The data for executing the interior orientation is collected from the
image in the software used; by first loading the image and its calibration
data, and then placing the mouse cursor on each of the fiducial marks in
turn. Clicking the mouse will register the film co-ordinates ξ, η for the
fiducial mark in question; the camera co-ordinates x, y for the same mark
were loaded from the calibration data.
As we have six unknown parameters, and four or eight fiducial marks,
i.e., eight or sixteen conditions, the determination of the parameters is
overdetermined. Software for interior orientation solves for the parame-
ters by a least-squares estimation procedure. This procedure will produce,
as outputs:
1. a table of residuals, one for each fiducial mark, showing how well
the interior orientation solution obtained is able to place the fiducial
marks of the oriented image on the fiducial marks of the camera’s
image plane — in other words, how well we have succeeded. The

Ó »  Š.î á
 Interior orientation with e-foto
69
presence of large residuals will indicate that we did something
wrong.
2. A value for the variance of unit weight σ20 , as well as a value for
its square root, the standard error of unit weight, also root-mean-
square error or RMSE, σ0 . These numbers are also a measure for
how well the interior orientation has succeeded. They too tell us, if
they are unduly large, that we must have done something wrong.
If we find a very large σ0 value, the first thing to do is look at the table
of residuals. Often the presence of a very large residual for one of the
co-ordinates, or both, of one fiducial mark will indicate something wrong
with that fiducial mark. Check for clerical errors in the calibration data
entry; then, re-do the measurement of that point, and re-do the interior
orientation. Now σ0 should be suitably small. Do not accept your result
until this is the case.

D 7.3 Interior orientation with e-foto

Interior orientation is the process to express the digital image’s co-
ordinate frame (“film co-ordinate frame”) into the camera co-ordinate
frame, deriving the transformation between the two co-ordinate frames,
as described in detail above. This section will guide you through
e-foto’s Interior Orientation photogrammetric project stage, or module, in a
step-by-step fashion. It requires a project file that was saved from the
previous module, the Project Manager module.

D 7.3.1 Getting started

If you start the e-Foto software, the start screen comes up (figure 2.6). The
main menu offers the options Project, Execute and Help. Using the Interior
Orientation module requires that you load a project file created and saved
by the Project Manager module: its new project setup process including
the entry of the photogrammetric camera calibration certificate data, and
entering the paths to the digital images to be used. If you haven’t done
so yet, we refer you to the tutorial on the Project Manager module.
1. Select the Project menu item from the main menu.
2. Load the previously saved project definition file using the “Load file”,
“Last project”, or “Open file” option, depending on e-foto version. The
file has an extension .epp. See figure 7.3.

Ó »  Š.î á
70 I NTERIOR ORIENTATION

D F IGURE 7.3. Loading a previously saved project file.

If you select the project file, the screen shown in figure 6.1 will come
up.
Now choose Execute ▷ Interior Orientation. You’re ready to go!

D 7.3.2 Starting interior orientation

A window will be presented asking you to select your image, figure 7.4.
Please select the first image in your set (this will be the one presented for
selection).
Next, you have to identify the fiducial marks according to their num-
bers.
Now you should move the mouse, and click, every fiducial mark in turn.
The followings steps must be executed in sequence, for every fiducial
mark:

D F IGURE 7.4. Select your image.

Ó »  Š.î á
 Interior orientation with e-foto
71

Zoom Fixed Detailview

Set mark
Show report
Fit View Interior orientation
Detailview

Fiducial mark co-ordinate list

D F IGURE 7.5. Interior orientation main view.

1. Click on the Zoom icon on the toolbar.

2. Make sure that the icon Fixed Detailview has been pressed down;
click on it if needed.
3. In the big picture, move the mouse cursor to the current fiducial
mark that you wish to measure.
4. Use the mouse to draw a small, yellow rectangle around the fiducial
mark. Click the top left corner, pull the mouse, with button held
down, to the bottom right corner — see how a little yellow rectangle
forms —, and release mouse button. Now the fiducial mark and its
surroundings should show up magnified on the screen.

Ó »  Š.î á
72 I NTERIOR ORIENTATION

F IGURE 7.6. Fiducial marks in the Detailview at 8x magnification: left an edge

D mark, right a corner mark.

5. Verify that, in the fiducial mark co-ordinate list, below the image,
the row of the table belonging to the current fiducial mark is the
active one; if not, click on the row you wish the co-ordinates to be
entered into the fiducial mark co-ordinate list.

Note: if you have measured a fiducial mark wrong, or not precisely

enough, you can also go back and re-measure by clicking on
the table row for that mark. The values in the row will be
overwritten as you measure.

6. Now move the mouse cursor precisely on top of the fiducial mark,
and click. The co-ordinate values for this fiducial mark are entered
into the co-ordinate list below the image, and the list cursor jumps
to the next row.
7. Clicking on the Fit View toolbar button should give you back the over-
view of the full image, ready to proceed to the next fiducial-mark
measurement.

Repeat for every fiducial mark, until all marks (four or eight) have
been measured.
Then

Ó »  Š.î á
 Interior orientation with e-foto
73

D F IGURE 7.7. Interior orientation report, showing residuals.

1. Press the toolbar icon Interior Orientation. Note, in the output, the
standard error of unit weight (also known as root-mean-square
error or RMSE) σ0 . Your experience with this imagery should tell
you how small this number can be if everything is correct, usually
of order 0.01 mm.
2. Press the toolbar icon View report, and the tab “V”. This will display
the residuals of the fiducial mark co-ordinates. They should all be
suitably small.
3. If you are happy, press Accept.

D 7.3.3 Completing interior orientation

◦ You have to repeat the interior orientation procedure for all images
in your set.
◦ After every image, or whenever you need to interrupt your work,
remember to save your intermediate results by exiting Interior Orien-
tation (by closing its main window by clicking on the little cross in
the corner) and then, in the main menu, clicking Project ▷ Save file.
◦ You may interrupt your work at any time. Interior orientation
results that have been approved by clicking Accept in the Interior
Orientation report, and after that, saved in the Project menu, will be
preserved.
◦ If you are unhappy about the quality of some of your work (it
improves with training!), you may return to Interior Orientation, open

Ó »  Š.î á
74 I NTERIOR ORIENTATION

the problem image, re-do the fiducial marks you are unhappy with
— you may set the fiducial mark to be updated by clicking on the
appropriate row in the fiducial mark co-ordinate list — , re-accept
and re-save.

Under the Project main menu item in the Images tab you can see which
images have their interior orientation completed and saved into the
project file, figure 6.8.

D 7.4 Debugging interior orientation

It is rather likely that, at first attempt, your interior orientation may
not succeed and will produce a much too big value for σ0 , the root-mean-
square error. There may be several reasons for this:
1. One or more of the values you entered from the camera calibration
document into the co-ordinate table of fiducial marks (table 4.1) is
in error.
2. Your measurement of the image co-ordinates of one or more of these
fiducial marks is in error.
3. Your identification of the fiducial marks — i.e., the system of num-
bering used — was wrong, and the values you measured in the
image, and the values tabulated in the calibration table, do not
correspond. This will lead to huge residuals; check carefully that, in
the images, the fiducial marks are indeed numbered as you assumed
when measuring.
Carefully go through the numbers for each fiducial mark again, starting
with the point producing the largest residuals. The RMSE value σ0
should be small, e.g., order 0.01 mm for the Finnish imagery; don’t settle
for anything bigger than this. Only once repeated attempts to bring the
largest residual further down fail, is it time to move on. The quality of
the results from the whole project depends on you getting this stage right.

Ó »  Š.î á
D Exterior orientation

8
Exterior orientation, or spatial resection, refers to establishing the con-
nection between terrain co-ordinates — three-dimensional, say ( X , Y , Z ) ,
with the Z axis pointing up — and the camera co-ordinates ( x, y) for one
given image. It requires a sufficient number of ground control points
i = 1 . . . n to be given both in the terrain, ( X i , Yi , Z i ) , i = 1 . . . n, and on
the photographic plate, (ξ i , η i ) , i = 1 . . . n, or equivalently, in camera
co-ordinates ( x i , yi ) .

D 8.1 Theory
The unknowns, or elements, or parameters, to be recovered by exterior
orientation are:
◦ The location co-ordinates ( X 0 , Y0 , Z0 ) of the camera, in the same
terrain co-ordinate system also used for the ground control points.
◦ The camera orientation angles (ω, ϕ, κ) also relative to the axes
( X , Y , Z ) of the terrain co-ordinate system, defined by a “corkscrew
rule”: ω rotates, around the X axis, Y towards Z ; ϕ rotates, around
the Y axis, Z towards X ; and κ rotates around the Z axis, X
toward Y . The ( X , Y , Z ) axes triad is right-handed, i.e., a corkscrew
turning from X to Y moves forward along the Z axis.
In the usual situation in aerial photography, where the camera looks
straight down and the terrain underneath has only small height (ele-
vation) differences, one can show that while recovery of the camera’s
height co-ordinate Z0 , and its rotation angle κ around the Z axis, can be
done precisely, both the horizontal co-ordinates X 0 , Y0 and the tilt angles
ω, ϕ can not be recovered very precisely: the recovered values are much
more uncertain and are correlated or anti-correlated, X 0 with ϕ, Y0 with
ω. It is only because aerial cameras have such a wide view angle, that

– 75 –
76 E XTERIOR ORIENTATION

Z
κ

Image
y
Focal length c

ω ϕ

Optical centre X 0 , Y0 , Z0
x

Terrain

F IGURE 8.1. Exterior orientation elements. These together describe the re-
lationship between the location of a terrain point in terrain
co-ordinates, and the corresponding image point in camera co-
D ordinates.

a reasonable recovery of all the exterior orientation elements is at all

possible.
The exterior orientation situation is depicted for two image acquisition
moments in figure 8.2. Note that, as we have six unknown parameters
to determine, the co-ordinates of three points should be sufficient, and
four points is already overdetermined, allowing not only a solution, but a
judgment on its quality as well, in the form of the variance of unit weight
σ20 coming out of the least-squares estimation computation.

D 8.2 Observation equations

The observation equations for exterior orientation are referred to as the
collinearity equations.

Ó »  Š.î á
 Observation equations
77

F IGURE 8.2. Exterior orientation of two images. A square in the terrain is

depicted as a square or as a trapezoid, depending on whether the
D camera is looking down vertically (right) or obliquely (left).

We derive these equations by looking at a terrain point — X in terrain

co-ordinates — and the corresponding image point. They are connected
by a straight line through the camera optical centre, which, in terrain
co-ordinates, is given by the vector
⎡ ⎤
X0
X0 = ⎣ Y0 ⎦ .
⎢ ⎥

Z0
Similarly the image point is given in the camera co-ordinate frame:
⎡ ⎤
x
x = ⎣ y ⎦,
⎢ ⎥

c
where c is the calibrated focal length, and x, y the measured image co-
ordinates after reduction for camera calibration and interior orientation.
Now, in case the camera is looking straight down and the aircraft is
flying due East — with the camera operator sitting on starboard! —,
there is the following relationship between these:
c c
x=− (X − X0 ) = (X − X0 ) ,
Z0 − Z Z − Z0

Ó »  Š.î á
78 E XTERIOR ORIENTATION

where Z is the height of the terrain point above sea level, and Z0 that of
the exposure optical centre above sea level. Thus, Z0 − Z is the height
def /
of the camera above the terrain. The ratio s = 1 : S = c ( Z0 − Z ) is an
important number called the scale factor of the exposure. The inverse
number S = 1 s , typically a large number, is called the scale number.
/

Of these three equations, only two are useful observation equations:

X − X0
x = c ,
Z − Z0
Y − Y0
y = c .
Z − Z0

In reality of course, the camera has an arbitrary, varying orientation

relative to terrain co-ordinates, and we introduce camera-rotated terrain
co-ordinates X of the terrain point, as follows1 :

X = R (X − X0 ) ,

where R is the rotation matrix between terrain and camera co-ordinate

axes, as follows:
⎡ ⎤ ⎡ ⎤
r 11 r 12 r 13 eT1
R = ⎣ r 21 r 22 r 23 ⎦ = ⎣ eT2 ⎦ = R (ω, ϕ, κ) .
⎢ ⎥ ⎢ ⎥

r 31 r 32 r 33 eT3

Here, we also expressed the rows of the matrix as column vectors

e1 , e2 , e3 . As a rotation matrix is orthogonal (see appendix A.3), these
vectors are unit vectors, and mutually ortogonal:

∥e1 ∥ = ∥e2 ∥ = ∥e3 ∥ = 1, ⟨e1 · e2 ⟩ = ⟨e1 · e3 ⟩ = ⟨e2 · e3 ⟩ = 1.

So, they form an orthonormal basis {e1 , e2 , e3 } of the space. They are in
fact oriented along the axes of the ( x, y, c)camera co-ordinate frame, as
depicted in figure 8.1.
Now, for a realistic, (ω, ϕ, κ)-rotated camera we obtain the collinearity
equations:

X ⟨e1 · (X − X0 )⟩
x=c =c =
Z ⟨e3 · (X − X0 )⟩
r 11 (ω, ϕ, κ)( X − X 0 ) + r 12 (ω, ϕ, κ)(Y − Y0 ) + r 13 (ω, ϕ, κ)( Z − Z0 )
=c ,
r 31 (ω, ϕ, κ)( X − X 0 ) + r 32 (ω, ϕ, κ)(Y − Y0 ) + r 33 (ω, ϕ, κ)( Z − Z0 )

1 The origin of these new co-ordinates is the camera optical centre, so formally one could
say that X0 = 0.

Ó »  Š.î á
 The rotation matrix
79
Y ⟨e2 · (X − X0 )⟩
y=c =c =
Z ⟨e3 · (X − X0 )⟩
r 21 (ω, ϕ, κ)( X − X 0 ) + r 22 (ω, ϕ, κ)(Y − Y0 ) + r 23 (ω, ϕ, κ)( Z − Z0 )
=c .
r 31 (ω, ϕ, κ)( X − X 0 ) + r 32 (ω, ϕ, κ)(Y − Y0 ) + r 33 (ω, ϕ, κ)( Z − Z0 )
(8.1)

D 8.3 The rotation matrix

The rotation matrix R can be written in terms of the Euler angles ω, ϕ, κ,
which serve as the actual unknowns describing camera orientation. This
is done as follows:
⎡ ⎤
cos ϕ cos κ sin ω sin ϕ cos κ + cos ω sin κ − cos ω sin ϕ cos κ + sin ω sin κ
R = ⎣ − cos ϕ sin κ − sin ω sin ϕ sin κ + cos ω cos κ cos ω sin ϕ sin κ + sin ω cos κ ⎦ .
⎢ ⎥

sin ϕ − sin ω cos ϕ cos ω cos ϕ

(8.2)
Substitution now yields the exact collinearity equations:
f fy
x = c gx , y=c g , (8.3)

with the definitions2

e

⎡ 1 ⎤ ⎡ ⎤
⟨ cos ϕ cos κ X − X0
⟩
def
f x = ⎣ sin ω sin ϕ cos κ + cos ω sin κ ⎦ · ⎣ Y − Y0 ⎦ =
⎢ ⎥ ⎢ ⎥

sin ω sin κ − cos ω sin ϕ cos κ Z − Z0

= cos ϕ cos κ ( X − X 0 ) + (sin ω sin ϕ cos κ + cos ω sin κ)(Y − Y0 ) +
+ (sin ω sin κ − cos ω sin ϕ cos κ)( Z − Z0 ) ,
e
⎡ 2 ⎤ ⎡ ⎤
⟨ − cos ϕ sin κ X − X0
⟩
def
f y = ⎣ cos ω cos κ − sin ω sin ϕ sin κ ⎦ · ⎣ Y − Y0 ⎦ =
⎢ ⎥ ⎢ ⎥

cos ω sin ϕ sin κ + sin ω cos κ Z − Z0

= − cos ϕ sin κ ( X − X 0 ) + (cos ω cos κ − sin ω sin ϕ sin κ)(Y − Y0 ) +
+ (cos ω sin ϕ sin κ + sin ω cos κ)( Z − Z0 ) ,
e

⎡ 3 ⎤ ⎡ ⎤
⟨ sin ϕ X − X0
⟩
g = ⎣ − sin ω cos ϕ ⎦ · ⎣ Y − Y0 ⎦ =
⎢ ⎥ ⎢ ⎥

cos ω cos ϕ Z − Z0
= sin ϕ ( X − X 0 ) − sin ω cos ϕ (Y − Y0 ) + cos ω cos ϕ ( Z − Z0 ) .

2 The rows of the matrix R are here written as column vectors, just in order to save
horizontal page space.

Ó »  Š.î á
80 E XTERIOR ORIENTATION

In the general case, this formulation has the problem of becoming

unstable for certain combinations of angles. This is fortunately not the
case for standard aerial vertical photogrammetry, where the angles ϕ
and ω are close to zero. Then, we have

cos ϕ ≈ cos ω ≈ 1, sin ϕ ≈ ϕ, sin ω ≈ ω, sin ϕ sin ω ≈ 0.

For the matrix above this means

⎡ ⎤
cos κ sin κ ω sin κ − ϕ cos κ
R ≈ ⎣ − sin κ cos κ ϕ sin κ + ω cos κ ⎦ .
⎢ ⎥

ϕ −ω 1

With this, the above equations 8.3 become

cos κ · ( X − X 0 ) + sin κ · (Y − Y0 ) + (ω sin κ − ϕ cos κ)( Z − Z0 )

x≈c , (8.4)
ϕ · ( X − X 0 ) − ω · (Y − Y0 ) + ( Z − Z0 )
− sin κ · ( X − X 0 ) + cos κ · (Y − Y0 ) + (ϕ sin κ + ω cos κ)( Z − Z0 )
y≈c . (8.5)
ϕ · ( X − X 0 ) − ω · (Y − Y0 ) + ( Z − Z0 )

This approximate simplification saves a lot of drudge when computing

the partial derivatives for the design matrix A .

D 8.4 Linearization
The expressions 8.4, 8.5 need to be linearized with respect to all six
unknowns X 0 , Y0 , Z0 , ω, ϕ, κ. Doing this requires determining 2 × 6 partial
derivatives ∂ x ∂ X . . . ∂ y ∂κ . Deriving these partials, which form the
/ /
0
design matrix A , is a lot of painstaking work but not really difficult3 .
The observation equations system (for a single point’s two camera co-
ordinates) then becomes

∆ X̂ 0
⎡ ⎤
A
⎡   ⎢ ∆Ŷ0 ⎥
⎢ ⎥
∂x ∂x ∂x ∂x ∂x ∂x ⎤ ⎢
∆x + vx ⎢ ∆ Ẑ0 ⎥ ,
⎥
∂ X0 ∂Y0 ∂ Z0 ∂ω ∂ϕ ∂κ ⎦ ⎢ ⎥
=⎣ ∂y ∂y ∂y ∂y ∂y ∂ y ⎢ ∆ω̂ ⎥ (8.6)
∆y + vy
∂X0 ∂Y0 ∂ Z0 ∂ω ∂ϕ ∂κ ⎢
⎢ ⎥
⎣ ∆ϕ̂ ⎦
⎥

∆κ̂

in which the ∆ (Delta) symbols refer to the differences of both observa-

tions x, y and unknowns X 0 . . . κ from their assumed approximate values.

3 It can even be done numerically, avoiding the risk of error in an analytical derivation.
This is a useful correctness check on your code.

Ó »  Š.î á
 Computing the partials
81
The matrix of partial derivatives — to be evaluated at best available
approximate values for the unknowns — is the design matrix A :
⎡ ∂x ∂x ∂x ∂x ∂x ∂x ⎤
∂X0 ∂Y0 ∂ Z0 ∂ω ∂ϕ ∂κ ⎦
A=⎣ ∂y ∂y ∂y ∂y ∂y ∂y . (8.7)
∂X0 ∂Y0 ∂ Z0 ∂ω ∂ϕ ∂κ

In eq. 8.6 we have marked the observations ∆ x, ∆ y and their residuals

v x , v y as stochastic by adding an underline, and the unknowns ∆ X̂ 0 . . . ∆κ̂
as estimators by adding a hat, as is formally correct in observation
equations.

D 8.5 Computing the partials

Obtaining expressions for the partial derivatives making up the design
matrix A , eq. 8.7, is done by applying the quotient rule of differentiation
to equations 8.4, 8.5: given two functions f ( p) , g ( p) of some parameter
p, we have
d f 1 df f dg
( )
dp g = g d p − g2 d p .
We call, like above in equation 8.4,
def ( )
f x = c cos κ · ( X − X 0 ) + sin κ · (Y − Y0 ) + (ω sin κ − ϕ cos κ) · ( Z − Z0 ) ,
def
g = ϕ · ( X − X 0 ) − ω · (Y − Y0 ) + ( Z − Z0 ) ,
.
so that x = f x g . Now we obtain, e.g., for p = X 0 :
/

∂x 1 f
= − g c cos κ + x2 ϕ,
∂X0 g
and for p = κ (note that g does not contain κ):
∂x 1
= − g c (sin κ ( X − X 0 ) + cos κ (Y − Y0 ) + (ω cos κ + ϕ sin κ)( Z − Z0 )) ,
∂κ
and so forth. . .

D 8.6 Full observation equations

As we see in eq. 8.6, a single ground control point photographed in the
image produces two observation equations, with two observations x and
y, but no less than six unknowns X 0 , Y0 , Z0 , ω, ϕ, κ. The design matrix A
containing the partial derivatives has size 2 × 6, two rows and six columns,
eq. 8.6, which we write like this:

ℓ + v = A x̂,

Ó »  Š.î á
82 E XTERIOR ORIENTATION

with
[ ] [ ] ⎡ ∂x ∂x ∂x ∂x ∂x ∂x ⎤
∆x vx ∂X ∂Y0 ∂ Z0 ∂ω ∂ϕ ∂κ ⎦
ℓ= , v= , A = ⎣ ∂ y0 ∂y ∂y ∂y ∂y ∂y ,
∆y vy
∂X0 ∂Y0 ∂ Z0 ∂ω ∂ϕ ∂κ

and4 [ ]T
x̂ = ∆ X̂ 0 ∆Ŷ0 ∆ Ẑ0 ∆ω̂ ∆ϕ̂ ∆κ̂ .

Now of course in reality we have more than one ground control point: if
we have p control points, we will have an observation vector of length
n = 2 p, and the A matrix will be have two rows for each point, being
2 p × 6 in size. If 2 p ≥ 6, i.e., at least three control points, the problem will
be well posed, and for 2 p > 6 even overdetermined.
As an example we give the observation matrix for the case of four
ground control points:

ℓ1
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
v1 A1
⎢ ℓ ⎥ ⎢ v ⎥ ⎢ A ⎥
⎢ 2 ⎥ ⎢ 2 ⎥ ⎢ 2 ⎥
⎥+⎢ ⎥=⎢ ⎥ x̂,
⎣ ℓ 3 ⎦ ⎣ v3 ⎦ ⎣ A 3 ⎦
⎢

ℓ4 v4 A4

where
⎡ ∂x ∂xi ∂xi ∂xi ∂xi ∂xi ⎤
[ ] [ ] i
∆xi v x,i ∂ X ∂Y0 ∂ Z0 ∂ω ∂ϕ ∂κ ⎦
ℓi , vi , A i = ⎣ ∂ y0 ∂ yi ∂ yi ∂ yi ∂ yi ∂ yi ,
∆ yi v y,i i
∂X0 ∂Y0 ∂ Z0 ∂ω ∂ϕ ∂κ

and x̂ as before, with i the ground control point number.

D 8.7 Initial approximate values for exterior orientation

parameters

One might think that the above observation equations 8.4 and 8.5 could
be further simplified, by the omission of ω and ϕ (being ≈ 0) to

cos κ · ( X − X 0 ) + sin κ · (Y − Y0 )
x≈c , (8.8)
( Z − Z0 )
− sin κ · ( X − X 0 ) + cos κ · (Y − Y0 )
y≈c . (8.9)
( Z − Z0 )
However, this will not work because then, in the linearized observation
equations, all the partial derivatives in the columns belonging to ϕ and

4 Here, we write the column vector x̂ as a transposed row vector, in order to save vertical
page space.

Ó »  Š.î á
Initial approximate values for exterior orientation parameters
83
ω will vanish, and therefore it will not be possible to estimate values for
∆
ˆϕ and ∆
ˆω — and the least-squares solution will not converge for these
unknowns!
However, these equations are useful nevertheless: they allow us to
derive a set of crude approximate values for the exterior orientation
parameters, that can be used to start their iterative improvement. We
write the simplified equations 8.8, 8.9 in matrix form:
[ ] [ ][ ]
x c cos κ sin κ X − X0
=
y Z − Z0 − sin κ cos κ Y − Y0

def
or, inverted, with the scale number S = 1 s = Z − Z0 c :
/ /

[ ] [ ][ ]
X − X0 cos κ − sin κ x
=S ,
Y − Y0 sin κ cos κ y

or [ ] [ ] [ ][ ]
X X0 S cos κ −S sin κ x
= + ,
Y Y0 S sin κ S cos κ y

Ó »  Š.î á
84 E XTERIOR ORIENTATION

or, elaborately rearranging terms5 ,

⎡ ⎤
X0
S cos κ
⎢ ⎥
[ ] [ ]⎢ ⎥
⎢ −S sin κ
⎢ ⎥
X 1 x y 0 0 0 ⎢ ⎥
⎥.
= (8.10)
Y 0 0 0 1 x y ⎢
⎢ Y0 ⎥
⎥
⎣ S sin κ
⎢ ⎥
⎦
S cos κ

We recognize eq. 8.10 as a set of observation equations, with

[ ]
X
ℓ=
Y

the “observations”: the horizontal terrain co-ordinates of a ground control

point; the design matrix
[ ]
1 x y 0 0 0
A= ;
0 0 0 1 x y

and the vector of unknowns

[ ]T
x= x1 x2 x3 x4 x5 x6 =
[ ]T
= X 0 S cos κ −S sin κ Y0 S sin κ S cos κ .

5 An even simpler rearrangement would be

⎡ ⎤
[ ] [ ] X0
X 1 0 x −y
⎢ Y ⎥
0
=
⎢ ⎥
⎣ S cos κ
⎢ ⎥
Y 0 1 y x ⎦
S sin κ

with only four unknowns. Now

⎡ ⎤ ⎡ ⎤
x1 X0
⎢ x ⎥ ⎢ Y ⎥
⎢ 2 ⎥ ⎢ 0
x=⎢ ⎥=⎢
⎥
⎣ x3 ⎦ ⎣ S cos κ
⎥
⎦
x4 S sin κ

and

κ(0) = atan2 ( x4 , x3 ) ,
√
(0)
Z0 = Z + c x32 + x42 .

This way of writing the problem is actually very similar to solving the parameters of a
[ ]T
Helmert transformation in two dimensions, with X0 Y0 the translation vector, κ
the rotation, and the flight height Z0 as a “quasi-scaling”.

Ó »  Š.î á
 Exterior orientation using e-foto
85
We write these “observation equations” 8.10 for all ground control points
i = 1 . . . n within the image, and construct the ordinary least-squares
−1
solution x = ( A T A ) A T ℓ. Then, the initial approximate values for the
individual exterior orientation elements are extracted as follows:
(0)
X 0 = x1 ,
(0)
Y0 = x4 ,
(8.11)
κ(0) = atan2 (− x3 , x2 ) = atan2 ( x5 , x6 ) ,
√ √
(0)
Z0 = Z ± c x22 + x32 =Z±c x52 + x62 ,

and of course ϕ(0) = ω(0) = 0.

(0)
In the Z0 equation, the plus sign gives the physically realistic solu-
6
tion . Here, Z is the average terrain height Z : we have made the further
approximation that the terrain is relatively flat.
In the exterior orientation (“Spatial Resection”) code of e-foto, part of
the above expressions 8.11 are applied for setting the initial values of the
parameters.

D 8.8 Exterior orientation using e-foto

Starting the e-foto software brings up the start screen, figure 2.6. The
main menu, figure 2.7, offers the options Project, Execute and Help. Using
the Exterior Orientation module requires that you load a project file created
and saved by the Project Manager module, and modified by the Interior
Orientation module. If you haven’t done so yet, we refer you to the previous
instructions on the Project Manager and Interior Orientation modules.
1. Select the Project menu item from the main menu.
2. Load the project definition file previously saved from the Interior
Orientation module using one of the options “Load file”, “Last project”,
or “Open file”, depending on e-foto version. The file has an extension
.epp. See figure 7.3.

If you select the project file, the screen shown in figure 6.1 will come up.
Now choose Execute ▷ Spatial Resection. You’re ready to go!
The main screen that comes up is in figure 8.3. Note the toolbar, figure

6 The fact that, mathematically, there is also a negative solution, explains why one
sometimes sees near-convergence to an otherwise sensible-looking solution, with small
residuals, but a negative flight height. . .

Ó »  Š.î á
86 E XTERIOR ORIENTATION

F IGURE 8.3. Exterior Orientation main screen. Note the toolbar on top and the
D list of terrain points at the bottom.

8.4. In the point co-ordinate table, figure 8.5, ones sees the ground control
points with originally only their terrain co-ordinates given.

D 8.9 Doing exterior orientation

Exterior orientation proceeds now as follows:
1. Enter the terrain point co-ordinates into a text file of suitable
format, and load it into e-foto’s Project Manager module.

Fit view Set flight

Measure mark
Zoom Exterior orientation View report

D F IGURE 8.4. Exterior Orientation toolbar.

Ó »  Š.î á
 Doing exterior orientation
87

D F IGURE 8.5. Point co-ordinates list.

2. Obtain a paper print of the photos or a map of the area on which

the points are marked, so you can find them from the image.
3. For every point marker:

(a) find it on the image on your screen using the information pro-
vided. It typically looks like a white square made of cardboard
or painted on the roadtop, see the left picture in figure 8.7.
Alternatively, the point is not clearly marked in the terrain
but identified by a description, see the right picture in this
figure.
(b) Zoom if necessary using the Zoom toolbar button: click and
drag the mouse to form a yellow outline that will be magnified
(c) activate point marking by clicking the “Measure mark” icon on
the left of the toolbar
(d) check once more that the point you are about to measure is
the active row in the point co-ordinate table; click on the row
number to activate the correct row if necessary
(e) place the crosshairs precisely over the point marker; if it is
shown as big pixels, estimate the centre visually
(f) click.
(g) Like in interior orientation, also here you can correct a wrong
measurement by activating the row for the point in the co-
ordinate table, and re-measuring.

4. Now, you should do the exterior orientation computation. The icon

for this will be greyed out; you should first tell the system the
direction of flight associated with your images. Click the icon with
a little aircraft and an arrow (Set flight), and follow instructions.
You are presented with a little turnable dial (figure 8.6), the title of
which refers to the “flight direction”.

Ó »  Š.î á
88 E XTERIOR ORIENTATION

D F IGURE 8.6. The ”flight direction” — more accurately, “camera azimuth” — dial.

The angle that the dial sets is κ(0) , the initial or approximate value
of the κ angle of camera orientation around the Z axis.
This again is defined as the angle, counted positively (counterclock-
wise) from the terrain co-ordinate frame’s X axis — which the e-foto
software calls E , for Easting — and the camera co-ordinate frame’s
x axis (or, as the e-foto software calls it, ξ). This latter axis points to
the right edge of the image.
Now, whether the right edge of the image is on the side of the
nose or the tail of the aircraft will depend on whether the camera
operator sits on starboard or port with respect to the camera. . . as
follows:
◦ If the operator sits on starboard, facing port, and the upper
edge of the images is on port, κ will be the heading of the
aircraft, counted counterclockwise from the East.
◦ If the operator sits on port, facing starboard, and the upper
edge of the images is on starboard, then κ will be the aircraft
heading reckoned counter-clockwise from the West.
. . . but of course you can always just experiment, until you find a
κ(0) that works. . .
A better way of understanding κ is, that it is the angle over which
the images are rotated, in the positive or counter-clockwise direc-
tion, from the “standard orientation” where the upper edge of the

Ó »  Š.î á
Doing exterior orientation
89
image is North, and the right edge East7 . Perhaps “camera azimuth”
would be a suitable name.
Note: if you are using the e-foto version from June 30, 2016, it
doesn’t actually have a “flight direction” dial. For this version,
you have to rotate the images before input using a graphical
editor. You have to do this in the Project Manager’s Images
tab, see section 6.5. The angle of rotation is the same κ(0)
discussed above. This is necessary to achieve convergence of
the exterior orientation iteration.

5. Now the Exterior orientation button is no longer greyed out. Click

it, and check the results, figure 8.9. Again, the standard error
of unit weight σ0 tells you the general quality of the solution; it
should be small. Check also the residuals — tab V —; they too
should be small. And remember to check the unknowns — tab
Xa in the report: these are the position co-ordinates X 0 , Y0 , Z0 of
the camera optical centre, and the orientation angles ω, ϕ, κ of the
camera. These values should all be reasonable. Reasonable values
are, e.g., small values for two of the three angles (ϕ and ω), and a
positive, realistic value for flight altitude Z0 .
The iteration should converge within five steps or so. If the iteration
doesn’t converge in the ten steps allowed by default, there must be
something wrong.
If you see huge, all-zero, or otherwise implausible values appear for
any of the co-ordinates of the camera optical centre, this probably
means that you have too few terrain points, or terrain points in
unfavourable positions, or the flight direction dial setting wrong, or
some other thing wrong: carefully check all the data you entered
into the Project Manager module’s many tabs.
6. If you are happy with the result, press Accept. Exit the Spatial
Resection module (by clicking on the cross in the window’s corner
— or in later versions, on the green Done icon) and click Project ▷
Save file.

7 This means that e-foto’s way of calling this dial “flight direction” is misleading: it may
be precisely opposite to the direction of flight!

Ó »  Š.î á
90 E XTERIOR ORIENTATION

F IGURE 8.7. Detail with ground control point. Left, Helsinki, right, Rio de
D Janeiro.

F IGURE 8.8. Overview picture of all terrain points in this image. Only the
points marked green are activated and will be used for exterior
orientation. The red points are deactivated. There is a tick box
for activating or deactivating every point in the point co-ordinates
D list’s leftmost column, see figure 8.5.

Ó »  Š.î á
 Completing exterior orientation
91

F IGURE 8.9. Exterior orientation adjustment report, displaying unknowns es-

timates (left) and residuals (right). If the number of iterations
was insufficient for convergence, or you see large residuals, or
D implausible values in the Xa tab, do not accept yet.

D 8.10 Completing exterior orientation

You have to exteriorly orient every image in your set. However, you may
do it one image at a time, and save the accepted result to the project file.
It appears to be difficult in e-foto to re-do exterior orientation after the
result has been accepted and saved. Therefore it might be a good idea
to create a backup of your project definition file every time you complete
one image.
Alternatively, the project definition file, with extension .epp, is in fact
an XML file which is humanly readable. If you are very careful, you may
manually edit out the result of one image’s Exterior Orientation. We
will not give further instructions here, but what to do should be clear to
anyone who can read XML files. Remember that a small error may lose
you your work up to that point, so don’t do this if you are unsure.

D 8.11 About geometric strength

Exterior orientation is done on a single image at a time, using ground con-
trol points (GCP) which appear on the image. The precision with which
exterior orientation can be done and the EO parameters determined,
depends on the geometry of these points within the area being imaged.
This is depicted in figure 8.10. Note that three points is sufficient, but
has no overdetermination, so a gross error in any of the control point

Ó »  Š.î á
92 E XTERIOR ORIENTATION

x x x x x x
x x
xxx x
xxx
x x xx
x x x x x x

Barely enough Good Excellent Poor (1)

(no error detection) Poor (2)

D F IGURE 8.10. Exterior orientation precision and ground control point geometry.

co-ordinates will not be noticed: the solution is not robust. This is gener-
ally not acceptable. Four points offer a minimum of overdetermination,
provided no three points lie on a straight line.
On the right hand side of figure 8.10 are two examples of a near-singular
or ill-conditioned geometry:
◦ if all points are clustered together, a set of exterior orientation
parameters will be obtained that represents only the immediate
neighbourhood of this cluster well. At some distance, the quality of
the solution becomes rapidly poorer.
◦ If all points are on a straight line, then again close to this line the
solution will be good, but away from the line, its quality gets poorer
quickly.

D 8.12 Ground control points

Before an imaging flight, ground control points of known location must be
marked or signalized in the terrain so they become visible in the images
as marks that can be easily and precisely measured. Signal types used
are white discs or white crosses of sufficient size (depending on flying
height or target resolution of the project). The discs or crosses can be
painted on roadtop asphalt or manufactured from cardboard. Size will
depend on the specified ground resolution.
It is important that the shape of the signal is symmetric, and that it is
mounted symmetrically on the benchmark being signalized. Note that,
with care, it is possible to measure the position of the mark in the image
with sub-pixel accuracy.
Obtaining the co-ordinates of ground control points would nowadays
always be done with GNSS. The proper technique to use to obtain pre-
cise positions this way is static relative GNSS using geodetic-quality
(dual-frequency) receivers, with post-processing using geodetic GNSS

Ó »  Š.î á
 Ground control points
93

D F IGURE 8.11. Ground control point: centering the marker.

processing software and precise satellite ephemeris.

The GNSS measurements should be connected to points known in a
modern, nationally valid reference frame defined with the aid of space
geodetic techniques and connected to the international reference frame.
Any realization of the ITRS will do; so will a national system like the
Ethiopian Adindan datum, connected with ITRS through a conventionally
fixed translation vector. The connection can be made
1. by including data collected by continuously operating reference
stations (CORS) connected to the international network, or
2. by measuring on points the co-ordinates of which have already been
determined, similarly by geodetic GNSS, in this system.
Co-ordinates thus obtained would be in some realization of “WGS84” (or
connected to it in some simple way) and be centimetre-level geocentric.
They could also be readily converted to map projection co-ordinates, e.g.,
in the UTM projection.
In figure 8.11 is shown a magnification of the image around a control
point, that is marked on the ground with a white cross around the bench-
mark, a signal. Such signals are placed on control points in preparation
for the flight. One sees that the pixel size is quite large compared to the
size of the control point and its signal. Yet, it is important that the cursor
is placed as precisely on the true point location as possible.
The figure shows how it is done. It requires some training and a little
visual imagination. The middle sub-picture shows a replica in red of
the white ground cross placed as well as possible to coincide with the
pixellated cross in the image. In your mind’s eye you should place this
red cross on the image every time you mark a ground control point. Then,
you choose the centre point of the red cross, and click to place the marker,
see the rightmost sub-figure.
Note that for exterior orientation one must have a set of points with

Ó »  Š.î á
94 E XTERIOR ORIENTATION

known three-dimensional co-ordinates in the terrain, all within the area

appearing in this one image. On the other hand, photogrammetry by
aerotriangulation is a way of determining three-dimensional point co-
ordinates, but not from one image only; it requires two or more images
to be used together. This brings us to the next, extensive subject, that
of stereoscopic reconstitution of three-dimensional models from several
overlapping images, or eventually a block of images covering the entire
project area.

Ó »  Š.î á
D Aerotriangulation

9
D 9.1 Description
Aerotriangulation (also known as phototriangulation) is the joint adjust-
ment of a project set or block of aerial photographs, together with tie
point information from the terrain, to obtain a three-dimensional model
of the area mapped. It requires a number of preparatory steps, up to
interior orientation for each image, the co-ordinates of points with known
ground coordinates, so-called ground control points (GCP) visible in one
or more of the images, as well as tie points which can be identified in
overlapping images and serve to interconnect them.
Numerical aerotriangulation is typically performed by bundle block
adjustment, i.e., a least-squares adjustment where the projective rela-
tionships between image and terrain co-ordinates serve as observation
equations. An often used numerical technique, in view of the very large
system of equations formed, is the iterative technique known as con-
jugate gradients, steepest descent, or an implementation known as the
Levenberg-Marquardt algorithm. This algoritm is essentially1 a lin-
earized, iterative solution algorithm stabilized by a technique similar to
Tikhonov regularization.
Contrary to exterior orientation, which is done one image at a time,
aerotriangulation uses all images, and their ground control points, to-
gether. Therefore the situation may arise where some of the images
cannot be externally oriented on their own, but, thanks to the addition
of tie points, the block adjustment is successful. Much depends on the
geometry of the tie points chosen, as well as the overlap between the
images in the block, both in the flight direction and perpendicular to it.

1 https://en.wikipedia.org/wiki/Levenberg%E2%80%93Marquardt_algorithm

– 95 –
96 A EROTRIANGULATION

x x x x x

x x
x x x

x x x x

F IGURE 9.1. Tie point geometry in aerotriangulation. The crosses form a tie
point geometry that is sufficient; adding the circles will make it
excellent. Also drawn are four ground control points (GCPs) as
D triangles.

In figure 9.1 an aerial photography block is depicted where the overlap

in the flight direction, between two images in the same strip, is approxi-
mately 60%, and in the perpendicular direction, between neighbouring
strips, it is about 20%. These values are fairly typical and give good
results based on long experience.
Two tie point geometries are drawn:
1. a sufficient one in which there are four tie points between images
in the same strip, and two between images in neighbouring strips;
and
2. a great one where there are no less than six tie points between each
pair of images in the flight direction.
Of course, existing ground control points can do double duty as tie points,
and one should not choose tie points close to them. Good tie points
are, e.g., the corners of building roofs which are well defined in both
photographs. Again, we do not need to know the ground co-ordinates of
the tie points. It suffices that they are identical in both (or all three, four)
images in which they appear.

Ó »  Š.î á
 Observation equations
97
If the tie point geometry of a photogrammetric block is sufficient, or
even good, this means that the relative orientations of the images within
the block — that is, the relative locations of their exposure centres, and
the orientation angles of the cameras at exposure — are well constrained
and can be well determined by the aerotriangulation.
However, absolute orientation of the block as a whole requires in ad-
dition ground control points. See again figure 9.1, where four proposed
ground control points (GCPs) are drawn as triangles. Four points are suf-
ficient, as they are for the exterior orientation of a single image (chapter
8), and provide some redundancy. Theoretically, the absolute minimum
number of GCPs to constrain the six parameters of absolute orientation
of the block is three points not on a straight line. The lack of redundancy
makes this unacceptable, however.
Modern photogrammetric software like ERDAS Imagine can generate lots
of tie points automatically using advanced image correlation techniques.
Unfortunately e-foto does not have this capability.

D 9.2 Observation equations

The observation equations are similar to the ones we already derived
for exterior orientation. However, note that now there is a much greater
number of unknowns: every camera location on which an image was
taken now contributes three unknowns of place X 0i , Y0i , Z0i and three
unknowns of camera attitude ω i , ϕ i , κ i , where the index i = 1 . . . p is the
image number.
We can now write the observation equations as follows, giving the
example of two images A and B containing three ground control points,
one of which — point number 2 — appears in both images:

ℓ v A
⎡  ⎤ ⎡
  ⎤ ⎡  ⎤ x̂
ℓ1A v1A A 1A 0   ]
[
⎢ ℓ ⎥ ⎢ v ⎥ ⎢ A 0 ⎥ x̂
⎢ 2A ⎥ ⎢ 2A ⎥ ⎢ 2A A
⎥+⎢ ⎥=⎢ . (9.1)
⎥
⎣ ℓ2B
⎢ ⎥
⎦ ⎣ v2B ⎦ ⎣ 0 A 2B ⎦ x̂B
ℓ3B v3B 0 A 3B
  
12×1
        
8×1 8×1 8×12

Here, e.g., each observation ℓ ji stands for the linearized camera co-
ordinates, i.e., ∆ x ji , ∆ y ji , measured for ground control point j in image
i . It thus consists of two real-number elements, and the total ℓ vector

Ó »  Š.î á
98 A EROTRIANGULATION

on the left-hand side — as, similarly, the vector of residuals v — is eight

real-number elements tall.
The partial design matrix or submatrix for every ground control point
j and image i has now two rows and six columns, as follows:

∂ x ji ∂ x ji ∂ x ji ∂ x ji ∂ x ji ∂ x ji
⎡ ⎤
⎢ ∂X ∂Y0,i ∂ Z0,i ∂ω i ∂ϕ i ∂κ i ⎥
A ji = ⎣ ∂ y0,i ∂ y ji ∂ y ji ∂ y ji ∂ y ji ∂ y ji ⎦ .
ji
∂ X 0,i ∂Y0,i ∂ Z0,i ∂ω i ∂ϕ i ∂κ i

Here, the unlinearized camera co-ordinates — approximate, and only to

be used for doing the partial derivations! — are given by the camera
model equations

cos κ i ( X j − X 0,i ) + sin κ i (Y j − Y0,i ) + (ω i sin κ i − ϕ i cos κ i )( Z j − Z0,i )

x ji = c ,
ϕ i ( X j − X 0,i ) − ω i (Y j − Y0,i ) + ( Z j − Z0,i )
(9.2)
− sin κ i ( X j − X 0,i ) + cos κ i (Y j − Y0,i ) + (ϕ i sin κ i + ω i cos κ i )( Z j − Z0,i )
y ji = c ,
ϕ i ( X j − X 0,i ) − ω i (Y j − Y0,i ) + ( Z j − Z0,i )

according to equations 8.4, 8.5. As always, the partial derivatives must

be evaluated for the approximate values used in the linearization.
The full matrix A now consists of these submatrices of size 2 × 6. In
our example equation 9.1 there are four such submatrices, as well as four
submatrices of the same size 2 × 6 containing zeroes.
The vector of unknowns is, like before, containing linearized quantities:

∆X ∆X
⎡ ⎤ ⎡ ⎤
ˆ0 A ˆ0 B
∆Y ∆Y
⎢ ˆ0 A ⎥ ⎥ ⎢ ˆ0 B ⎥ ⎥
[ ] ⎢ ⎢
∆Z ∆Z
⎢ ⎥ ⎢ ⎥
x̂ A ⎢ ˆ0 A ⎥ ⎢ ˆ0 B ⎥
x̂ = , with x̂ A = ⎢ ⎥ , x̂B = ⎢ ⎥.
x̂B ⎢
⎢ ∆ω̂ A ⎥
⎥
⎢
⎢ ∆ω̂B ⎥
⎥
∆ϕ̂ A ⎦ ∆ϕ̂B ⎦
⎢ ⎥ ⎢ ⎥
⎣ ⎣
∆κ̂ A ∆κ̂B

Linearization means that we choose approximate values for the unknowns,

like in subsection 2.4.3:
(0) (0)
⎡ ⎤ ⎡ ⎤
( X 0)A ( X 0 )B
⎢ (Y0 )(A0) ⎢ (Y0 )(B0)
⎢ ⎥ ⎢ ⎥
⎥ ⎥
⎢ ⎥ ⎢ ⎥
(0) (0)
[ (0)
]
xA (0) ⎢ ( Z0 ) A (0) ⎢ ( Z0 )B
⎢ ⎥ ⎢ ⎥
x(0) = , with x A = ⎢ ⎥, xB = ⎢ ⎥.
⎥ ⎥
(0) ⎢ ω(0) ⎢ ω(0)
xB ⎢ A
⎥
⎥ ⎢ B
⎥
⎥
(0) (0)
⎣ ϕA ⎣ ϕB
⎢ ⎥ ⎢ ⎥
⎦ ⎦
(0) (0)
κA κB

Ó »  Š.î á
Observation equations
99
Then, e.g.,

ˆ0 A = ( X 0 ) + ∆ X (0)
X A
ˆ0 A ,
(0)
ϕ̂B = ϕB + ∆ϕ̂B ,

and so on. Now, the approximate values for the observations are obtained
by propagating the approximate unknowns through the non-linear, exact
camera model equations, based on the exact matrix R given in equation
8.2:
(0) (0)
(0) ( f x ) ji (0) ( f y ) ji
x ji = (0)
, y ji = (0)
,
g ji g ji

with

(0)
( f x ) ji =
⎡ (0) (0) ⎤ ⎡ (0) ⎤
⟨ cos ϕ i cos κ i X j − X 0,i ⟩
( ) ( ) ( ) ( ) ( ) ⎥ ⎢ ( ) ⎥
= ⎣ sin ω i 0 sin ϕ i 0 cos κ i 0 + cos ω i 0 sin κ i 0 ⎦ · ⎣ Y j − Y0,i0 ⎦ =
⎢
(0) (0) (0) (0) (0) (0)
sin ω i sin κ i − cos ω i sin ϕ i cos κ i Z j − Z0,i
( )
(0) (0) (0)
= cos ϕ i cos κ i X j − X 0,i +
( )( )
(0) (0) (0) (0) (0) (0)
+ sin ω i sin ϕ i cos κ i + cos ω i sin κ i Y j − Y0,i +
( )( )
(0) (0) (0) (0) (0) (0)
+ sin ω i sin κ i − cos ω i sin ϕ i cos κ i Z j − Z0,i ,
(0)
( f y ) ji =
⎡ (0) (0) ⎤ ⎡ (0) ⎤
⟨ − cos ϕ i sin κ i X j − X 0,i ⟩
( ) ( ) ( ) ( ) ( ) ⎥ ⎢ ( ) ⎥
= ⎣ cos ω i 0 cos κ i 0 − sin ω i 0 sin ϕ i 0 sin κ i 0 ⎦ · ⎣ Y j − Y0,i0 ⎦ =
⎢
(0) (0) (0) (0) (0) (0)
cos ω i sin ϕ i sin κ i + sin ω i cos κ i Z j − Z0,i
( )
(0) (0) (0)
= − cos ϕ i sin κ i X j − X 0,i +
( )( )
(0) (0) (0) (0) (0) (0)
+ cos ω i cos κ i − sin ω i sin ϕ i sin κ i Y j − Y0,i +
( )( )
(0) (0) (0) (0) (0) (0)
+ cos ω i sin ϕ i sin κ i + sin ω i cos κ i Z j − Z0,i ,
⎡ (0) ⎤ ⎡ (0) ⎤⟩
⟨ sin ϕ i X j − X 0i
(0)
g ji = ⎣ − sin ω(i 0) cos ϕ(i 0) ⎦ · ⎣ Y j − Y0i
(0) ⎥
⎦ =
⎢ ⎥ ⎢
(0) (0) (0)
cos ω i cos ϕ i Z j − Z0i
( )
(0) (0)
= sin ϕ i X j − X 0i −
( )
(0) (0) (0)
− sin ω i cos ϕ i Y j − Y0i +
( )
(0) (0) (0)
+ cos ω i cos ϕ i Z j − Z0i .

Ó »  Š.î á
100 A EROTRIANGULATION

Then, the linearized observations (“delta-observations”) are

(0)
[ ] [ ]
def ∆ x ji x ji − x ji
ℓ ji = = (0) ,
∆ y ji y ji − y ji

the input for the adjustment computation.

D 9.3 Tie points

D 9.3.1 Measurement

Tie points may be any points that show in two or more images and are
well defined. Often, roof corners of buildings make good tie points.
If you are manually measuring tie points, follow these simple rules:
1. Measure a set of tie points that — together with the ground control
points — are well spread out over the area to be mapped. It is
useless to create tie points that are too close together or too close to
a ground control point.
2. Only measure real points, uniquely defined in three dimensions.
A “point” that should not be used as a tie point is shown in figure
12.5, an intersection of road lines on two different levels. Also, don’t
measure points that might move between exposures: cars, animals,
shadows, trees waving in the wind.

D 9.3.2 Theory

If you introduce tie points, this will lead to the further introduction of
three co-ordinate unknowns for every tie point. These will appear in
the vector x̂ in equation 9.1, and corresponding elements will appear
in the design matrix A . Note that these elements, for tie point k, are
now obtained by partial differentiation with respect to the three tie-point
co-ordinates X k , Yk , Z k , where k appears in the place of j in equations
9.2:
ℓki + vki = A ki x̂k ,

with [ ]
∆ x ki
ℓki = ,
∆ yki
the camera co-ordinates for tie point k in image i ,
⎡ ⎤
X̂ k
x̂k = ⎣ Ŷk ⎦
⎢ ⎥

Ẑ k

Ó »  Š.î á
 Aerotriangulation in e-foto
101
the (unknown) terrain co-ordinates of the tie point, and
⎡ ∂x ∂ xki ∂ xki
⎤
ki
A ki = ∂ X k ∂Y k ∂Zk ⎦
∂ yki ∂ yki ∂ yki .
⎣
∂X k ∂Y k ∂Zk
For this, we need to partially differentiate

xki = c·
cos κ i ( X k − X 0i ) + sin κ i (Yk − Y0i ) + (ω i sin κ i − ϕ i cos κ i )( Z k − Z0i )
· ,
ϕ i · ( X k − X 0i ) − ω i · (Yk − Y0i ) + ( Z k − Z0i )
yki = c·
− sin κ i ( X k − X 0i ) + cos κ i (Yk −Y0i ) + (ϕ i sin κ i + ω i cos κ i )( Z k − Z0i )
· .
ϕ i · ( X k − X 0i ) − ω i · (Yk − Y0i ) + ( Z k − Z0i )

Every tie point thus produces two rows in the system of observation
equations for every image in which it is visible, and three elements in the
vector of unknowns irrespective of the number of images in which it is
visible. This means that the design matrix A also gains two rows per
image, and three columns.

D 9.4 Aerotriangulation in e-foto

We must construct a common three-dimensional model for several images,
in either a relative or an absolute co-ordinate frame. The model for doing
this contains as unknown parameters the six location and orientation
parameters X 0 , Y0 , Z0 , ω, ϕ, κ of the camera in the terrain co-ordinate
frame for every image used.

See figure 9.2 for the main window. Constructing the model requires
1. a sufficient number of known points or ground control points, the
terrain co-ordinates of which were loaded from file in the Project
Manager’s Points tab, and for which camera co-ordinates may have
been measured and saved in the Exterior Orientation module.
2. However, these points may or may not suffice for connecting each
image strongly enough with the other images to allow a strong
determination of the model parameters. Therefore, additionally we
may need to determine photogrammetric tie points, i.e., points that
can be seen in two or more images and thus “tie them together”.
These points should be evenly distributed throughout the overlap
area of the pictures. These tie points will be numbered successively

Ó »  Š.î á
102 A EROTRIANGULATION

D F IGURE 9.2. Phototriangulation main window.

after the ground control points: PF24, PF25, etc. Unlike control
points, tie points have no known co-ordinates; however, in the course
of the aerotriangulation, terrain co-ordinates will be determined
(estimated) for them, and these will be saved into the Project Manager
module’s Points list.
In order to do phototriangulation, one brings up the main screen by
choosing Phototriangulation from the Execute main menu. There are many
panels in the main window; only two images are displayed at a time, the

Set Results Show full

flight Mark Zoom image
directions
Move
Add tie point Calculate

D F IGURE 9.3. Aerotriangulation tool bar.

Ó »  Š.î á
 Aerotriangulation in e-foto
103

D F IGURE 9.4. Exterior Orientation elements obtained.

Left Image and the Right Image, which can be selected separately.
The user interface is very similar to that of Exterior Orientation: also
here, for some versions of e-foto, one first has to specify a “flight direction”
using a dial-like interface (figure 8.6), invoked by clicking on the aircraft-
with-arrow toolbar icon. Use here the same directions you used for the
individual photographs in Exterior Orientation2 . Note that, as a block of
images used for phototriangulation may consist of multiple strips in which
the aircraft flew in opposite directions, you may have to enter different
values for different images. Setting these directions correctly will make
the phototrangulation adjustment computation converge rapidly. See
figure 9.4, showing the exterior orientation elements computed after only
eight iteration steps.
After setting the flight directions, the computation icon — showing an
aircraft in the act of photographing the terrain — is no longer greyed out.
Click on it.
The module offers a window (figure 9.5) where you can choose which
images to take along in the computation, and which points, both ground
control and tie points. You can move any or all of these between the left
window — which lists what is available — and the right window — listing
what is actually used — before proceeding with the calculation.
You will have to choose the number of iterations allowed, as well as
the precision criteria applied to judge if the computation has converged,
separately for camera location co-ordinates and for camera orientation
angles. These usually do not have to be changed; only the number of
iterations may be increased if convergence is slow.
It may happen that the module complains that the geometry is not
strong enough, and that tie points need to be added; the toolbar offers
buttons for doing this.

2 And consider that, also here, the angle to be entered is κ0 , which may be either the
flight direction or the direction opposite to it. . .

Ó »  Š.î á
104 A EROTRIANGULATION

D F IGURE 9.5. Inclusion / exclusion of images and points.

Technical remark: e-foto’s Phototriangulation module will offer you the

option of using a local topocentric reference for doing its computa-
tions in. While this seems like a good idea, this feature is — in
some e-foto versions — buggy and you should not use it: choose
the “No” option. If accidentally you choose the “Yes” option, you
will obtain nonsensical values (very large, or all zero values, huge
RMSE, no convergence) for the exterior orientation elements. And
somehow, e-foto will remember this nonsense: you have to exit
the programme completely and start it again to be able to get
reasonable results.

D 9.5 Debugging an aerotriangulation adjustment

The end result of a successful phototriangulation is high quality exterior
orientation elements for all images in the block. One should especially
verify that
1. the height of the aircraft Z0 is realistic, and roughly the same for

Ó »  Š.î á
 Debugging an aerotriangulation adjustment
Ó
»  Š.î á

D F IGURE 9.6. Aerotriangulation point list. Points in two images are listed, with in the middle the list of all points (control points and tie points).

105
106 A EROTRIANGULATION

all images; in this example Brazilian case (figure 9.4) it is about

1318 m
2. the camera orientation angles ω, ϕ, κ (given here in degrees) are
realistic: ϕ and ω no more than a few degrees, and κ roughly the
same for all images in the same strip. For adjacent strips, one
expects the κ values to be opposite, i.e., κ′ = κ ± 180◦
3. the computation has converged, and
4. the root-mean-square error (RMSE) is acceptable (in the example
it is ±0.29 m, which is a little large). Do give allowance though for
a large block to produce a larger RMSE than the uncertainties of
individual ground control points in individual images.
If you are performing aerotriangulation on a large block with many
images, it is not likely that you will achieve a correct result on first try.
There will be errors in the input data:
1. The ground control point co-ordinates in the terrain reference sys-
tem of some of your ground control points may be wrong
2. The measurement of the image co-ordinates for these ground control
points may be wrong
3. Wrongly measured tie points
4. Something wrong in the e-foto set-up.
The important thing now is: don’t panic! Finding and eliminating the
error requires a systematic approach. First of all, you should realize
that the aerotriangulation computation does not need to be done on the
whole set of images (which may be very large); you can at first do the
computation on sub-sets of images, or even on individual photographs.
The latter case corresponds to exterior orientation (chapter 8) of that
photograph. It requires, for a robust result, the presence of at least four
ground control points, of which no three are positioned on one line.
Generally the requirements for processing sub-sets of photographs in a
way that will detect data errors are:
1. There should be four (or more) ground control points within the
area covered by the images together. These points need not be in
the overlap areas of two or more images. These should be spread
reasonably around the area, with no three points on one line.
2. There should be four or more common points in the overlap area
of each pair of images. These may be ground control points or tie

Ó »  Š.î á
Debugging an aerotriangulation adjustment
107
points. If this condition is not met, you should be measuring more
tie points in your image pairs.
It is clear from the above, that for successful debugging, it is good to
have an abundance of ground control points! So, include all points you
can find that have their terrain co-ordinates measured, and if you are
yourself responsible for measuring them, make a generous measurement
plan. The time spent on ground measurements will pay itself back when
processing the imagery.
1. Firstly, find the individual images that contain four or more ground
control points. Perform exterior orientation (chapter 8). At least
some of these images should give a correct result, as judged by the
root-mean-square error (RMSE). If not, i.e., if all images give poor
results, you must suspect that there is something wrong with the
general set-up of e-foto (chapter 6); go through everything carefully
again.
2. For those images that do not produce a good result, go through each
of the ground control points (GCP), checking that

(a) the given terrain co-ordinates in the GCP table are correct,
and
(b) the image co-ordinates measured in exterior orientation are
correct, and expecially that the points were all identified cor-
rectly in the image.

3. If that doesn’t help, you should remove each of the ground control
points in turn from the computation, to see if this gives a better
result. If you get a good result by removing a GCP, chances are
that it is precisely that point which is in error. This technique is
called “cross-validation”. Of course you can only do this if you have
a sufficient number of GCPs appearing on the image, i.e., five at
least3 .
4. Next, for those images for which the above treatment is inapplicable,
find pairs of images that contain four or more control points, and
a sufficient number of common points — measure additional tie
points if you need to. Perform the aerotriangulation computation
on those pairs. Again, your aim should be to get a correct result,

3 Because, if you only have four GCPs, removing one of them, any one, will leave you
with three points only, which will always allow exterior orientation to be performed
without producing any contradictions, even if errors are present. . .

Ó »  Š.î á
108 A EROTRIANGULATION

i.e., a result with a small RMSE. If not, go through the individual

ground points again, like described above, items 1-3.
5. Next, consider triplets or even larger sub-sets of images. Apply the
same treatment.
Finally, when all images have produced acceptable results, either indi-
vidually, in pairs, in triplets, or in larger sub-sets, you can attempt to
perform aerotriangulation on the whole block. The above describes a
debugging technique not unlike that used for finding errors in computing
code. Good luck!

Ó »  Š.î á
D Stereo restitution

10
Traditionally, aerial photographs have been acquired and processed in
pairs, allowing the formation of scenes that can be viewed in stereo by
the human eye. In the traditional photogrammetric approach, a pair of
photographs is placed inside a stereo restitution instrument that allows
them to be precisely oriented relative to each other, so that a stereoscopic
image would appear to the human eye. Then, the human operator would
be able by further mechanical manipulation to move a marker or cursor
within both pictures, effectively moving it three-dimensionally within the
viewed scene, and allowing it to be placed on chosen terrain details. The
mechanics of the stereo restitution instrument would then allow three-
dimensional co-ordinates to be recovered for plotting or map-making.
Today, this whole process is done digitally, using photographs in dig-
ital form. The formation of the 3-D scene is done computationally and
presented on a computer screen. Stereo vision however, the sensation
of depth and three-dimensionality, continues to require the human eye-
brain combination, although also there, computing technology is making
progress. . .

D 10.1 Relative orientation

For stereo restitution to work, one does not need to solve for the location
and orientation of the two camera locations used in any absolute sense.
One only needs to do a so-called relative orientation. This process, in-
volving a sufficient number of common points in the terrain seen in both
photographs, establishes the vector connecting the two camera locations
in space, and the camera orientations at both ends, but in a co-ordinate
frame that is still arbitrary and not connected to the terrain.
Note that the common points used in relative orientation do not need to

– 109 –
110 S TEREO RESTITUTION

have known co-ordinates. It suffices that they are visible in both images,
e.g., corners of roofs are well defined and typically used.

D 10.2 Absolute orientation

In absolute orientation we use the absolute terrain co-ordinates ( X , Y , Z )
of known points to obtain absolute camera locations and orientations
in the same, 3-D, terrain co-ordinate frame. This is not necessary for
successful stereo vision; it is however necessary for the successful mea-
surement of 3-D co-ordinates in a stereo pair, obtaining co-ordinates in a
known system.
Absolute orientation is very similar to exterior orientation for a single
camera: but here, seven orientation elements are to be determined, three
translations (shifts of the co-ordinate origin), three rotations, and the
length of the baseline (distance between the two camera locations) in the
model. And, contrary to exterior orientation, here the complete model with
its two camera positions and the baseline between them, is translated and
rotated in space. But just like in exterior orientation, also here, ground
control points with known co-ordinates are used in the computation. But,
as we have here three-dimensional model co-ordinates for these ground
control points, the number of condition equations is actually 3 p for p
points. This means that using three points will give an overdetermination
of 9 − 7, making the solution robust1 .

D 10.3 Feature extraction

D 10.3.1 Parallax elimination

In order to successfully extract features from a pair of images using stereo

vision, they need to be properly placed on the screen with respect to
each other. First of all they should be oriented correctly, i.e., the vector
between the two exposure centres of the image pair should have the same
direction as the vector connecting the pupils of your eyes.
Then, especially the parallax in the y direction (“up-down”) needs to be
removed by shifting one of the images. See figure 10.1.
This makes the baseline between the two camera positions in the stereo
model parallel with the line between your left and right eye pupils. Then,

1 One should still require these three points to not be on one line. More than three points
would even be better, and the wider they are distributed throughout the model area, the
better.

Ó »  Š.î á
 Feature extraction
111

F IGURE 10.1. Eliminating the y-direction parallax. In the top figure, there is a
large y parallax, and stereo vision is not possible. In the bottom
figure, the right side image has been moved in the y direction to
D eliminate the parallax.

if everything else is in order, you will suddenly see the stereo image
appear.
We can also describe this more formally: the epipolar lines in both
images must be brought into superimposition and parallel to the eye base,
see figure 10.2.
Also the x-direction parallax needs to be eliminated, but this is less
critical as the human eyes are more flexible in converging. See figure
10.3.
After eliminating the parallax in both directions, you can not only view
the image pair in stereo, but more importantly, you can view the cursor
in stereo. This is called the floating mark. While, on a personal computer
work station, the cursor can be moved left and right using the mouse, the
vertical motion within the model is typically tied to the mouse wheel2 .

2 In the old optomechanical photogrammetric workstations the two horizontal motions

were controlled by the left and right handwheels, wheras the vertical motion was

Ó »  Š.î á
112 S TEREO RESTITUTION

Epipolar lines Image 2

Image 1 Stereo base

Camera 2
Camera 1

Epipolar
plane

Terrain point

F IGURE 10.2. Epipolar plane and lines in a pair of aerial photographs. The
epipolar plane contains the ground point, the two optical centres
D and the two image points of the ground point.

The motion is implemented by shifting the cursor due right (+ x direction)

in the right image, and by the same amount due left (− x direction) in the
left image. The stereo impression to the human eye is, that the mark
moves vertically down, towards, and ultimately through, the terrain
surface.
It is important that you become comfortable with the floating cursor,

F IGURE 10.3. Eliminating the x-direction parallax. Left, before elimination;

D right, after elimination.

controlled by a footwheel. See figure 2.3 on page 11.

Ó »  Š.î á
 Using the Stereoplotter module in e-foto
113
and move it up and down to land softly on the terrain, while it is also
horizontally positioned on a chosen terrain feature. As you move around
the image, you will notice that the parallax elimination may have to be
repeated.

D 10.3.2 Features of interest

Now that we are set up to perform three-dimensional measurements in

our pair of images, we can use this to digitize visible features for mapping.
This is done by identifying the feature, recording its type and sub-type
according to a catalogue of features to be mapped, and measuring the
point or points making up the feature or its outline.
Typically features are categorized as points, lines, and polygons or
areal features. Points are features without any interesting extent, like
street lights. Lines typically are roads, canals, railroads, i.e., objects only
having an interesting linear extent connecting two locations. Polygons
again are all things with areal extent: buildings, parks, fields, . . . the
closed polygon being their outline.
Of course this classification may be dependent upon map scale: on a
small-scale map an entire city may become a point feature.
Extracting features:
◦ Land use and cover
◦ Vegetation or crop type
◦ Bedrock or soil type and texture, soil depth
◦ Landform type, slopes
◦ Drainage conditions, flood hazard
◦ etc.

D 10.4 Using the Stereoplotter module in e-foto

Note: before you can use the Stereoplotter module, you must have deter-
mined exterior orientation elements for all images of the project,
either by using the Exterior Orientation module, or by using the
Phototriangulation module.

The Stereoplotter module lets you extract sets of three-dimensional point

co-ordinates for features visible in an image pair. Features may be points,
lines, or polygons (i.e., surface areas, like buildings, land parcels etc.).

Ó »  Š.î á
114 S TEREO RESTITUTION

D F IGURE 10.4. The floating mark above, at, and below the object surface.

In order to register a feature point, be it a point feature or a single point

of a line or polygon feature, one must register camera co-ordinates in
two images. From these, three-dimensional terrain co-ordinates may be
computed if the camera’s location and orientation parameters are known,
e.g., by determination in the Exterior Orientation (Spatial Resection) module,
or in the Phototriangulation module.
Like in expensive commercial stereoplotters, also here one may move a
crosshair cursor in three dimensions through the terrain space, and place
it on a feature in three dimensions directly. On old optomechanical stereo
plotters, there were two handwheels for the horizontal X and Y motion,
and a footwheel for Z . On our personal computer it is simpler: the mouse
motion does X and Y and the mouse wheel does Z . Three-dimensional
navigation requires however stereo vision, which again requires that the
two images are placed on the screen approximately correctly with respect
to each other.

Important: in the Stereoplotter module one needs to be able to move

the left and the right image relative to each other, to remove any
parallax that would interfere with stereo vision. This is done by
holding down the mouse left button (Button 1) and at the same
time either the SHIFT key or the CTRL key on the keyboard.
First of all, click the Move button in the toolbar (the little hand; it
will only work with this cursor). After that
SHIFT-Mouse-1: drag the left image only

CTRL-Mouse-1: drag the right image only.

Before starting the work, you should do some basic settings, cf. Figure
10.6:
Pair: the image pair you are working on
Stereo mode: choose here Anaglyph. These are the inexpensive red and
cyan coloured glasses we use for stereo viewing (the alternative

Ó »  Š.î á
 Using the Stereoplotter module in e-foto
115

D F IGURE 10.5. Stereoplotter main window.

offered is polarized glasses; choose this only if your work station

emits polarized images and you have suitable glasses for viewing).

Do not touch the other settings. Direct ▷ Reverse may be useful if the
images were loaded the wrong way round (or you folded your cardboard
glasses the wrong way).
In the stereoplotter toolbar (figure 10.7) we have
◦ Zoom, which lets you place a little yellow rectangle around the area
you want to zoom into: mouse down for the top left corner, drag, and
mouse up again for the bottom right corner (using mouse button 1)
◦ Fit view, which will give you back a view on the full photographs
◦ Move, which lets you move the images in two directions by clicking

Ó »  Š.î á
116 S TEREO RESTITUTION

D F IGURE 10.6. General tools.

and dragging
◦ Mark lets you set a marker in either image. For accuracy, make sure
that the image is enlarged enough (by using Zoom).

D 10.5 Editing features

Then, we have the machinery for entering features, a set of three panes
depicted in Figure 10.8. What you should do first, when entering a
feature, is decide what type it is: point, linear (line), or covering an area
(polygon). If it is a line or a polygon, you will be offered a list of alternative
descriptors: if it is a line, is it a paved street, a railway, a river? Or if it is
a polygon, is it a building, a square, a park, or a lagoon? And finally, you
may give the feature a name. This is done in the middle pane; the data
entered appears in the left pane.
The right pane is a block of buttons explained in Tableau 10.1.

Mark Zoom Fit view

Move

D F IGURE 10.7. The stereo plotter toolbar.

Ó »  Š.î á
 Inserting a point
117
D T ABLEAU 10.1. Editing panel.

1. The standard workflow:

Add new features such as point, line
Add new feature
or polygon

Toggle. If in the down position,

Insert point mode indicates that new points will be
added to the current feature

Exit Stereoplotter Exit the software

2. Correcting and modifying your work:

Select a feature / point to be edited
Select feature / point
(“current feature”)

Remove the current point of the

current feature. The feature and
Remove point
point to be removed must be selected
in the leftmost pane.
Toggle. If in the down position,
Edit point mode indicates that the current points will
be modified

Remove feature Remove the current feature

End feature Stop editing the current feature

Delete all features (empties the

Remove all features
feature data base!)
3. Loading and saving work:
Load previously measured features
Load features
from file

Save current work, i.e., all features

Save features
currently in memory

Export features as
Export features in text format
text

D 10.6 Inserting a point

In e-foto, inserting a point in the corresponding locations in two images is
tricky. You should proceed in the following fashion.
1. First select the feature that you want to register, and enlarge it
sufficiently by zooming in (Zoom button on toolbar)

Ó »  Š.î á
118 S TEREO RESTITUTION

Remove feature

Add new feature Remove all features

End feature

Insert point mode

Edit point mode

Remove point

Exit Stereoplotter

Select feature / point Export features as text

Load features Save features

D F IGURE 10.8. Panels used in entering feature data.

2. Check that the two images in the anaglyph superposition are placed
correctly with respect to each other. Remove any parallax in the y
direction completely, and in the x direction enough, so that you can
comfortably see the stereo effect.
3. Select the feature’s type and sub-type, and give it a name (if you
don’t, e-foto will just number your features of each type sequen-
tially). It is a good idea to register all features of the same type
together, so you have to enter this data only once. It is entered into
the middle pane.
4. Activate the Mark button in the toolbar (the crosshair icon) and
place the crosshair cursor on the first point of the feature — in
three dimensions. You should see the crosshair “hanging” over the
terrain, in mid-air. Move it sideways by moving the mouse, and
vertically by rolling your mouse’s wheel.
5. Now, click to register the point.
6. Note that the three-dimensional terrain co-ordinates (“object co-
ordinates”) X , Y , Z are displayed immediately under the anaglyph
superposition.
After this, you can proceed to measure all the points of this feature,
assuming that they are all on the same height. For the roof of a building
this will be the case. For other objects this may not be assumed: e.g., a

Ó »  Š.î á
 Working example
119
street will generally vary in height. Then, you have to use the mouse
wheel to place the floating mark on every feature point before measuring
it.

D 10.7 Working example

As an example we use the first pair of images from Rio de Janeiro. We
select the railway station at Maracanã, and first of all, using the Move
cursor (the little hand) and the CTRL key, we eliminate the y direction
parallax and reduce the x parallax so that comfortable stereo viewing
becomes possible:

Next, we click on Add new feature and Insert point mode. Also, on the tool
bar the crosshair (“Mark”) button should be selected.
Position the crosshairs on the first corner of the station building using
the mouse, move the mark vertically using the mouse wheel until it is at
the same level as the station roof, and click.

Repeat the procedure by moving the crosshair cursor to the second

corner, adjusting height with the mouse wheel as necessary, and click.

Ó »  Š.î á
120 S TEREO RESTITUTION

Same for the third point. . .

. . . and the fourth.

Terminate the feature by clicking the End feature button:

Now you should have the four point co-ordinate sets in the feature list:

Ó »  Š.î á
 Working example
121

F IGURE 10.9. A bigger sample of stereo plotter production showing several

D urban features in the Rio area.

Note that these are three-dimensional terrain co-ordinates in the refer-

ence frame defined by the ground control points used in Exterior Orientation
or Phototriangulation3 .
You may choose to save this data to a disk file for future reference using
the Save features button. Alternatively, Export features as text creates a text
file that can be loaded and edited in an ordinary text editor.
In order to speed up the motion of the crosshair cursor, which may be a
bit slow on small machines, you may want to disable the Detailview pane
in the user interface. Additionally you can remove the Overview pane and
make the main window smaller.
Technical remark: many people find working with anaglyphic glasses
cumbersome, unpleasant, or imprecise. Some people don’t even
see stereo with them. Then, if you don’t have access to a work
station with special hardware (e.g., using polarization), it is a

3 You may verify this by measuring a ground control point as a point feature.

Ó »  Š.î á
122 S TEREO RESTITUTION

poorly guarded secret that with a little training, one can place the
floating mark three-dimensionally even without any glasses at all.
One uses, instead of anaglyphic glasses, the “Detailview” window
on the right hand side of the screen. It enlarges the same detail
seen in the left and right images, and, after precisely eliminating
the y parallax (by SHIFT-Mouse-1 or CTRL-Mouse-1 with the hand
cursor), one uses the mouse wheel to position the twin cursors on
the same point in both images. Consider, though, that the refresh
of the Detailview window is a bit slow, so give it time.
The method can be speeded up by using the circumstance that the
points of a rooftop outline are usually all on the same level. This
means that one can use the hand cursor to travel from point to
point without using the mouse wheel. This is how Figure 10.9 was
produced.

Ó »  Š.î á
D Measuring in aerial images

11
D 11.1 Image pyramids
It is typical in photogrammetric software that uses statistical analysis
techniques on its image data, to construct an image pyramid first. This
means that, in addition to the original image, an image is constructed at
half the resolution, occupying one quarter of the disk space; then, one at
one-quarter of the resolution; etc. See figure 11.1. One has to agree on a
technique to derive the pixel values for every layer of the pyramid from
the layer directly below it: simplest is just averaging 2 × 2 pixels. More
sophisticated methods exist too.

D F IGURE 11.1. An image pyramid.

– 123 –
124 M EASURING IN AERIAL IMAGES

Note that e-foto does not use image pyramids. The widely used ERDAS
LPS software however does.

D 11.2 Automatic feature matching by correlation

The statistical techniques that are typically used have to do with au-
tomatic feature matching. E.g., the automatic generation of tie points
in image pairs for aerotriangulation is done this way: the correlation
between two patches of the image is computed as1

Cov{∆ x, ∆ y}
Corr{∆ x, ∆ y} = √ , (11.1)
Var{∆ x} Var{∆ y}
with
ˆ d ˆ d
def
Cov{∆ x, ∆ y} = ( g 1 ( x, y) − g1 )( g 2 ( x + ∆ x, y + ∆ y) − g2 ) dx d y,
−d −d
ˆ dˆ d
( g 1 ( x, y) − g1 )2 dx d y,
def
Var{∆ x} =
−d −d
ˆ dˆ d
( g 2 ( x + ∆ x, y + ∆ y) − g2 )2 dx d y,
def
Var{∆ y} =
−d −d

in which g 1 ( x, y) , g 2 ( x, y) are the two images, g1 , g2 their averages

[ ]T
∆x ∆ y
over the whole patch, and is the shift vector between the two
images needed to maximize this correlation value. The value is always
in the interval [−1, 1]. What is typically seen is that, for most values of
the shift vector, the correlation is fairly close to zero, but for one value of
the shift vector, it becomes close to one. This means we have achieved a
match.
The patch over which the correlation is computed is only 2 d × 2 d pixels
in size, but still, for a typical aerial image pair, this is a quite massive op-
[ ]T
∆x ∆ y
eration, due to the large number of alternative shift vectors
that need to be inspected. One could limit the work by keeping d small,
but then you will see many “false positives”, high correlation values that
nevertheless do not mean a match has been found — and consider the
case where the image contains repeating features: this would easily fool
the algorithm. But then, so it would the human eye.

1 Folks who have studied statistics will notice the similarity with the statistical definition

of correlation: here, the place of the expectancy operator is taken by the double integral
over the patch area. The overbar notation used for average is also just shorthand for
that.

Ó »  Š.î á
 Least-squares matching
125
The pyramid approach helps here: perform the correlation calculation
first on the smallest image at the top of the pyramid, to find an approxi-
[ ]T
mate vector ∆ x1 ∆ y1 ; then, use this to limit the search in the next
lowest layer of the pyramid to the small square
[ ] [ ] [ ]
∆ x2 ∆ x1 i
= + , i, j = −1 . . . 1.
∆ y2 ∆ y1 j

In this way, the computational work is limited, even for a generous patch
semi-size d , and of order log n rather than n2 .

D 11.3 Least-squares matching

The correlation method can be refined by a method called least-squares
matching, where we allow, besides differences in brightness and contrast
between the patches (perhaps due to changed lighting conditions between
exposures), also a geometric distortion in the form of an affine trans-
formation, which can account for the differences in deformations in the
images caused by terrain patch slopes.
The free parameters describing all these differences are optimized
in the least-squares sense. The model connecting the pixel brightness
patterns to be correlated g 1 , g 2 in patches of the two images is not, as
in section 11.2, g 1 ( x, y) and g 2 ( x + ∆ x, y + ∆ y), but the more complicated
(Silveira et al., 2011)

g 1 ( x, y) and α g 2 (ax + b y + ∆ x, cx + d y + ∆ y) + β. (11.2)

Here ∆ x, ∆ y are the co-ordinate offsets between the centres of the patches,
α (alpha) describes the difference in contrast, and β (beta) the difference
in brightness. The parameters a, b, c, d together describe the geometrical
affine transformation.
Once homologuous point pairs have been identified, one can search
for nearby points by recursively applying the same procedure, stepping
in four different directions. The range of ∆ x, ∆ y values to look at will
be limited if the terrain is smooth. In this way, the whole area can be
matched.

D 11.4 Scale Invariant Feature Transform (SIFT)

The SIFT method, Scale Invariant Feature Transform, was originally
designed by Lowe (1999). The method is too complex to explain here in

Ó »  Š.î á
126 M EASURING IN AERIAL IMAGES

F IGURE 11.2. Difference of Gaussians. Original image, Gaussian blur width 8,

Gaussian blur width 16, and their difference, presenting detail
only between these two resolution levels.
Adding together the difference images (bottom right) thus ob-
tained for all widths k = 1, 2, 4 . . . 2n will produce back the original
D image.

detail, but involves the building of a pyramid of images, produced from the
original image by applying a Gaussian blur of widths in increasing powers
of 2, and, subtracting in every step, the similarly constructed Gaussian
blur image for the one lower power of 2 (“difference of Gaussians”, see
figure 11.2). This neatly separates the details on the various spatial
scales, defined by a binary “scale number” k, into layers of a pyramid.
In this pyramid, there are three co-ordinates: the x and y pixel co-
ordinates of the original image, and the “blurring scale” k. A point is
declared a point of interest if this “difference-of-Gaussians” value assumes
a local minimum in this three-dimensional space.
An alternative variant of SIFT, using wavelets instead of differences of

Ó »  Š.î á
 Automatic feature recognition
127
Gaussians, is also popular. Wavelets are well suited for this hierarchical
decomposition by scale in powers of 2.

D 11.5 Automatic feature recognition

Another application of pyramids is automatic feature recognition. This
is more challenging and requires a library of known and recognizable
features. This is used, e.g., in automatic construction of digital elevation
models (DEMs) from image pairs, by recognizing edge-like features —
terrain ridges — which must at least be included in any DEM.

Ó »  Š.î á
D Digital Elevation Models

12
D 12.1 Infrastructure applications
Height information, of good quality, geographical distribution and spatial
resolution, is a resource that is essential for the development of a country,
more precisely, the construction of its infrastructure. The reason for
this is that fluids, e.g., water, flow under the influence of gravity, and
the direction in which they flow, and the energy that is released when
they flow, depend on the height differences in the terrain. This applies
to drinking or irrigation water, naturally flowing water from rainfall, in
rivers and in the soil, and to sewage. It even applies to traffic, requiring
roads and railroads to be built with only moderate changes in height
along their trajectories.
There are a number of techniques for establishing a height system, or
vertical datum, in a country. More traditional techniques include precise
levelling, which however is laborious. More rapid, modern techniques
use satellite positioning; however, in order to produce the proper kind
of heights related to the gravity field, i.e., orthometric or normal heights
“above sea level”, one needs a physical model of the figure of sea level, i.e.,
a geoid model. We shall discuss this problem later on.
The detailed description or modelling of heights in a smaller area often
takes the form of digital elevation models, or DEMs. These DEMs are
needed in connection with large infrastructure projects: roads, railroads,
bridges, reservoir dams, irrigation works, mobile telephony link towers,
etc. etc.
DEMs are typically raster files describing a larger or smaller area. An
alternative format is triangulation, e.g., Delaunay1 triangulation: points
are given connected by edges that form triangles approximating the shape

1 Boris Delaunay (1890 – 1980) was a Russian mathematician, mostly remembered for

– 129 –
130 D IGITAL E LEVATION M ODELS

D F IGURE 12.1. Digital elevation models as rasters (left) or triangulation models.

of the terrain. This is often called a TIN (triangulated irregular network)

model. What is nice about triangulation is that it can describe optimally
terrains containing both smooth and rugged areas, as the inter-point
spacing can be varied from place to place.

D 12.2 Presentation of map materials

DEMs can also be used to enhance the presentation of map materials.
An obvious way is to add height contours to a map: this allows the user
to read off the height of any point, as well as judge the convenience of
routes, etc. Also, in physical geography maps, it is common to add height
colours to maps between the contours, making the heights of the terrain
more readily visible to the untrained eye.
A third technique often used in visualization is relief shading, using
an artificial light source, created by software, that shines on the terrain

h( x, y)
s
s +

+
−
s
+
x
y

D F IGURE 12.2. Height contours (left) and perspective view.

inventing a technique for optimally connecting a network of points on a surface into a

triangulation of the surface. He was also an accomplished mountaineer.

Ó »  Š.î á
 Extraction from imagery
131

F IGURE 12.3. Example of a shaded relief representation of a digital elevation

model. The light source is here placed in the North-West at a 60◦
elevation angle, as is conventional in cartography. © Eric Gaba @
Wikimedia Commons, https://commons.wikimedia.org/wiki/File:
D Tutorial_optimization_shaded_relief_map_05.jpg, detail.

from some oblique direction, making the relief visible2 . The merit of this
approach is, that is visualizes not just the heights but also the slopes of
the terrain.
Of course the most obvious technique for representing terrain heights
is as a perspective image, showing it as a bird’s eye view, again from some
oblique direction.
Both the shading technique and the perspective technique exploit deep
properties of the human vision system.

D 12.3 Extraction from imagery

Extracting a digital elevation model from a set of exteriorly oriented aerial
images can be done manually or more or less automatically. The manual
process is laborious and no longer used. It was based on following, in a
stereo restitution instrument, the height contours of the landscape using
the floating mark at a constant height. The contours were registered
digitally, e.g., by printing out a paper tape or punched cards. Together,
the contours formed an elevation model of the terrain.
Today’s DEMs are almost always in raster format, like the aerial images

2 See, e.g., http://www.reliefshading.com/

Ó »  Š.î á
132 D IGITAL E LEVATION M ODELS

from which they are derived. Constructing the raster requires going
through a pair of images and finding corresponding locations, so-called
homologuous points. This is a challenging task to do automatically or
semi-automatically. One proceeds in the following steps:
1. One starts by providing a number of seed points, points for which
three-dimensional terrain co-ordinates X , Y , Z are known. These
could be, e.g., the ground control points of the aerial photography
block.
2. The software then undertakes a search process outward from the
seed points, seeking for neighbouring little patches of terrain visible
in both images. Correspondence between patches can be established,
e.g., by correlating the patches. Then the terrain height is found
for which this correlation is maximal. The new point added to the
data base is then the centre of this patch. This can be done with
good computational efficiency because the height is expected not
to change much from that of the previous point. Nevertheless the
number of numerical operations involved is huge, due to the large
number of pixels in each image.
3. In order to cut down on the calculation effort, one may restrict
the calculation to a sparser grid if it is known that the terrain is
reasonably smooth.
4. After a grid of points with known heights has been formed through-
out the area, an interpolation is undertaken creating a smooth
model surface with a height value for every pixel in the DEM.

D 12.4 Automatic feature matching

In the literature are a number of approaches for establishing the corre-
spondence of points or patches in overlapping image pairs, recently an
object of scientific interest due to the development of machine vision. The
following techniques can be distinguished:
1. SIFT, or Scale Invariant Feature Transform
2. Methods based on correlating patches
3. Methods based on topological relations.
Of these, the method based on correlating patches, which was discussed
in section 11.2, can be refined into least-squares matching, as described
in section 11.3. Also SIFT is discussed there, section 11.4.

Ó »  Š.î á
 Digital elevation models (DEM) with e-foto
133

Interpolate
Open seed editor
DEM to grid
Open stereo plotter
Load extracted DEM Save DEM
Automatic DEM Load interpolated DEM
extraction
Abort current operation
Save DEM Exit / done

D F IGURE 12.4. Main menu of DEM generation module.

D 12.5 Digital elevation models (DEM) with e-foto

The purpose of the DEM Extraction module is to construct a digital terrain
model (DTM) out of a set of aerial photographs that have been exteriorly
oriented — either individually in the Exterior Orientation module, or to-
gether in Phototrangulation. The DEM is obtained in the terrain co-ordinate
system X , Y , Z in which the exterior orientation was obtained, i.e., the
reference system of the given ground control points. It is thus a require-
ment that the images are interiorly and exteriorly oriented, e.g., by the
Phototriangulation module.
The way the DEM is being constructed is by starting from a set of
given seed points, see Silveira (2011). The technique used is differential
in nature.
The DEM generation consists of two parts:
1. Obtaining a set of homologuous points for each stereo pair in the
project, and their terrain co-ordinates X , Y , Z — “DEM Extraction”
2. Interpolate these points to a regular grid — “DEM Interpolation”.

Ó »  Š.î á
134 D IGITAL E LEVATION M ODELS

F IGURE 12.5. Quasi-point at a non-level road intersection. This is not a unique

D three-dimensional point.

D 12.6 Doing the DEM generation

D 12.6.1 Seed points

The “seed editor” that comes up when clicking on the button is shown
in Figure 12.6. E-foto generates a small number of “seed points” auto-
matically, actually the ground control points. The user interface should
be rather familiar by now. The idea is to place a number of seeds, ho-
mologuous points in each stereo pair, from which the software moves
gradually outward, in a search process of areal expansion finding further
homologuous point pairs with their three-dimensional co-ordinates to be
used for building the DEM. If the seed points generated automatically
are not sufficient, one can manually add some more.
Good seed points meet the following requirements:
1. they are spread evenly throughout the area of overlap of the image
pair. Place new seed points especially there, where there are large
empty areas between the already existing seed points
2. they are sharply defined in the terrain. E.g., the corners of buildings
or other structures, but at ground level, or corners of white road
markings like pedestrian crossings.
3. Make sure the surrounding terrain is smooth, so the areal expansion
process will work. Don’t use points on the roofs of buildings3 ; they
won’t allow areal expansion into the surrounding terrain.
4. Don’t measure “quasi-points”, like in figure 12.5, which are not
really unique three-dimensional points, as may occur at non-level
road crossings. Also, don’t measure anything that moves between

3 One can however load the information from feature extraction (the Stereoplotter module)
into DEM Extraction.

Ó »  Š.î á
 Doing the DEM generation
135

Toolbar
Save Exit
Seed table Load

New seed Remove Cancel

seed

F IGURE 12.6. Seed editor main window. The seed points are in yellow, and are
listed in the table bottom left. Seed points can be added using
D the buttons bottom right.

exposures, like cars, animals, trees in the wind, or even the shadows
of lamp-posts!
An important feature not shown in the figure is the tick box in the mid-
dle marked “Show matching pairs”. If you tick this box, the result of the
previous DEM extraction / areal expansion run — i.e., the homologuous
points already found — will be displayed as red crosses; see figure 12.7.
This allows you to judge in which areas further seed points are needed to
make a more complete coverage possible.

D 12.6.2 Areal expansion (“DEM Extraction”)

Areal expansion of the set of homologuous points proceeds from existing

points, by testing the four neighbouring points in the North, South, East
and West directions at a distance given by the “Region Growing step”, see
figure 12.8. It compares the pattern found in a patch of the image (a
template) around the point with the corresponding pattern in a patch

Ó »  Š.î á
136 D IGITAL E LEVATION M ODELS

F IGURE 12.7. Seed editor with “Show matching pairs” activated, so the red areas
show the coverage already obtained using Areal Expansion from
the given set of seed points. Several seed points have been added
D to improve coverage.

of the other image, a process called cross-correlation, eq. ??. A variant

of the technique, eq. 11.2, allows for arbitrary geometric deformations
between the two patches in the form of an affine transformation: this
method is known as least-squares matching. In both techniques, the free
parameters are estimated by the least-squares technique, minimizing
the sum of squares of differences and at the same time maximizing the
Pearson4 correlation between the patches.
This process of gradual areal expansion can of course only proceed in
areas that are visible from both images of a pair. It is clear that in our
test area in the city of Rio de Janeiro (and in urban areas in general)
this is a problem: high buildings frequently occlude the ground in one of
the images, and these ground points cannot be reached by the algorithm.
Adding more seeds helps a little, but is not a complete solution to this.
Also discontinuous jumps in terrain height will block progression of the
areal expansion.

4 Karl Pearson (1857 – 1936) was an influential English statistician.

Ó »  Š.î á
 Doing the DEM generation
137

Growth step
Seed point

F IGURE 12.8. The areal expansion process. In this image, the growing step is
two pixels, so a sparse grid of determined heights is constructed
that still needs to be interpolated, see section 12.7. The pink
D background shows obtained coverage, like the red in figure 12.7.

The further menus of DEM Extraction are depicted below. Each contains
default values for the parameters with which the algorithm is known to
function generally well based on experience. Of course more adventurous
users can experiment with these values.

D 12.6.3 Extraction sub-menus

The Automatic Extraction sub-menu is shown in figure 12.9.

In this, Region Growing step is the number of pixels at a time the areal
expansion proceeds in the NSEW directions. Thus, a sparse but regular
grid of points is constructed that will need interpolation — the second
step, discussed below — for full coverage by the DEM generated. This
value may be increased to as much as 10 in smooth terrain, to speed up
the process.
Also for smooth terrain, one may downsample the images to speed
up the process. However, this will also reduce the resolution of the
orthophoto image produced, so should be used with care.

Ó »  Š.î á
138 D IGITAL E LEVATION M ODELS

D F IGURE 12.9. The Automatic Extraction sub-menu.

There are two different approaches to automatic extraction offered: least-

squares matching, and the simpler cross-correlation. Each has its own
sub-menu, Least-Squares Matching in figure 12.10 and Cross-Correlation in
figure 12.11.
The result obtained, after a prolonged run, looks like Figure 12.12, a
graph showing in yellow the homologuous points that were found by the
process, and in black, locations not reached.
The statistics for matching accuracy (Pearson correlation) are given as a
histogram, figure 12.13.
As you see, coverage is far from complete. Some of the large, black
areas may be filled in by adding “seed points” in the seed point editor;
you should place these seed points in the middle of fairly level areas.
However, some areas, like high-rise, densely built urban areas, are just
impossible to process. Here, the option of importing digitized features
from the feature editor is useful.

D 12.7 Doing DEM Interpolation

DEM Extraction will produce a point set throughout the model that will
still be sparse: if you magnify out a small area, you will see something
looking like a sieve, with lots of little black holes. Therefore after DEM
Extraction, one should run DEM Interpolation, which fills in the missing
pixels in-between the pixels for which extraction has produced heights.
The result will be a grid file of point heights very similar to the image
files used. This DEM file can be saved and used in many applications,

Ó »  Š.î á
 Doing DEM Interpolation
139

F IGURE 12.10. Least-Squares Matching sub-menu. The expert-setting items

“Convergence SHIFT, SCALE, SHEAR” are related to the affine
transformation introduced between the patches as discussed in
D section 12.4.

including the production of orthophoto maps from the images.

The interpolation result may look like figure 12.14.

F IGURE 12.11. Cross-Correlation sub-menu. As in the Least-Squares Matching sub-

menu, the user may enter the minimum Pearson correlation
value required for a match, the minimum amount of greytone
variability (“std”, standard deviation) required to make the patch
or “template” useable for correlation, and the way its size may
D be grown if there is too little such variability.

Ó »  Š.î á
140 D IGITAL E LEVATION M ODELS

F IGURE 12.12. Result of Areal Expansion (Region Growing), for the Rio the
D Janeiro area.

D 12.8 Manual DEM entry

It is clear that for urban areas — like all three test image sets — the
automatic generation of a digital elevation model will not work very well.
It is best suited for areas without buildings and smooth terrain, like open

D F IGURE 12.13. Matching accuracy histogram.

Ó »  Š.î á
 Cleaning up the digital terrain model
141

D F IGURE 12.14. DEM interpolation result by e-foto. Coverage is still incomplete.

fields, savannah or desert. Forest might work, but the outcome could be
that the DEM contains the forest canopy rather than the ground.
One available option is to import a DEM that has been produced by
other means, e.g., airborne laser scanning or synthetic aperture radar
(SAR) from aircraft or satellite.
For urban areas, the obvious alternative is to import a feature file that
has been produced by the Stereoplotter module, containing the three-di-
mensional descriptors of point, line and polygon features like buildings,
roads, parks etc. The DEM Extraction module can then interpolate an
elevation model from this, good enough for use in orthophoto production.
See figure 12.15.

D 12.9 Cleaning up the digital terrain model

D 12.9.1 Why cleaning up?

For an urban area, a digital elevation model (DEM) will contain many
buildings, which is not what we usually call “terrain”. So, if we are inter-
ested in generating an DTM, a “digital terrain model”, that represents

Ó »  Š.î á
142 D IGITAL E LEVATION M ODELS

D F IGURE 12.15. Manually generated DEM based on urban features.

terrain features as commonly understood — e.g., for generating terrain

height contours to be included on a topographic map — then we must
remove these buildings by editing the DTM.
However, if the only purpose of the DTM is to be used for ortho-
rectification of the imagery, than this is unnecessary, and in fact not
even desirable. If the DTM contains the buildings (i.e., it is a DEM), then
in ortho-rectification, the roof edge of every building will be shifted to the
correct X , Y location, which is precisely what we want. Every building
will be represented by its roof outline on the orthophoto map produced.
The ERDAS Imagine software contains a facility for the manual removal
of buildings, basically by drawing a polygon around the building at terrain
level, and then filling the polygon with values on that level.
e-foto does not offer any such facility, but, on the other hand, the e-foto
DTM generation leaves the building areas uncovered anyway. These
uncovered areas may now be interpolated by a repeat application of the
DTM interpolator, using a sufficiently large interpolation radius. For
this, you have to re-load the already interpolated DTM file again into the
interpolator.

Ó »  Š.î á
 Saving the digital surface model
143

D F IGURE 12.16. Saving the digital surface model file.

D 12.10 Saving the digital surface model

After completing the interpolation, you will want to save the digital
surface model obtained into a file. This file will have the extension .dsm;
you should write the full filename including extension before saving
it, see figure 12.16. This file will be needed in the following operation,
ortho-rectification.

D 12.11 Other DEM construction technologies

D 12.11.1 Radar

Interferometric Synthetic-Aperture Radar (InSAR) from satellites is a

powerful and precise technique to build DEMs. It is based on sending a
radar pulse to Earth from a satellite on two different overpasses. The two
different locations of the satellite antenna in space allow the construction,
by interferometry, of a model of the terrain heights underneath. Alterna-
tively and for higher resolution on the ground, two separate antennas on
an aircraft are used in a single overflight.

D 12.11.2 Airborne laser scanning (LiDAR)

In this technique the terrain under the aircraft is scanned, typically

whiskbroom wise, by a range-measuring laser beam. For airborne laser

Ó »  Š.î á
144 D IGITAL E LEVATION M ODELS

scanning, the preferred technique is a pulsed laser beam. The angle

under which the beam goes out is steered by a rapidly rotating mirror
and is a linear function of time for every sweep.
This technique allows, among many other things, the construction of
geocentric terrain height models of high precision (several cm, depending
on the nature of the terrain) and resolution (sub-m2 ). Key to the success
of airborne laser scanning is the positioning by GNSS of the scanning
aircraft placing the height models obtained in a geocentric reference
frame. Compared to using photogrammetry the workflow is largely auto-
matic, explaining in part its current popularity. Precise calibration, and
especially connecting the laser scanner’s origin with the electromagnetic
centre of the GNSS antenna, is however demanding.
Modern instruments allow the recording and use of multiple reflections,
e.g., from the forest canopy, mid-level foliage, understorey, and forest
floor. Then one can choose to construct either a digital elevation model
(including the canopy) or a digital terrain model (only including the
ground). In urban areas, for constructing a terrain elevation model,
buildings will have to be removed, either manually or automatically.

Ó »  Š.î á
D Orthophoto mapping

13
D 13.1 Rationale
Before a set of aerial photographs, obtained as part of an aerial mapping
project, can be used for actually drawing a map, it is necessary to extract
the information in them in the form of three-dimensional co-ordinates.
This is typically done by going through the sequence of interior and
exterior orientation and formation of a stereo model, using given points
in the terrain of which the true geocentric co-ordinates are known. After
this, three-dimensional co-ordinates of arbitrary terrain points can be
recovered from the photographs.
Digital elevation models or DEMs are a prerequisite for orthophoto
mapping.
If we know the height of a point in the photograph above the refer-
ence surface (e.g., sea level), we may use this information to derive the
true horizontal position ( X , Y ) from the position in the photograph, and
from our knowledge of the absolute position ( X 0 , Y0 , Z0 ) and absolute
orientation of the camera optical centre. See figure 13.1.
Another way to look at this is that we want to map the landscape as it
would look from infinitely far away, looking straight down from the zenith.
This is known in cartography as an orthographic projection (greek: ortho
= straight).
The procedure for doing so may be automated, allowing the automatic
drawing, by optical projection, of the aerial photograph onto a light
sensitive emulsion with every pixel being in the correct ( X , Y ) position.
This method is known as orthophoto mapping.
According to the World Bank, orthophoto mapping is the preferred
method for cadastal mappings (Timo Linkola, personal communication;
see also, e.g., Hailu and Harris (2014)).

– 145 –
146 O RTHOPHOTO MAPPING

Correction in image

Camera

Optical
centre

Terrain point Terrain

Ortho
projection
height Z
Projected point X , Y Reference surface (sea level)

F IGURE 13.1. Orthophoto mapping. Notice how buildings are “falling over”
outside the image centre (photo detail from Rio de Janeiro). The
correction in the image, computed using a digital elevation model
D of the terrain, is applied in an orthophoto mapping instrument.

D 13.2 Advantages and drawbacks

The great advantage of the method of orthophoto mapping is, that it is
rapid, requiring little additional manual work after the acquisition of the
imagery and its stereo restitution. However, a digital elevation or surface
model (digital terrain model) or DEM, Z ( X , Y ) of the area to be mapped
must be available. Traditionally such a model was constructed from the
stereo restitution itself, which is able to provide three-dimensional co-
ordinates for every terrain point captured on two images. Nowadays often
pre-existing DEMs are used, typically acquired using the technique of
airborne laser scanning (also called LiDAR, Light Detection And Ranging,
by analogy with radar).
A disadvantage of orthophoto mapping is, that the plotted information
is purely visual, coming directly from the aerial photograph without any

Ó »  Š.î á
 Ortho-rectification with e-foto
147

D F IGURE 13.2. Ortho-rectification in e-foto.

interpretation. The user is expected to do the interpretation. In practice,

interpreted information is added to the map from independent sources:
place names, the road network, elevation contours, etc. Such a map may
be rapidly produced yet be almost as useable as a traditionally produced
map.
Orthophoto maps are extensively used in order to provide a first, rough
cut at mapping an area on which little information pre-exists, and where
time is of the essence. An example of this is a military situation where
no good maps of enemy territory are available. Another use case is large,
thinly populated areas which have not been previously mapped at large
scale, and where the collection of detailed information by traditional
means would be prohibitively expensive against its expected utility.

D 13.3 Ortho-rectification with e-foto

In this step, we eliminate from the photographic images the effect of
“falling over” away from the image’s centre due to the height of objects in
the terrain. I.e., we re-draw the images with the content of every pixel
shifted, in space, vertically downward to sea level, in the image, toward
the nadir, the point straight under the aircraft, by an amount depending

Ó »  Š.î á
148 O RTHOPHOTO MAPPING

on the pixel’s height.

E-foto will do the ortho-rectification computations for every image in turn,
which will take a little time. Then, it presents its results in the image
viewer.

Ó »  Š.î á
D Applications

14
D 14.1 Cadastral system, zoning, and aerial mapping
In a modern, high-investment economy, real estate property and real
rights or encumbrances on it — such as mortgage — are essential finan-
cial instruments. They need to be reliably recorded, including property
boundaries. For this, there is the cadastral system. The precision with
which boundaries need to be recorded, depends on the value of the land:
in areas of low land value, such as deserts, the precision requirement is
low, whereas in areas of high value, such as urban areas and city centres,
the precision requirement will be high. This is referred to as various
measurement classes.
As one example we give here the classification presented in the Finnish
regulatory document “Drafting a Local Zoning Map”, also known as
JHS 185. See table 14.1.
It is possible to use aerial mapping techniques for taking inventory of
real-estate boundaries, if these boundaries are already marked in the
terrain by physical features. Such features may be hedges, stone fences,
streams, or whatever physical distinguishing marks that are somewhat
permanent and visible from the air. This is the so-called “General Bound-
aries” system which is in use in many Anglo-Saxon countries. It works
well provided the precision required is not very great, and if the system
is supported by law and agreement by land owners to place or grow these
boundary markers between their properties and accept their legitimacy.
Advantages are speed and economy.
Where the land is more valuable and mapping precision requirements
are higher, greater precision and definiteness of boundary markers are
required, and proper surveys have to be carried out. This can be done with
traditional geodetic survey techniques, but also from the air. However,

– 149 –
150 A PPLICATIONS

D T ABLE 14.1. Measurement classes, from the Finnish publication JUHTA

(JHS 185). Note that lake or sea shore (measurement class 3)
is especially valued in Finland for its recreational use.

Survey areas are divided into three measurement classes. The measurement class defines the
accuracy of measurement and graphical rendering.
Measurement class 1: Urban areas where land is very valuable and where there is a valid
local zoning plan with a binding parcel division, or a construction ban awaiting the
drafting of such a plan.
In surveys intended to be incorporated into the municipal GIS for use in technical
planning requiring great accuracy, a higher accuracy level may be used (measurement
class 1e).
Map scale 1 : 500 or 1 : 1000. Parcel boundary co-ordinate accuracy (mean error or
standard deviation) ±0.085 m.
Measurement class 2: Urban areas for which the local zoning plan to be drafted does not
require binding parcel division.
Map scale 1 : 1000 or 1 : 2000. Parcel boundary co-ordinate accuracy ±0.14 m.
Measurement class 3: Areas zoned as lake or sea shore, lake or sea shore areas, and other
areas where the land is clearly more valuable than agricultural or forest land, e.g.,
so-called dispersed settlements (https://en.wikipedia.org/wiki/Dispersed_settlement).
Map scale is usually 1 : 2000. Parcel boundary co-ordinate accuracy ±0.21 m.
A scale of 1 : 4000 or 1 : 5000 may be accepted if a zoning plan can be drafted with such
a map without essentially compromising the requirements to be placed on it.
Measurement class 4: All other (unzoned) areas.

A digital map has no scale. The accuracy of the collected data corresponds to that on a map of
a certain scale. One should not print out graphical products from a digital data base at a scale
greater than the accuracy of the data allows.
In choosing the measurement method and the scale of the map, one must take into account
surveys carried out earlier in the area or its surroundings, and the extent and character of the
area.

the work of marking (signalizing) the boundaries so they are visible from
the air can be substantial. It may be speedier and cheaper to directly
measure the boundaries using, e.g., an automated tacheometer and prism
staff, or GPS in real-time kinematic (RTK) mode. The latter requires a
second GPS receiver or base station on a known location nearby, adding
to cost. A careful comparison of logistics and costs, both of hardware and
of labour, should be made before the choice of method.

Ó »  Š.î á
 Land-use inventory and aerial mapping
151
D 14.2 Land-use inventory and aerial mapping
Here the techniques discussed in the previous chapter find application.
◦ Mapping current land use patterns
◦ Assessing demand on irrigation water, transport capacity, electric
grid power, . . .
◦ Agricultural production estimates for domestic use and export,
preparation for shortages due to drought
◦ Urbanization, also uncontrolled, and related infrastructure de-
mand.

D 14.3 Maps in support of land-use planning

◦ Large infrastructures; roads, railroads, bridges, dams, canals, aque-
ducts, airports
◦ Electricity, high-voltage lines, consumption / load planning. Hydro,
solar
◦ Soil quality and use.

D 14.4 Maps in natural resource management

◦ Geophysical surveys: gravity, magnetism, terrain, ores and mineral
resources, deep aquifers, . . .
◦ Vegetation and forest inventory
◦ Natural habitat protection
◦ etc.

D 14.5 Water resources, hydro power, height systems

In mountainous countries the building of reservoir dams both for regulat-
ing river water and its availability for irrigation, and for the generation of
electrical power, is attractive on a large scale. However, the engineering
task requires a proper mapping of the area affected especially regarding
heights. This concerns not only the area of the dam itself, but especially
the area behind the dam where water will accumulate over a long area of
river valley, and the area downstream serviced by the irrigation water.
Deficiencies in this knowledge may lead to unpleasant surprises.
There needs to be a good vertical reference system. This can be estab-
lished by precise levelling, which is however laborious and challenging

Ó »  Š.î á
152 A PPLICATIONS

especially in mountainous terrain. Satellite positioning (GNSS) is an

alternative method for establishing a vertical reference, but requires
the availability of a geoid model. Either method will establish a sparse
network of points or lines of known height throughout the area. Areally
dense knowledge of the terrain height in the area, in the form of digital
elevation models, can then be acquired using aerial survey. Such knowl-
edge is needed in order to predict what areas will be covered by water
when the reservoir fills up, in support of evacuation, resettlement, and
compensation. It also allows estimation of the total water volume, energy
stored, and areas reachable by irrigation water.
More precise geodetic survey techniques are required to monitor the
structural integrity of the dam, and its deformations and those of the
surrounding Earth’s crust. Filling up reservoirs typically causes small
deformations and even microseismicity, which need to be monitored for
safety reasons. Aerial survey techniques may be useful for this, though
their precision is limited. Airborne laser scanning may be an attractive
option. In all cases, a precise geodetic base network tied to the non-
deforming surroundings is a requirement.
As a matter of general interest, the amount of water stored behind
reservoir dams worldwide amounts to some 30 millimetres sea-level
equivalent (Chao et al., 2008)!

Ó »  Š.î á
D Datums in geodesy and
photogrammetry

D 15.1 Geodetic reference systems

15
In order to describe, in geodesy and photogrammetry, the locations of
points in numbers, we need co-ordinate reference systems. A reference
system describes the nature of the co-ordinates that give the location of a
point in the system, consisting of
1. the nature of the co-ordinates used: are they rectangular
co-ordinates, spherical co-ordinates, ellipsoidal co-ordinates, etc.
2. the way the origin and the orientation of the co-ordinate axes are
chosen: is the origin in, or near the geocentre (the centre of mass
of the Earth), or is it on or near the Earth’s surface? Is one of
the co-ordinate system’s axes oriented along the Earth’s rotation
axis, or along the local plumbline? And does another axis point to
a location on the Earth’s surface, or to a location on the celestial
sphere? Or to the local North, or along some instrumentally defined
zero direction?
Every answer to these questions defines a co-ordinate reference system.
Note that we haven’t yet talked about geodetic networks and observations
at all, which we must do if we are to achieve our goal of quantifying point
locations.

D 15.2 Geodetic reference frames and datums

Geodetic measurements, in the form of geodetic measurement networks,
allow us to to tie a co-ordinate reference system to the physical Earth,
producing a co-ordinate reference frame. This is referred to as realizing
the reference system.
However, when actually trying to compute a geodetic network, it turns

– 153 –
154 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

out that there usually are things that need to be fixed in order to arrive
at a unique solution.
E.g., when adjusting — i.e., computing — a levelling network, one must
conventionally fix the height of one point in order to find unique height
values for all the points in the network. This is an example of a datum
defect: a number that must be provided before a unique solution becomes
possible. Typically, the point will be at sea level, close to a tide gauge
monitoring local sea-level variations. The height value chosen for the
point will make mean sea level at this tide gauge equal to zero. This
conventional choice defines a geodetic datum.
For two- or three-dimensional co-ordinate reference systems, the datum
fixing issue becomes more complicated, as we shall see.

D 15.3 Geocentricity
In geodesy and photogrammetry we express co-ordinates, of ground con-
trol points, of the photographing aircraft, or of the terrain being mapped,
in some co-ordinate reference frame. This frame will be uniquely tied to
the Earth by measurements, and by fixing the geodetic datum.
Traditionally, datums were always local, or at best national or regional.
This was because measurement technologies available were terrestrial
in nature. Using theodolites and spirit levels, one could only measure
within small areas on the Earth’s surface. Thus, these techniques could
not provide an overview over large areas, or of the Earth as a whole.
This changed with the introduction of satellite positioning technologies,
i.e., GNSS, Global Navigation Satellite Systems. The first modern such
system was the American GPS, the Global Positioning System. These
technologies allow the creation of geocentric datums, tied to the Earth’s
centre of mass as their origin, and to the Earth’s rotation axis for their
orientation.
Such global datums are based on so-called terrestial reference systems
(TRS), e.g. the International Terrestial Reference System (ITRS) of the
International Association of Geodesy (IAG). This system has been the
basis for a series of global datums or realizations called International
Terrestial Reference Frames (ITRF).

Ó »  Š.î á
 Connecting local and global reference frames
155
D 15.4 Connecting local and global reference frames
The problem arises how to connect these global, geocentric reference
frames or datums — which are typically highly accurate — with the
traditional, non-geocentric, often national, datums. This connection must
be made in situations where, e.g., the imaging aircraft has been positioned
in flight using GNSS technology, but the co-ordinates of ground control
points visible in the imagery are provided in a traditional datum. In
the days before satellite positioning you wouldn’t have to worry about
this. Nowadays, however, GNSS on board the aircraft produces geocentric
( )
locations X 0 , Y 0 , Z 0 of the aircraft for the moment each image was
exposed. These locations must be transformed before use to map positions
X , Y and heights above sea level Z in the same reference frame as used for
the ground control points. And this is also the reference frame — for both
map co-ordinates and heights above sea level — that the photogrammetric
mapping mission will use for its end product.
There are two issues here:
1. in which co-ordinate reference frame the ground control is given,
and
2. in which co-ordinate reference frame do we wish the final result of
the mapping mission to be presented.
Relatively simple techniques based on local rectangular co-ordinates are
still commonly used when all point co-ordinates are given in the same
datum. Typically this will be a local and non-geocentric datum. Then,
many approximations are valid due to the limited size of the study area,
without loss of precision.

D 15.5 Topocentric co-ordinates

D 15.5.1 Definition

Topocentric1 co-ordinates are a special case of local co-ordinates, where

the origin of the co-ordinate frame is some given point on the Earth’s sur-
face, e.g., the location of a geodetic instrument or measurement station.
Topocentric co-ordinates are sometimes called “local astronomical”.
There are as many local systems as there are points of origin, or in-
strument locations, that can be chosen on the Earth’s surface: in fact,
infinitely many.

1 From the Latin topos, meaning place.

Ó »  Š.î á
156 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

z P

x ζ . Z
A
y P
C x z

C y

Y
O

F IGURE 15.1. Topocentric or local astronomical co-ordinates ( x, y, z), and how

D they connect to geocentric co-ordinates ( X , Y , Z ).

D 15.5.2 Rectangular and spherical

In the rectangular version of topocentric co-ordinates, the system’s axes

are as follows:
1. the z axis points to the local zenith, straight up.
2. The x axis points to the local North.
3. The y axis is perpendicular to both others and points East.
This is sometimes referred to as a NEU (“North, East, Up”) system. In
figure 15.1 the situation of a local topocentric system in the global context
is depicted.
The spherical co-ordinates of the same topocentric system are:
◦ the azimuth, or compass direction angle, A . In geodesy, reckoned
from North ( A = 0◦ ) through East ( A = 90◦ ).
◦ The zenith angle ζ (zeta), reckoned from the zenith downward;
alternatively the elevation angle (eta) η = 100 gon − ζ = 90◦ − ζ = π2 − ζ,
reckoned from the horizon upward.
◦ The distance s.

Ó »  Š.î á
 Topocentric co-ordinates
157
The transformation between a point’s topocentric spherical co-ordinates
and its topocentric rectangular co-ordinates is
⎡ ⎤ ⎡ ⎤
x s cos A sin ζ
⎣ y ⎦ = ⎣ s sin A sin ζ ⎦ .
⎢ ⎥ ⎢ ⎥

z s cos ζ

The inverse transformation is

√
s = x2 + y2 + z2 ,
√
x2 + y2
ζ = arctan z ,
y
A = 2 arctan √ .
x+ x2 + y2

The last formula is known as the half-angle formula and avoids the prob-
lem of finding the correct quadrant2 for the azimuth angle A . The result
is in the interval (−180◦ , 180◦ ], and negative values may be incremented
by 360◦ to make them positive.

D 15.5.3 Transforming from geocentric to topocentric and back

Often we will want to connect directions and orientations in a local

topocentric system and a global geocentric system, e.g., when using GNSS
positioning locally in a differential (relative) mode.
Let ( X , Y , Z ) be a geocentric co-ordinate system and ( x, y, z) a topocen-
tric instrument co-ordinate system. In an instrument co-ordinate system
the x axis points to the zero direction of the instrument instead of North.
As examples, in the case of a theodolite, this would be the zero direction
on the horizontal circle, or in the case of an aerial camera, the left-to-right
direction in the image in normal orientation.
[ ]T
Now we can symbolically write, with the notations X = X Y Z
[ ]T
and x = x y z , the same vector in the geocentric and topocentric
systems, respectively:

x = R 1 R 2 R 3 (X − X C ) ,

where the rotation matrices R 3 , R 2 , R 1 act in succession to transform X

into x. See figure 15.2. XC denotes the co-ordinates of the local origin C ,
expressed in the geocentric system.

2 In computer code, one may use the function atan2 ( y, x) for the same purpose.

Ó »  Š.î á
158 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

Astronomical Φ, Λ
co-ordinates z
Plumb- x
z
line y
x
π
2 −Φ
x y
Geoid −x
Z profile

z′ Λ
Y ′
z
y′
X Λ y′
x′
Z x′

Y
X

F IGURE 15.2. From the geocentric to the topocentric system. The matrix R 1
mentioned in the text is left out here.

The inverse transformation chain of this is

T
X = XC + R 3T R 2 R 1T x,

as can be easily derived by multiplying the first equation from the left by
T
the matrix R 1T = R 1−1 , then by the matrix R 2 , and then by the matrix R 3T ,
and finally by moving XC to the other side.

R3 rotates the co-ordinate frame around the z axis from the

Greenwich meridian to the local meridian of the observation
site, rotation angle Λ (lambda, astronomical longitude):
⎡ ⎤
cos Λ + sin Λ 0
R 3 = ⎣ − sin Λ cos Λ 0 ⎦. (15.1)
⎢ ⎥

0 0 1

Seen from the direction of the z axis we see (figure 15.3), that

x′ = x cos Λ + y sin Λ,
y′ = − x sin Λ + y cos Λ.

(The correct algebraic signs should always be established

by the aid of a sketch! Also, the directional conventions of
different countries may differ.)

Ó »  Š.î á
 Traditional geodetic datums on the reference ellipsoid
159
y
y′

x′
z

λ x

D F IGURE 15.3. Rotation by an angle λ around the z axis.

R2 turns the co-ordinate frame around the y axis, in such a way

that the z axis points to the North celestial pole instead of
to the zenith. The rotation angle needed for this is 90◦ − Φ,
where Φ (phi) denotes astronomical latitude. Furthermore,
the new x axis needs to be mirrored to point to the North
instead of the South. This gives us a “rotation-plus-mirroring
matrix”
⎡ ⎤⎡ ⎤
−1 0 0 sin Φ 0 − cos Φ
R2 = ⎣ 0 1 0 ⎦⎣ 0 1 0 ⎦=
⎢ ⎥⎢ ⎥

0 0 1 + cos Φ 0 sin Φ
⎡ ⎤
− sin Φ 0 cos Φ
= ⎣ 0 1 0 ⎦. (15.2)
⎢ ⎥

+ cos Φ 0 sin Φ

R1 rotates the co-ordinate frame around the new z axis or verti-

cal axis by the amount A 0 , after which the x axis points to
the azimuth of the instrumental zero, e.g., the zero point of
the instrument’s horizontal circle:
⎡ ⎤
cos A 0 + sin A 0 0
R 1 = ⎣ − sin A 0 cos A 0 0 ⎦.
⎢ ⎥

0 0 1

D 15.6 Traditional geodetic datums on the reference ellipsoid

The classical approach to geodetic datums is based on the use, and local
fit, of a reference ellipsoid. Based on local terrestial measurements only,
one cannot make this reference ellipsoid genuinely or precisely geocentric.
However, by using astronomical geodetic measurements — of astronom-
ical latitude and longitude of network points and azimuth of network

Ó »  Š.î á
160 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

Plumbline deflection
Φ, Λ ϕ, λ

Plumbline
Ellipsoidal normal
Geodetic measurement network

Φ, Λ Astronomical co-ordinates Reference ellipsoid

ϕ, λ geodetic co-ordinates Geoid (mean sea level)

Datum point

D F IGURE 15.4. Geodetic and astronomical co-ordinates.

sides — it is possible to give the datum a proper orientation with respect

to the Earth as a whole. E.g., the North direction within the datum will
be the true North astronomically.
The ellipsoid as a reference surface for geodetic measurements will,
thus constructed, not be uniquely defined. Not even the measurements
of astronomical latitude and longitude will help: these determine the
direction of the plumbline or vertical, which, due to Earth curvature,
will change from place to place, already over a distance of a hundred
metres. These changes are partly irregular, due to the gravity field of
the Earth being itself irregular because of the irregular distribution of
masses inside the Earth.
One can however construct a reference ellipsoid which locally fits as
well as possible to the figure of the Earth. This means that in every
location within the area of study, the ellipsoidal normal is as close as
possible to the plumbline. In other words, the plumbline deflections are
as small as possible.
There are two widely used techniques for achieving this:
1. A traditional geodetic datum is defined by, in addition to the shape
parameters (a, f ) of the reference ellipsoid used, also fixing

◦ The components of the plumbline deflections ξP , η P (xi, eta) in

some starting point or datum point P (see figure 15.4) and
◦ The height of the geoid NP above the reference ellipsoid in
datum point P .

Ó »  Š.î á
Traditional geodetic datums on the reference ellipsoid
161
This means that, in total, five (5) parameters must be fixed con-
ventionally to starting values in a datum point in order to obtain
a datum definition. In this way the reference ellipsoid surface on
which the geodetic calculations are done is positioned in space.
An example of such a geodetic datum is ED50 (European Datum
1950), which used the Frauenkirche, a church in Munich, Germany
as its datum point.
2. We minimize the regional plumbline deflections and geoid heights
in a least-squares sense:
n
∑ n
∑
minimize (ξ2i + η2i ) and N i2 ,
i =1 i =1

where ξ i , η i , N i are the plumbline deflections and geoid heights of

all the measurements point (triangulation points) i = 1 . . . n in the
area. With this approach, their mean values over all these points
will be set to zero.
In other words, with this approach there is not a single datum point
but an ensemble of datum points. The common centre of mass of all
these points together serves as the datum point.
In both approaches one may also, instead of using a starting point or
ensemble of starting points in the measurement area, fix the centre of the
reference ellipsoid. Also then, three parameters, the three co-ordinates3
X 0 , Y0 , Z0 of this centre relative to the Earth’s centre of mass, need to be
fixed. We again have a total of five parameters: a, f , X 0 , Y0 , Z0 .
In the example case of the ED50 datum, the values X 0 , Y0 , Z0 are
of order of magnitude 100 m. This is typical for traditional geodetic
datums: one cannot, by traditional measurement techniques, determine
the absolute deflection of the plumbline at the datum point, i.e., its
deviation from the normal on a geocentric elllipsoid, any better than this.
For the sake of argument, let the size of this deviation amount to
1′′ . This will cause at the centre of the Earth — a distance of 6378 km
6378 km
removed from the datum point — an “error” of ≈ 28 m4 . This
(ρ · 3600)
equals the size of the offset of the centre of the reference ellipsoid from

3 Not to be confused with the terrain co-ordinates of the camera optical centre, for which
we use the same notation.
4 Here, ρ is the length of one radian expressed in degrees; i.e., ρ = 57.3, and ρ · 3600 =
206266.

Ó »  Š.î á
162 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

North
a
eric
th Am Atlantic
Nor No meas.)
Eu
ro
pe

North
American + ED
NAD +
datum + Centre of mass
(NAD)

Geoid
European
datum
(ED50, ED87)

D F IGURE 15.5. Regional datums.

the centre of mass of the Earth. And this offset will be different for every
different geodetic datum, see figure 15.5.
Modern datums or co-ordinate reference frames are based on satellite
positioning techniques. This includes the international geodetic commu-
nity’s ITRS (International Terrestrial Reference Frame), like ITRF2008,
or in Europe, EUREF89, and the various WGS84 datums created by the
United States Defense Department’s operators of the Global Positioning
System GPS. For such datums, due to the use of satellites orbiting the
Earth, the co-ordinate origin will automatically be in the Earth’s centre
of mass. Locations are thus always obtained geocentrically. Only the
uncertainties of the measurement process itself, including the satellite
orbit determination, may still cause offsets from geocentricity on the
several centimetre level.

D 15.7 Transforming between geodetic datums

Traditionally, geodetic datums are defined on the reference ellipsoid,
by performing a geodetic network adjustment in which the location of

Ó »  Š.î á
Transforming between geodetic datums
163
the network, and its orientation, are fixed with the aid of additional,
astronomical information. This datum fix is done by
1. fixing, for either a single point, or for an ensemble of points, the
direction of the ellipsoidal normal ϕ, λ and the height h from the
ellipsoid. Or alternatively, by
2. fixing the location of the centre X 0 , Y0 , Z0 of the reference ellipsoid
used, relative to the Earth’s centre of mass.
Here we will show how these two alternative ways of datum fixing are
related to each other.
Rectangular geocentric co-ordinates may be written as follows, in spher-
ical approximation:
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
X cos ϕ cos λ X0
⎣ Y ⎦ = (R + h) ⎣ cos ϕ sin λ ⎦ + ⎣ Y0 ⎦ ,
⎢ ⎥ ⎢ ⎥ ⎢ ⎥

Z sin ϕ Z0

where ϕ, λ, h are latitude, longitude and height above the reference ellip-
[ ]T
soid, and the centre of this sphere is located at X 0 Y0 Z0 .
[ ]T
The precisely geocentric co-ordinates X Y Z of the same point
can now be written out in two different datums (i.e., referred to two
different reference ellipsoids). In other words, the vector expressions
⎡ ⎤ ⎡ ⎤ ⎡ (1) ⎤
X ( ) cos ϕ(1) cos λ(1) X0
(1) ⎢ ⎥ ⎢ (1) ⎥
Y = R + h ⎣ cos ϕ(1) sin λ(1) ⎦ + ⎣ Y0 ⎦
⎢ ⎥
⎣ ⎦
(1)
Z sin ϕ(1) Z0

and ⎡ ⎤ ⎡ ⎤ ⎡ (2) ⎤
X ( ) cos ϕ(2) cos λ(2) X0
( ) ⎢ ⎥ ⎢ (2) ⎥
⎣ Y ⎦ = R + h 2 ⎣ cos ϕ(2) sin λ(2) ⎦ + ⎣ Y0 ⎦
⎢ ⎥
(2)
Z sin ϕ(2) Z0
must be identical. Here, the superscripts (1) ja (2) denote geodetic co-ordi-
nates ϕ, λ, h computed in two [different datums,
] on two different reference
ellipsoids, with their origins X 0 Y0 Z0 in two different geocentric
locations. We assume here that the orientations of the co-ordinate axes
are the same in both datums.

Ó »  Š.î á
164 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

Let us define the following difference quantities:

⎡ ⎤ ⎡ (2) ⎤ ⎡ (1) ⎤
∆X0 X0 X0
⎢ (2) ⎥ ⎢ (1) ⎥
⎣ ∆Y0 ⎦ = ⎣ Y0 ⎦ − ⎣ Y0 ⎦,
⎢ ⎥
(2) (1)
∆ Z0 Z0 Z0
∆ h = h(2) − h(1) ,
∆ϕ = ϕ(2) − ϕ(1) ,
∆λ = λ(2) − λ(1) ,

then we obtain by subtraction (note the delta notation):

⎛ ⎡ ⎤⎞
( ) cos ϕ(2) cos λ(2)
0 = ⎝ R + h(2) ⎣ cos ϕ(2) sin λ(2) ⎦⎠−
⎜ ⎢ ⎥⎟

sin ϕ(2)
⎛ ⎡ ⎤⎞ ⎡ ⎤
( ) cos ϕ(1) cos λ(1) ∆X0
− ⎝ R + h(1) ⎣ cos ϕ(1) sin λ(1) ⎦⎠ + ⎣ ∆Y0 ⎦ =
⎜ ⎢ ⎥⎟ ⎢ ⎥

sin ϕ(1) ∆ Z0
⎧ ⎡ ⎤⎫ ⎡ ⎤
⎪
⎨ cos ϕ cos λ
⎪
⎬ ∆X0
= ∆ (R + h) ⎣ cos ϕ sin λ ⎦ + ⎣ ∆Y0 ⎦ .
⎢ ⎥ ⎢ ⎥

∆ Z0
⎪ ⎪
sin ϕ
⎩ ⎭

Now, by partial derivation (assuming all ∆ quantities to be small, so that

linearization is allowed):
⎛ ⎡ ⎤⎞ ⎡ ⎤
cos ϕ cos λ cos ϕ cos λ
∂ ⎜
⎝(R + h) ⎣ cos ϕ sin λ ⎦⎠ = ⎣ cos ϕ sin λ ⎦ ,
⎢ ⎥⎟ ⎢ ⎥
∂h
sin ϕ sin ϕ

and similarly with respect to ϕ and λ:

 ⎛  ⎡ ⎤⎞ ⎡ ⎤ ⎡ ⎤
cos ϕ cos λ ∆h ∆X0
∂ ∂ ∂
[ ]
0= ⎝(R + h) ⎣ cos ϕ sin λ ⎦⎠ ⎣ ∆ϕ ⎦ + ⎣ ∆Y0 ⎦ =
⎜ ⎢ ⎥⎟ ⎢ ⎥ ⎢ ⎥
∂h ∂ϕ ∂λ
sin ϕ ∆λ ∆ Z0

⎡  ⎤ ⎡ ⎤ ⎡ ⎤
cos ϕ cos λ − (R + h) sin ϕ cos λ − (R + h) cos ϕ sin λ ∆h ∆X0
= ⎣ cos ϕ sin λ − (R + h) sin ϕ sin λ + (R + h) cos ϕ cos λ ⎦ ⎣ ∆ϕ ⎦ + ⎣ ∆Y0 ⎦ .
⎢ ⎥⎢ ⎥ ⎢ ⎥

sin ϕ + (R + h) cos ϕ 0 ∆λ ∆ Z0

Here we have used, as argued above, the identity of the geocentric

[ ]T
co-ordinates X Y Z in the two different datums.

Ó »  Š.î á
 Transforming between geodetic datums
165
Ellipsoidal normal
ϕ1 , λ1 ϕ2 , λ2

1 2
h2 H Plumbline
ξ1 Φ, Λ
h1
ξ2
N1
.
N2
. Geoid

. Ell
ips
o id 1
El
lip
soi
d2
Datum
transformation

F IGURE 15.6. The effect of a datum transformation (reference ellipsoid shift)

on the geodetic latitude and longitude ϕ, λ, the deflections of the
plumbline ξ, η, the geoid undulation N, and the ellipsoidal height
D h of points.

Because h ≪ R :
⎡ ⎤ ⎡ ⎤⎡ ⎤
∆X0 − cos ϕ cos λ +R sin ϕ cos λ +R cos ϕ sin λ ∆h
⎣ ∆Y0 ⎦ ≈ ⎣ − cos ϕ sin λ +R sin ϕ sin λ −R cos ϕ cos λ ⎦ ⎣ ∆ϕ ⎦ =
⎢ ⎥ ⎢ ⎥⎢ ⎥

∆ Z0 − sin ϕ −R cos ϕ 0 ∆λ
⎡ ⎤⎡
∆N
⎤
− cos ϕ cos λ +R sin ϕ cos λ +R cos ϕ sin λ
= ⎣ − cos ϕ sin λ +R sin ϕ sin λ −R cos ϕ cos λ ⎦ ⎣ −∆ξ ⎦ .
⎢ ⎥⎢ ⎥
∆η
− sin ϕ −R cos ϕ 0 − cos ϕ

This equation gives the relationship that exists between the amount of
non-geocentricity of the reference ellipsoid used, and the datum defined
by it in the point (ϕ, λ, h). At the same time it is also the equation
by which the differences ∆ h, ∆ϕ, ∆λ between two geodetic datums in a
starting point may be converted to translation components of the origin
∆ X 0 , ∆Y0 , ∆ Z0 . In other words the equation by which transformation
parameters can be converted from their topocentric to their geocentric
form.
We have alse expressed the topocentric translation vector,
]T ]T
∆η
[ [
∆ h ∆ϕ ∆λ , into the alternative form ∆ N −∆ξ , − cos ϕ
where appear the geoid undulation N and the deflections of the plumbline

Ó »  Š.î á
166 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

ξ, η, all three evaluated in the datum point. See figure 15.6. Note that5

N = h−H
ξ = Φ−ϕ
η = (Λ − λ) cos ϕ

where
N geoid undulation from the reference ellipsoid (“geoid height”),
h height of the point above the ellipsoid (“ellipsoidal height”),
H height of the point above the geoid (“orthometric height”),
ξ, η deflections of the plumbline, i.e., the differences in direction be-
tween the astronomical vertical (“plumbline”) and the normal on
the reference ellipsoid — ξ in the North-South and η in the West-
East direction,
Φ ,Λ astronomical latitude and longitude (direction of the plumbline),
ϕ,λ geodetic latitude and longitude (direction of the ellipsoidal normal).

Because the quantities H, Φ, Λ are physical quantities6 — neither the

direction of the local plumbline relative to the starry sky, nor the height of
the point above the geoid or “sea level”, depend on the reference ellipsoid
chosen — it follows that

∆ N = ∆h
∆ξ = −∆ϕ
∆η = −∆λ cos ϕ

from which the alternative form above follows.

The inverse formula of the above is:
⎡
∆N
⎤ ⎡ ⎤
∆h
⎣ −∆ξ ⎦ = ⎣ ∆ϕ ⎦ =
⎢ ⎥ ⎢ ⎥
∆η
− cos ϕ ∆λ

5 The extra factor cos ϕ in the η equation comes from the meridian convergence. At
higher latitudes a given change in longitude ∆λ corresponds to an ever smaller distance
on the Earth’s surface, and thus to an ever smaller change in the direction of the vertical.
6 Orthometric height H can be determined by spirit levelling from the coast (together
with gravimetric measurements for reduction), and astronomical latitude and longitude
Φ, Λ by astronomical measurements. Neither technique assumes a reference ellipsoid.

Ó »  Š.î á
 Helmert transformations in three dimensions

⎡ ⎤⎡
167
−R cos ϕ cos λ −R cos ϕ sin λ −R sin ϕ
⎤
∆X0
1⎢
⎢ + sin ϕ cos λ + sin ϕ sin λ − cos ϕ ⎥ ⎢
⎥
= ⎦ ⎣ ∆Y0 ⎦ .
⎥
R⎣ sin λ cos λ
+ cos ϕ − cos ϕ 0 ∆ Z0

These equations simplify somewhat, if we write the topocentric transla-

[ ]T
7
tion into the form ∆ N −R ∆ξ −R ∆η , because then the matrices
are clean rotation matrices:
⎡ ⎤ ⎡ ⎤
∆N ∆h
⎣ R ∆ξ ⎦ = ⎣ R ∆ϕ ⎦=
⎢ ⎥ ⎢ ⎥

R ∆η R cos ϕ∆λ
⎡ ⎤⎡ ⎤
− cos ϕ cos λ − cos ϕ sin λ − sin ϕ ∆X0
= ⎣ − sin ϕ cos λ − sin ϕ sin λ + cos ϕ ⎦ ⎣ ∆Y0 ⎦ .
⎢ ⎥⎢ ⎥

− sin λ + cos λ 0 ∆ Z0

D 15.8 Helmert transformations in three dimensions

Above the transformation between two geodetic datums on reference
ellipsoids was described as a translation, a shift, in space only, i.e., a
three-parameter transformation. For traditional geodetic datums de-
fined on non-geocentric reference ellipsoids this is usually a good enough
approximation.
However, satellite positioning technology (GNSS, Global Navigation
Satellite Systems), changed the game: the superior positioning precision
makes also more precise transformation models necessary, and the model
that offers itself is the seven-parameter Helmert transformation in three-
dimensional space.
The Helmert transformation formula in three dimensions is8

X′ = µR (X − ∆X0 ) , (15.3)

where the three-dimensional, geocentric vectors of location are

⎡ ⎤ ⎡ ⎤
X′ X
X′ = ⎣ Y ′ ⎦ , X = ⎣ Y ⎦,
⎢ ⎥ ⎢ ⎥

Z′ Z

7 Also referred to as the “North, East, Up” form — though here, arbitrarily, “Up” comes
first.
8 Note the similarity with the collinearity equations 8.1 for connecting terrain and
camera co-ordinates! There too, we have a shift vector, a rotation matrix, and a scale
factor.

Ó »  Š.î á
168 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

⎡ ⎤ ⎡ ⎤ ⎡ ⎤
∆X0 X 0′ X0
∆X0 = ⎣ ∆Y0 ⎦ = ⎣ Y0′ ⎦ − ⎣ Y0 ⎦
⎢ ⎥ ⎢ ⎥ ⎢ ⎥

∆ Z0 Z0′ Z0
is the shift between the origins of the X and X′ systems, and

R = R 3 (α3 ) R 2 (α2 ) R 1 (α1 )

is a “chaining” of three rotations. µ (mu) is the scale factor. Equation

15.3 is called the three-dimensional Helmert or similarity transformation.
The elements of the R matrix are complicated expressions in the sines
and cosines of the angles α1 , α2 , α3 and we will not derive them here.

D 15.9 Transformation in case of small rotation angles

Often the axes of two co-ordinate systems are almost aligned with each
other. In that case, the angles of rotation are small9 and one may make
the approximations sin α ≈ α and cos α ≈ 1. Then all equations simplify.
If furthermore we assume that the scale factor µ is close to unity, we may
write
µ = 1 + ∆µ.

Furthermore
⎡ ⎤
1 0 0
R 1 (α1 ) ≈ ⎣ 0 1 α1 ⎦ ,
⎢ ⎥

0 −α1 1
⎡ ⎤
1 0 −α2
R 2 (α2 ) ≈ ⎣ 0 1 0 ⎦,
⎢ ⎥

α2 0 1
⎡ ⎤
1 α3 0
R 3 (α3 ) ≈ ⎣ −α3 1 0 ⎦.
⎢ ⎥

0 0 1

9 An example from aerial photogrammetry is the camera tilt angles ω, ϕ, which are
always small, so the camera z, or c, axis will be almost parallel with the local terrain
vertical axis Z , or H .

Ó »  Š.î á
 Case: transformation between ED50 and EUREF89
169
Whan all α i are small, one may also assume that all products α i α j ≈ 0,
and it follows

R = R 3 (α3 ) R 2 (α2 ) R 1 (α1 ) ≈

⎡ ⎤
1 α3 −α2
≈ ⎣ −α3 1 α1 ⎦ = I + ∆R,
⎢ ⎥

α2 −α1 1

where I is the 3 × 3 unit matrix, and

⎡ ⎤
0 α3 −α2
∆R = ⎣ −α3 0 α1 ⎦
⎢ ⎥

α2 −α1 0

is a skew-symmetric (antisymmetric) matrix: ∆R T = −∆R .

From this follows the correction formula for co-ordinates:

X′ − X = (∆µ + ∆R )(X − ∆X0 ) =

⎡ ⎤
∆µ α3 −α2
= ⎣ −α3 ∆µ α1 ⎦ (X − ∆X0 ) , (15.4)
⎢ ⎥

α2 −α1 ∆µ

where the parameters ∆µ, α1 , α2 , α3 , ∆X0 , as well as the left-hand side

X′ − X, are all small — but X is large.
The form 15.4 is that of the general similarity transformation in the
case of two co-ordinate reference system realizations that are close to-
gether, e.g., between two different ITRS realizations. In this case the
rotation angles α i are of order of magnitude fraction of a second of arc,
and the translation or shift vector ∆X0 is under 10 cm.

D 15.10 Case: transformation between ED50 and EUREF89

ED50 (European Datum 1950) is a traditional geodetic datum for Western
Europe, created shortly after the Second World War by combining and re-
adjusting the primary triangulation networks of the Western European
countries. The reference ellipsoid used in the joint adjustment was the
Hayford or International Ellipsoid of 1924.
EUREF89 is a modern, satellite based, geocentric geodetic reference sys-
tem, created in the 1990s by the IAG’s Subcommission for the European
Reference Frame (EUREF) for Europe. The modern geodetic datums of
the European member states are based on it.

Ó »  Š.î á
170 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

D T ABLE 15.1. Helmert transformation parameters EUREF89 → ED50 deter-

mined for the Finnish territory by Ollikainen (1993). The param-
eters differing significantly (> 3σ level) from zero are in red and
underlined.

Parameter Value Uncertainty (s.d.) Unit

∆X 93.477 ±3.345 m
∆Y 103.453 ±5.534 m
∆Z 123.431 ±2.736 m
α1 −0.246 ±0.168 ′′

α2 0.109 ±0.106 ′′

α3 0.068 ±0.112 ′′

∆µ −2.062 ±0.417 ppm

This transformation between them is of type 15.3, into a precisely

geocentric system. Because the Hayford ellipsoid — on which the ED50
datum is defined — is not (precisely) geocentric, it becomes necessary
to translate the origin of the co-ordinate system to the centre of mass
of the Earth, the geocentre. The translation components are of order of
magnitude of a hundred metres, and also the scale change is significant,
cf. Ollikainen (1993) page 15 and table 2 on page 13:
⎡ ⎤(2) ⎡ ⎤ ⎡ ⎤(1) ⎡ ⎤
X 1 α3 −α2 X ∆X
⎣ Y ⎦ = (1 + ∆µ) ⎣ −α3 1 α1 ⎦ · ⎣ Y ⎦ + ⎣ ∆Y ⎦ ,
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥

Z α2 −α1 1 Z ∆Z

where in the case EUREF89 → ED50 the transformation parameters are

according to the least-squares fit by Ollikainen, for the Finnish territory,
given in table 15.110 .
ED50 (European Datum 1950) is a traditional European reference
frame that is the basis of KKJ, the older Finnish national map grid co-
ordinate system, and which dates from before the satellite era. As the
table shows, it is strongly non-geocentric. The rather large precision
numbers given in the table reflect the circumstance that co-ordinates
determined over a large area by traditional geodetic techniques just aren’t
very precise.
For illustration, the parameters that are larger than three times their
own uncertainty (standard deviation) are marked red. These are sta-

10 For other parts of Europe, slightly different values apply, determined by fits to modern
GNSS measurements within those territories.

Ó »  Š.î á
 Transformations between pairs of modern geodetic datums
171
tistically significant at the three-sigma (3σ) level. The rotation angles
α1 , α2 , α3 are not significant at this level, and one could argue that they
could be left out without seriously degrading the solution.

D 15.11 Transformations between pairs of modern geodetic

datums

Between pairs of modern geodetic datums of the satellite era — typically

named ITRF yy, with the acronym ITRF standing for International Ter-
restrial Reference Frame, and yy for the year of realization — there are
general seven-parameter Helmert transformations of the form
⎡ ⎤(2) ⎡ ⎤(1) ⎡ ⎤⎡ ⎤(1) ⎡ ⎤
X X 0 −α3 α2 X ∆X
⎣ Y ⎦ = (1 + ∆µ) ⎣ Y ⎦ +⎣ α3 0 −α1 ⎦ ⎣ Y ⎦ +⎣ ∆Y ⎦ .
⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥

Z Z −α2 α1 0 Z ∆Z

In this equation, the parameters ∆µ, α1 , α2 , α3 , ∆ X , ∆Y , ∆ Z are actually

treated as linear functions of time, i.e.,

P ( t) = P ( t 0 ) + ( t − t 0 ) Ṗ,

where t is the epoch for which the co-ordinates in datum (2) must be
computed, and t 0 is the epoch of tabulation of the parameter on the web
site11 of the IERS (International Earth Rotation and Reference Systems
Service).

D 15.12 Transformations between local satellite based and

global datums

Since the advent of satellite positioning, also local and national datums
are usually created using this technology. The global satellite based
datums, type ITRF yy, are seldom directly suitable for national or regional
use. The user community expects the co-ordinates of points in the local
datum to be fixed, i.e., not change with time. This means that, for the
highest, geodetic precision, the local datum frame must be attached to
the tectonic motion of the local plate, typically a few centimetres per year.
As an example may serve the European situation. Europe is located
on the Eurasian tectonic plate. The co-ordinate reference system used

11 Currently, e.g., http://itrf.ensg.ign.fr/trans_para.php

Ó »  Š.î á
172 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

(by agreement within EUREF, the IAG Subcommission for the Euro-
pean Reference Frame) is called ETRS, or ETRS89. ETRS stands for
European Terrestrial Reference System. National datums for individual
European countries, created by national GNSS measurement campaigns,
are realizations of this system.
However, geodetic satellite positioning will give locations in the same
system as in which the precise orbital elements of the GPS satellites are
given, so-called precise ephemeris, e.g., ITRF2005. Here, ITRF stands
for International Terrestrial Reference Frame, and 2005 is the year of
determination or realization of this reference frame. Then, the following
transformation formula is commonly used (note the slightly different
notation):
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ETRS89
X X T1
⎣ Y ⎦ ( t) = ⎣ Y ⎦ ( t) + ⎣ T 2 ⎦ +
⎢ ⎥ ⎢ ⎥ ⎢ ⎥

Z Z T3
ETRS89 ITRF2005 ITRF2005
⎡ ⎤ETRS89 ⎡ ⎤
0 −Ṙ 3 Ṙ 2 X
+ ⎣ Ṙ 3 0 −Ṙ 1 ⎦ × ( t − 1989.0) × ⎣ Y ⎦ ( t) ,
⎢ ⎥ ⎢ ⎥

−Ṙ 2 Ṙ 1 0 Z
ITRF2005 ITRF2005

where the dot on the R parameters means derivation with respect to

time. Note that this formula is approximate and applies only (to good
approximation) within the European area.
Values for the T and Ṙ parameters are given in Boucher and Altamimi
(2007), tables 3 and 4. They are orders of magitude smaller than in
the earlier described example of a transformation between a modern
(ETRS89) and a traditional (ED50) European datum. Both systems
discussed here, ETRS89 ja ITRF2005, are in fact geocentric on the centi-
metric level of approximation.

D 15.13 Local and global datums; cases

We will discuss the issue using three different cases: the Finnish KKJ sys-
tem, the newer geocentric Finnish system EUREF-FIN, and the Ethiopian
Adindan datum.
In the following, “WGS84” is shorthand for the ITRS realisation
(ITRF yy) in which the satellite ephemeris used in the computation are
given. For the best precision, this too needs to be handled correctly.

Ó »  Š.î á
Local and global datums; cases
173
KKJ: This system is strongly non-geocentric: the offsets of the centre of
the International Ellipsoid of 1924 it uses, relative to the centre
of the ellipsoid of WGS84, are (Ollikainen, 1993) ∆ X = 93 m, ∆Y =
103 m, ∆ Z = 123 m. These offsets are in the direction WGS84 →
KKJ. Additionally there are non-zero rotations and a non-unity
scale factor.
A further complication is that, while KKJ is a map projection sys-
tem (the Gauss-Krüger projection on the International Ellipsoid
of 1924) based upon the European datum ED50, in the map plane
there is a further Helmert transformation applied to achieve ap-
proximate compatibility (on the metre level) with an older system
in use before 1970, when KKJ was officially introduced.
The “raw” transformation between KKJ and WGS84 only achieves
an accuracy on the metre level. Thus, the transformation must be
considered approximate. The National Land Survey has however
created a triangulated affine transformation (between KKJ and
EUREF-FIN, see below) based upon a Delaunay triangulation
of the Finnish territory, which achieves better (several cm level)
precision.
EUREF-FIN: This system is geocentric and based on satellite posi-
tioning. There are no translation parameters (shifts of origin)
between WGS84 and EUREF-FIN, but there is a rotation matrix:
EUREF-FIN is the Finnish national realization of ETRS89, the
European Terrestrial Reference System, which at epoch 1989.0 co-
incided with ITRS (the international geodetic community’s version
of WGS84), but has since rotated away due to the tectonic motion
of the Eurasian plate of the Earth’s crust, which ETRS eliminates.
EUREF-FIN co-ordinates can be transformed to and from “WGS84”
— i.e., to and from any realization of ITRS — with high precision.
The map projection used for mapping the whole country is UTM
(Universal Transverse Mercator) zone 35 (central meridian 27◦ )
on the GRS80 reference ellipsoid.
Adindan: Also this datum is strongly non-geocentric: ∆X =
−162 m, ∆Y = −12 m, ∆ Z = 206 m (in the direction Adindan
→ WGS84) for the Ethiopian territory (Thomas Dubois, personal
comm. April 30, 2013). This transformation is conventionally

Ó »  Š.î á
174 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

defined and may be assumed exact12 . So Ethiopia uses a

non-geocentric GNSS-based national datum, the connection of
which to WGS84 is exactly known13 . The rotation parameters are
set to zero and the scale to 1. The recommended projection for
mapping the whole country is UTM Zone 37 on the Clarke 1880
ellipsoid.

One should be careful using local, non-geocentric datums, especially if

the transformations to and from WGS84 are not precise, as is always
the case for datums based on traditional measurements. In that case,
GNSS positioning of the aircraft is a complication we don’t really need,
as its superior precision is then lost down the processing chain. The
problem in this case is that ground co-ordinates can only be approximately
transformed to geocentric WGS84 co-ordinates, and conversely, aircraft
image exposure co-ordinates ( X 0 , Y0 , Z0 ) from GNSS can also only be
obtained approximately in this local datum. In this case, one should
use these transformed image exposure co-ordinates only as approximate
values and allow the photogrammetric software to improve them by least-
squares adjustment.
In fact, in this case there is no reason to introduce these aircraft image
exposure co-ordinates at all: better to let the software handle the whole
computation in local project co-ordinates.
This will give good results in practice for photogrammetry, but is not
really an option for airborne laser scanning missions (section 16.1.1)
which rely critically on the practice of GNSS positioning the aircraft: in
that case a good solution is to handle the whole computation chain in
WGS84 or a geocentric datum close to it14 . A final transformation then
obtains co-ordinates in the local datum.
It is also not an option if we have precisely geocentrically positioned
ground control, but wish the mission result to be nevertheless in a lo-

12 For the older, traditional Adindan datum, older documents list ∆ X = −166 m, ∆Y =
−15 m, ∆ Z = 204 m, with uncertainties of 5, 5, and 3 m. Rotation parameters and scale
were presumably estimated and deemed insignificant against their uncertainties.
13 . . . at least at the epoch of its determination, not many years ago. Over time, the
movement of the African plate will assert itself, and Ethiopia will have to define a,
presumably AFREF-based, precise national geodetic datum in which this effect has
been eliminated.
14 In practice one chooses the ITRF datum in which also the GNSS satellites’ precise
ephemeris used were provided by the international geodetic community.

Ó »  Š.î á
 The Proj4 map-projection and datum software
175
Clarke 1880
Z

UTM37
(39◦ E)
WGS84 Y
162 m
206 m
Y
X Adindan
12 m
X

GRS80

D F IGURE 15.7. The Ethiopian “Adindan” datum in relation to WGS84.

cal datum. Then we need to use the appropriate transformation and

projection logic, and make sure these are precise.

D 15.14 The Proj4 map-projection and datum software

All the major geospatial software packages contain a great many national
geospatial datums and geoid models, and for those using free software,
there is proj4 (http://trac.osgeo.org/proj/). However, it is important to use
a transformation that is locally sufficiently precise. This may mean the
extra work of deriving a local transformation based on a set of points
known in both local and WGS84 co-ordinate systems.
In the EPSG data base used in proj4, we have, e.g., for the Finnish KKJ:
# KKJ / Finland Uniform Coordinate System
<2393> +proj=tmerc +lat_0=0 +lon_0=27 +k=1
+x_0=3500000 +y_0=0 +ellps=intl
+towgs84=-96.062,-82.428,-121.753,4.801,0.345,-1.376,1.496
+units=m +no_defs <>

Ó »  Š.î á
176 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

F IGURE 15.8. EGM08 geoid for East Africa and the Middle East. For scale, the
geoid depression south of India is approximately −100 m. © U.S.
National Geospatial-Intelligence Agency. http://earth-info.nga.
D mil/GandG/wgs84/gravitymod/egm2008/

Here we see the definition of the transformation from local to WGS84:

a translation vector, a rotation matrix, and a scaling:

′′ ′′ ′′
+towgs84=∆ X ,∆Y ,∆ Z ,α1 ,α2 ,α3 ,∆µppm ,

to be used as (http://proj.maptools.org/gen_parms.html):
⎡ ⎤ ⎡ ⎤
X X
1 + 10−6 · ∆µppm ⎣ Y ⎦
( )⎢
⎣ Y ⎦ = +
⎢ ⎥ ⎥

Z Z
WGS84 local
⎡ ′′ ′′ ⎤⎡ ⎤ ⎡ ⎤
0 −α3 α2 X ∆X
2π ⎢ ′′ ′′
+ α 0 −α1 ⎦ ⎣ Y ⎦ + ⎣ ∆Y ⎦ .
⎥⎢ ⎥ ⎢ ⎥
360 · 60 · 60 ⎣ 3′′ ′′
−α2 α1 0 Z
local
∆Z
′′
Note that ∆µppm is given in ppm (parts per million) and the rotations α i
in seconds of arc, and these need all to be converted.
The projection is defined as a generic transversal Mercator
(+proj=tmerc) with a re-scaling factor of one (+k=1) (i.e., plain

Ó »  Š.î á
 The geoid
177
Gauss-Krüger), a central meridian of 27◦ East (+lon_0=27), a “false
Easting” of 500 km (+x_0=3500000, note that the “3” is the Finnish zone
number which is prepended), and all on the International (+ellps=intl)
or Hayford reference ellipsoid. . .
The equivalent definition for Adindan would be (http://epsg.io/20137)

# Adindan / UTM zone 37N

<20137> +proj=utm +zone=37 +ellps=clrk80
+towgs84=-162,-12,206,0,0,0,0
+units=m +no_defs <>

where now the projection is UTM (no further detailed prescriptions

needed, except +zone=37) and the reference ellipsoid is +ellps=clrk80.
A warning is in place here on the use of proj4: the software is not
intended for transforming geodetic-quality co-ordinates. This shows, e.g.,
in the cavalier use of the term “WGS84” in the code, as well as the lack
of time dependent transformation parameters. Geodetic-quality work
requires that all point co-ordinate data, be they ground control points
or in-flight exposure locations, have been carefully transformed to — or
originally obtained in — exactly the same geodetic datum. If any of that
co-ordinate data has been obtained by GNSS, the datum to be used will
also be a modern, GNSS-based datum.

D 15.15 The geoid

The currently best global geoid, or geopotential, model is EGM08, pub-
lished in 2008 by the U.S. National Geospatial-Intelligence Agency. It
has a resolution of approximately 2.5 minutes of arc, corresponding to
approximately 4 km in position. It is already incorporated in modern
geodetic and photogrammetric software.
Many countries have local geoid models, which are typically more de-
tailed than EGM08, and tied to the national height system. An Ethiopian
precise geoid determination effort — or rather, its modern hi-tech equiva-
lent — is documented in Bedada (2010). It uses the innovative approach,
made possible by information technology, of computing directly the geopo-
tential in three dimensions, by interpolating in a pre-computed 3-D grid,
for any point on or above the terrain where GPS provides the geocentric
( )
co-ordinates X , Y , Z . Instrumental in achieving a precise result was
the use of the Ethiopian Airborne Gravity Survey data (Bedada, 2010), as

Ó »  Š.î á
178 D ATUMS IN GEODESY AND PHOTOGRAMMETRY

well as topographical data from the Shuttle Radar Topography Mission

(http://www2.jpl.nasa.gov/srtm/).
The pre-computed grid can be incorporated into GNSS receivers and
used in real time in the field.

Ó »  Š.î á
D Advanced subjects

16
D 16.1 Airborne laser scanning (ALS)
D 16.1.1 Airborne laser scan terrain height surveys

Airborne laser scanning has in a short while become the method of choice
to build detailed, precise digital elevation models (DEMs) over large areas.
In airborne laser scanning, three-dimensional terrain data is directly
geolocated geocentrically with the aid of a GNSS/IMU assembly, i.e., a
satellite positioning receiver (GNSS) plus an inertial measurement unit,
together tracking geocentric location and attitude of the aircraft in real
time.
The laser scanner typically is of the whiskbroom type, which changes
the nature of the exterior orientation problem: exterior orientation needs
to be performed for every measurement epoch, many times a second. Do-
ing this with ground control points would be very challenging: GNSS/IMU
is inevitable. It is however possible to specify verification points in the
terrain, or even verification surfaces, e.g., Dahlqvist et al. (2011).
Generally, the on-board GNSS system measures the location of the
GNSS antenna — more precisely, the electromagnetic centre of the an-
tenna, which must be determined by antenna calibration. This centre
may not be well defined, especially close to the large metal body of the
aircraft which causes reflections of the GNSS signal (“multipath”). GNSS
antennas are mounted on top of the aircraft fuselage for a good view of
the sky.
The laser scanning device on the other hand is mounted on the bottom
of the aircraft, for a good view of the ground. The inertial measurement
unit, or IMU, which, integrated with GNSS, mainly keeps track of the ro-
tational motions of the aircraft platform, can be mounted anywhere. One
should however try to place all three units, laser scanner, GNSS receiver,

– 179 –
180 A DVANCED SUBJECTS

Z δκ
z
δϕ
IMU
GNSS
δω
δx0
y
Laser scanner Y

x
Optical centre X 0 , Y0 , Z0

X
Terrain co-ordinates X , Y , Z

F IGURE 16.1. Geometry of airborne laser scanning. All three devices are
D mounted close together on the aircraft.

and IMU, as close as possible to each other, and rigidly interconnected.

This is because the airframe is elastic and will undergo deformations
under varying load.

D 16.1.2 System calibration

Now, the following offsets will exist:

1. A three-dimensional shift vector, here given in aircraft body co-ordi-
nates ( x, y, z):
δx0 = xGNSS − xALS

between the GNSS antenna location and the laser scanner’s origin.
2. A rotational bias triad

δω = ωIMU − ωALS ,
δϕ = ϕIMU − ϕALS ,
δκ = κIMU − κALS

between the aircraft body axes orientation as maintained by the

GNSS-IMU system, and the actual orientation of the laser scanning
sensor. These angles are typically small.

Ó »  Š.î á
 Airborne geophysical surveys
181
All these six offsets could in principle be determined by careful measure-
ment, but this has turned out to be difficult1 . In practice, field calibration
is applied, where these quantities are simply added as unknowns to be
determined. This means that, unlike the situation in traditional pho-
togrammetry where we have one set of exterior orientation unknowns

( X 0 ) i , (Y0 ) i , ( Z0 ) i , ω i , ϕ i , κ i

for every image exposure i , here we have one set of six unknowns

δ x0 , δ y0 , δ z0 , δω, δϕ, δκ

for the whole mission.

This remark applies not only to airborne laser scan campaigns, but
just as well to modern aerial photogrammetry projects where the aircraft
platform is located and oriented using a GNSS/IMU combination, which
is quite common nowadays. However, in this case it may be possible to
improve the exterior orientation parameters per image exposure using
good quality ground control points. It is then worthwhile keeping these
individual elements in the computation.

D 16.2 Airborne geophysical surveys

Airborne geophysical surveys are used to study a great many quantities
of interest, like also the magnetic field, gamma radiation from minerals
in the underlying rock, ground penetrating radar, etc.
An interesting example is measuring gravity from the air, something
that has become possible only because it is now possible to precisely
position the aircraft using satellite positioning (GNSS). As an example,
we mention the Ethiopian Airborne Gravity Survey executed in 2008
(Bedada, 2010). See figure 16.2.
Note that in the case of airborne gravity survey measurements, GNSS
gives not only the location of measurement, but also a correction to the
measurement: the motion of the aircraft causes varying disturbances to
the gravity measurement by the on-board gravimeter. Using GNSS, it is
possible to determine and eliminate this disturbance.
The gravimeter used in airborne gravimetry is very similar to that
used in sea gravimetry: it is strongly damped in order to make it perform

1 E.g., the aircraft airframe deforms between standing on the ground and being airborne,
and with changes in the amount of fuel and payload on board.

Ó »  Š.î á
182 A DVANCED SUBJECTS

F IGURE 16.2. The tracks of the Ethiopian Airborne Gravity Survey of 2008.
The gravity anomalies in mGal (10−5 m/s2 ) are colour coded. Total
D flight path is 90 000 km.

reasonably well in spite of the motions of the platform. It is mounted on a

gyroscopically stabilized platform inside the aircraft, so that it measures
the acceleration of free fall only along the direction of the local vertical or
plumbline.
The precision of measurement achievable by an airborne gravimeter
is of order 2 − 5 mGal, i.e., a 2 − 5 millionth part of ambient gravity, or
2 · 10−5 − 5 · 10−5 m/s2 . The motions of the aircraft cause accelerations that
may easily be 10 000 times greater, and these have to be eliminated. This
is where GNSS comes in. If it measures at a rate of one position x i per
second, the spatial acceleration at time t i can be estimated as

a i = 2x i − x i−1 − x i+1 ,

and the part of this acting in the local vertical direction n i at that point
of the track is simply a i = ⟨n i · a i ⟩.
Note that the gravity values obtained, typically expressed as grav-
ity anomalies2 , are at flight height, when most users (geophysicists,

2 Gravity anomalies are obtained by subtracting from the measured acceleration of

gravity, the gravity acceleration computed from a smooth, three-dimensional, global,
ellipsoidal or normal gravity field model, for the location, and especially the height, of

Ó »  Š.î á
Airborne geophysical surveys
183
geodesists) are interested in values at terrain height. Transforming the
complete set of measurements from flight height to terrain height (or sea
level) is called (harmonic) downward continuation, and belongs to the
domain of physical geodesy. The accuracy of the gravity anomalies at
terrain level is slightly poorer than at flight level.

measurement. Gravity anomalies contain the locally rapidly varying deviations of true
gravity from the smooth model gravities of the normal field, and thus extract the local
features of interest from the global overall field.

Ó »  Š.î á
D Rotation matrices

D A.1 Introduction
A
Always when we change the orientation of the axes of a co-ordinate
system, we have, written in rectangular co-ordinates, a multiplication
with a rotation matrix.
Let us investigate the matter in two dimensions in the ( x, y) plane,
figure A.1.
The new x co-ordinate, after rotation by an angle α, is

x′P = OU = OR cos α,

where
OR = OS + SR = xP + PS tan α = xP + yP tan α.
By substitution we find

x′P = ( xP + yP tan α) cos α = xP cos α + yP sin α.

Similarly the new y co-ordinate is

yP′ = OT = OV cos α,

′ y
y
Q P
α
′
x
V
α
T α U

α x
O S R

F IGURE A.1. Rotation in the plane

– 185 –
186 R OTATION MATRICES

where
OV = OQ − V Q = yP − PQ tan α = yP − xP tan α,

so substitution yields

yP′ = ( yP − xP tan α) cos α = − xP sin α + yP cos α.

We summarize in a matrix equation:

[ ] [ ][ ]
x′ cos α sin α x
= .
y′ − sin α cos α y

The place of the minus sign in this matrix is the easiest to ascertain by
making a paper sketch: draw both pairs of axes, mark the angle α, and
infer graphically whether, for a point on the positive x axis (i.e., y = 0),
the new y′ co-ordinate is positive or negative.
In the above case
[ ]
′
[ ] x
y = − sin α cos α = − sin α · x,
0

i.e., y′ < 0 for α > 0. So the minus sign is indeed in the lower left corner
of the matrix.

D A.2 Chaining matrices in three dimensions

In a three-dimensional co-ordinate system we may write a two-
dimensional rotation matrix as follows:
⎡ ⎤ ⎡ ⎤⎡ ⎤
x′ cos α sin α 0 x
⎣ y ⎦ = ⎣ − sin α cos α 0 ⎦ ⎣ y ⎦ ,
⎢ ′ ⎥ ⎢ ⎥⎢ ⎥

z′ 0 0 1 z

i.e., the z co-ordinate is just copied, z′ = z, while x and y transform into

each other according to the above formula.
If there are several transformations in sequence, we obtain the final
or compound transformation by “chaining” the transformation matrices.
I.e., if
r′′ = S r′ , r′ = R r,

in which
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
cos α sin α 0 1 0 0 x
R = ⎣ − sin α cos α 0 ⎦ , S = ⎣ 0 cos β sin β ⎦ , r = ⎣ y ⎦ etc.,
⎢ ⎥ ⎢ ⎥ ⎢ ⎥

0 0 1 0 − sin β cos β z

Ó »  Š.î á
 Orthogonal matrices
187
SR

z z = z′
z′
′′
R z
S y′′

y′ β y′
y y
α
x
x x′ x′′ = x′

F IGURE A.2. The associativity of rotations.

then (associativity):
r′′ = S (R r) = (SR ) r,

i.e., the matrices themselves may be multiplied with each other to obtain
the compound transformation.
Remember that
RS ̸= SR,

in other words, matrices and transformations are not commutative1 ! See

figure A.2.

D A.3 Orthogonal matrices

Rotation matrices are orthogonal, i.e.

RR T = R T R = I ; (A.1)

their inverse matrix is the same as their transpose.

E.g.,
[ ]−1 [ ]( [ ])
cos α sin α cos α − sin α cos (−α) sin (−α)
= = ,
− sin α cos α sin α cos α − sin (−α) cos (−α)

completely understandable, because this is a rotation around the same

axis, by the same amount, but in the opposite direction.
eq. A.1 can be written in index notation:
n
{
∑ 1 if j = k
R i j R ik = .
i =1
0 if j ̸= k

1 Two-dimensional rotations are in fact commutative; they can be described also by

complex numbers, or by simple rotation angles to be added.

Ó »  Š.î á
188 R OTATION MATRICES

We can also say that the columns R · i of a rotation matrix are orthonormal,
their norm (length) is 1 and they are mutually orthogonal. This can be
seen for the case of our example matrix:

 2 √
∑ 2
∥ R ·1 ∥ = √ R i1 = cos2 α + (− sin α) = 1,
i =1

 2 √
∑
∥ R ·2 ∥ = √ R i2 = sin2 α + cos2 α = 1,
i =1
2
∑
⟨ R ·1 · R ·2 ⟩ = R i1 R i2 = cos α · sin α + (− sin α) · cos α = 0.
i =1

Often we encounter less obvious orthogonal matrices:

1. The mirroring matrix for an axis, e.g., the y axis:
⎡ ⎤
1 0 0
. ⎢
M2 = ⎣ 0 −1 0 ⎦ ,
⎥

0 0 1

inverting the direction, or algebraic sign, of the y co-ordinate:

⎡ ⎤ ⎡ ⎤ ⎡ ⎤
x′ x x
⎢ ′ ⎥
⎣ y ⎦ = M2 · ⎣ y ⎦ = ⎣ − y ⎦ .
⎢ ⎥ ⎢ ⎥

z′ z z

2. The matrix that interchanges the x and y axes (a permutation):

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
0 1 0 x′ x y
def
P12 = ⎣ 1 0 0 ⎦ , i.e., ⎣ y′ ⎦ = P12 · ⎣ y ⎦ = ⎣ x ⎦ .
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥

0 0 1 z′ z z

3. Inversion of all axes:

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
−1 0 0 x′ x −x
X =⎣ 0 −1 0 ⎦ , i.e., ⎣ y′ ⎦ = X · ⎣ y ⎦ = ⎣ − y ⎦ .
⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥

0 0 −1 z′ z −z

Both M and P differ from rotation matrices in this way, that their determi-
nant is −1, when for rotation matrices it is +1. The determinant of the X
n
matrix is (−1) , with n the number of dimensions (in the above example,
3). A determinant of −1 means that the transformation changes a right
handed co-ordinate axes frame into a left handed one, and conversely.

Ó »  Š.î á
Orthogonal matrices
189
If we multiply, e.g., M2 and P12 , we obtain
⎡ ⎤
0 1 0
M2 P12 = ⎣ −1 0 0 ⎦ .
⎢ ⎥

0 0 1

The determinant of this is +1. However, it is again a rotation matrix:

⎡ ⎤
cos 90◦ sin 90◦ 0
M2 P12 = R 3 (90◦ ) = ⎣ − sin 90◦ cos 90◦ 0 ⎦!
⎢ ⎥

0 0 1

All orthogonal transformations having positive determinants are rota-

tions.
(We state still without proof, that all orthogonal transformations can be
written as either a rotation around a certain axis, or a mirroring through
a certain plane.)

Ó »  Š.î á
 Bibliography

Tulu Besha Bedada. Absolute geopotential height system for Ethiopia. PhD
thesis, University of Edinburgh, 2010.
http://www.era.lib.ed.ac.uk/handle/1842/4726. 177, 181

Claude Boucher and Zuheir Altamimi. Memo : Specifications for reference

frame fixing in the analysis of a EUREF GPS campaign.
http://users.auth.gr/kvek/20070327-MEMO-ver6.pdf, 2007. 172

Jorge Luis Nunes e Silva Brito, Rafael Alves Aguiar, Marcelo Teixeira Silveira,
Luiz Carlos Teixeira Coelho Filho, Irving da Silva Badolato, Paulo
André Batista Pupim, Patrícia Farias Reolon, João Araujo Ribeiro, Jonas
Ribeiro, Orlando Bernardo Filho, and Guilherme Lúcio Abelha Mota.
E-FOTO: Development of an Open-Source Educational Digital
Photogrammetric Workstation. pages 356–361, Barcelona, Espanha, October
2011. ISBN 978-1-61208-165-6. URL http://www.thinkmind.org/index.php?
view=article&articleid=icsea_2011_15_10_10194. 2, 17

Duane C. Brown. Decentering distortion of lenses. Photogrammetric

Engineering, 32(3):444–462, 1966. URL
https://eserv.asprs.org/PERS/1966journal/may/1966_may_444-462.pdf. 36, 53

B. F. Chao, Y. H. Wu, and Y. S. Li. Impact of Artificial Reservoir Water

Impoundment on Global Sea Level. Science, 320:212–214, 2008. 152

Alexander Eugen Conrady. Decentered lens systems. Monthly Notices of the

Royal Astronomical Society, 79:384–390, 1919. 36

Satu Dahlqvist, Petri Rönnholm, Panu Salo, and Martin Vermeer. Evaluating
the Correctness of Airborne Laser Scanning Data Heights Using
Vehicle-Based RTK and VRS GPS Observations. Remote Sensing, 3(9):
1902–1913, 2011. ISSN 2072-4292. doi: 10.3390/rs3091902. URL
http://www.mdpi.com/2072-4292/3/9/1902. 179

Michael Gruber, Martin Ponticelli, Stefan Bernögger, Bernhard Reitinger,

Hannes Hegenbarth, Zanin Cosic, and Alexander Wiechert. UltraCam Eagle,

– 191 –
192 B IBLIOGRAPHY

understanding the new sensor. In ASPRS 2012 Annual Conference,

Sacramento, California, March 19-23, 2012. http://www.asprs.org/a/
publications/proceedings/Sacramento2012/files/Gruber.pdf. 32

Maurice S. Gyer. Methods for computing photogrammetric refraction

corrections for vertical and oblique photographs. Photogrammetric
Engineering and Remote Sensing, 62(3):301–310, March 1996. http:
//info.asprs.org/publications/pers/96journal/march/1996_mar_301-310.pdf,
accessed Sep 7, 2017. 25

Zerfu Hailu and David Harris. Rural land registration in Ethiopia increased
transparency for 26,000,000 land holders. In Annual World Bank Conference
on Land and Poverty, Washington DC, 2014.
http://www.oicrf.org/pdf.asp?ID=13778. 145

JUHTA. Drafting a Local Zoning Map (in Finnish). Web page, Advisory Board
for Information Management in Public Administration, JHS 185.
http://www.jhs-suositukset.fi/suomi/jhs185, read May 25, 2016. xv, 150

K. Kraus. Photogrammetry: Geometry from Images and Laser Scans. Second

edition. Walter de Gruyter Textbook. De Gruyter, 2007. 459 pages, transl. I.A.
Hartley and S. Kyle. 7, 25, 26

Linifiniti Consulting CC. The Free Quantum GIS Training Manual, 2008,
maintained. URL https://docs.qgis.org/2.8/en/docs/training_manual/. CC BY
4.0, Accessed May 8, 2019. 17

D.G. Lowe. Object recognition from local scale-invariant features. In

International Conference on Computer Vision, pages 1150–1157, Corfu,
Greece, 1999. 125

C. Mueller and K. Neumann. XL-3/W4:1–9, 2016. doi:

10.5194/isprs-archives-XL-3-W4-1-2016. URL http://www.
int-arch-photogramm-remote-sens-spatial-inf-sci.net/XL-3-W4/1/2016/. 33

Matti Ollikainen. Transforming GPS Co-ordinates to the National Map Grid

Co-ordinate System KKJ (in Finnish). Tiedote 8, Finnish Geodetic Institute,
Helsinki, 1993. xv, 170, 173

Optical Metrology Centre. Camera Calibration Methods, pages 4–1 to 4–15.

undated. xnatmap.org/adnm/pub/camcal/hbook4.pdf. 37

Public Domain Aeronautical Software. Properties Of The U.S. Standard

Atmosphere 1976, 2014. http://www.pdas.com/atmos.html, accessed Sep 7,
2017. 25

G.H. Schut. Photogrammetric refraction. Photogrammetric Engineering, 35(1):

79–86, 1969. 25

Ó Š.î á
B IBLIOGRAPHY
193
M.T. Silveira. Considerações Técnicas sobre o Submódulo de Extração do MDE
da Versão Integrada do e-foto (versão educacional). Technical report, Rio de
Janeiro, 2011. 133

M.T. Silveira, R.Q. Feitosa, K. Jacobsen, and J. L. Brito. Correspondência

Eficiente de Descritores SIFT Para Construção de Mapas Densos de Pontos
Homólogos em Imagens de Sensoriamento Remoto. Boletim de Ciências
Geodésicas, 17:130–160, 2011. 125

Ó Š.î á
 Index

A B
Adindan datum (Ethiopia), 60, 93, 177 Bahir Dar, Ethiopia, 4, 39, 61
adjustment balloon, 8
least-squares, 95 base map, topographic, 9
theory, 11 base mapping, 3
aerial camera base network, geodetic, 152
analogue, 65 basics of photogrammetry, 2
digital, 66 beta βB (Greek letter), 125
aerial mapping, 149, 151 block, photogrammetric, 41
aerial survey, 152 boundary
aerotriangulation, 2, 29, 47, 94–97, 101, parcel, 150
106–108 real-estate, 149
numerical, 95 Brown, Duane, 36
affine transformation, 67, 125 bundle block adjustment, 95
African Reference Frame (AFREF), 174
airborne geophysical survey, 181 C
airborne gravity survey, 181 cadastral mapping, 145
airborne laser scanning (ALS), 2, 28, 141, cadastral system, 1, 149
144, 146, 152, 174, 179 camera, 8
aircraft aerial
body co-ordinates, 180 traditional, 34
motorized, 1 digital
airship off the shelf, 34
lighter than air, 1 digital photogrammetric, 32
alpha αA (Greek letter), 22, 125 height, 23, 75
anaglyph superposition, 118 location, 14, 45, 97, 103, 109
anaglyphic glasses, 1, 8 metric, 35
antenna calibration, GNSS, 179 orientation, 9, 14, 45, 75, 79, 88, 103,
apparent uplift, 23 106, 109
approximate value, 15, 29, 53, 80, 88, 98, traditional photogrammetric, 31
99, 174 camera calibration, 2, 31, 35, 66, 74, 77
associativity model, 12
of matrix multiplication, 187 camera co-ordinates, 65–67, 69, 75, 77, 80,
atmosphere, effect of, 12 88
atmospheric refraction, 21 linearized, 97
Automatic Extraction, 137 of feature point, 114
aviation, 8 of ground control point, 101
azimuth, geodetic, 156 of tie point, 100

– 195 –
196 ABCDEFGHIJKLMNOPQRSTUVWXYZ I NDEX

camera model, 98, 99 deformation

Camera obscura (dark room), 8 in image, 125
canopy, forest, 141, 144 of airframe, 180
central meridian, 27 of image patch, 136
centre of mass of reservoir dam, 152
common, 161 Delaunay triangulation, 129
of the Earth, 153, 154, 161–163, 170 delta δ∆ (Greek letter), 80
centre, of reference ellipsoid, 161, 163 DEM Extraction, 133, 137, 138
charge-coupled device (CCD), 32, 66 DEM Extraction module (e-foto), 133, 141
Clarke 1880 (reference ellipsoid), 60 DEM Interpolation, 133, 138
collinearity equations, 76, 78 design matrix, 16, 80, 81, 84, 100
exact, 79 determinant, 188, 189
commutativity diagonal matrix, 13
of matrix multiplication, 187 difference of Gaussians, 126
conjugate gradients, 95 digital elevation model (DEM), 127, 129,
construction 130, 141, 145, 146, 152, 179
infrastructure, 1 digital map, 150
continuously operating reference station digital surface model (DSM), 143
(CORS), 93 digital terrain model (DTM), 2, 133, 141
co-ordinate frame dispersed settlement, 150
camera, 65, 68 distortion
film, 67 decentering, 37
co-ordinate measurement machine, 65 radial, 38
co-ordinate reference frame, 153, 154 downward continuation, harmonic, 183
co-ordinate reference system, 153 drinking water, 129
co-ordinate system Dutch-Indies, 9
geocentric, 157 dynamic range (of image sensor), 33
terrain, 75 Dürer, Albrecht, 8
three-dimensional, 186
topocentric, 157 E
instrument, 157 Earth curvature
corkscrew rule, 75 effect, 27
cross-correlation, 136, 138 reduction, 27, 30
crosshair cursor, 114, 119, 121 e-foto (software), 2, 3, 5, 17, 49, 59, 65, 69,
curvature 85, 88, 89, 91, 97, 106, 107, 117,
of Earth’s surface, 26 118, 124, 134, 142, 148
curvature, of Earth’s surface, 12 e-foto development team, 38
EGM08 (geopotential model), 176, 177
D electricity, 40, 151
date line, 51 generation, 151
datum electromagnetic centre,
geocentric, 154, 155 of GNSS antenna, 144, 179
geodetic, 154 elevation angle, 156
traditional, 160 ellipsoidal co-ordinates, 153
global, 154 ellipsoidal normal, 160, 163
local, 155 direction, 166
national, 155 emacs (text editor), 63
non-geocentric, 155 emulsion, light-sensitive, 7
traditional, 155 encumbrance, 149
vertical, 129 ensemble, of datum points, 161
datum defect, 154 epipolar line, 111
datum point, 160, 161 ERDAS Imagine (software), 17, 97, 142
Debian Linux, 18, 58 Espoo, Finland, 4
debugging, 104, 107, 108 estimates, least-squares, vector of, 13
deciduous tree, 45 eta ηH (Greek letter), 65, 156, 160
Defense Department, United States, 162

Ó Š.î á
I NDEX ABCDEFGHIJKLMNOPQRSTUVWXYZ
197
Ethiopian Airborne Gravity Survey, 177, Gauss-Krüger map projection, 173, 177
181 GDAL (software), 17
ETM+ sensor, 34 gedit (text editor), 63
ETRS89, 172 General Boundaries system, 149
Euler angles, 28, 29, 79 geocentric co-ordinates
EUREF89, 162, 169 rectangular, 163
EUREF-FIN, 173 geodesy, 1
European Datum 1950 (ED50), 161, 169 physical, 183
European Petroleum Survey Group geodetic measurement, 153, 160
(EPSG), 17, 61 geodetic network, 153
European Reference Frame, IAG geodetic survey, 149
Subcommission for the (EUREF), geographic information system (GIS), 17
169, 172 geography
European Terrestrial Reference System physical, 130
(ETRS89), 172, 173 geoid height, 160, 161, 166
Extensible Mark-up Language (XML), 62 geoid model, 29, 129, 152
exterior orientation, 2, 56, 75, 85, 86, 91, geometry, 7, 8
95, 97, 106, 107, 145, 179, 181 geophysical survey, 151
computation, 87 airborne, 181
elements, 34, 75, 103, 104, 113 glass plate, photographic, 7
Exterior Orientation module (e-foto), 56, 85, glasses
101, 103, 113, 121, 133 anaglyphic, 121
polarized, 115
F Global Navigation Satellite Systems
feature, 110, 116, 117 (GNSS), 28, 92, 144, 152, 154,
feature extraction, 2, 134 155, 157, 167, 174, 179, 181, 182
feature file, 141 Global Positioning System (GPS), 28, 154,
feature matching, automatic, 124 162
feature recognition, automatic, 127 GNSS
fiducial cross (FC), 37 geodetic measurement, 93
fiducial mark, 35, 37, 65–74 relative measurement, 92
field calibration, 181 static measurement, 92
figure of the Earth, 1 gravimetry
film co-ordinates, 65, 66, 68, 69 airborne, 181
film grain, 33 sea, 181
film, photographic, 7 gravity anomaly, 182, 183
flight direction, 45 ground control point (GCP), 26, 28, 29, 40,
dial, 87 47, 75, 81, 84, 85, 92, 96, 97, 100,
flight line, 32, 39 101, 107, 121, 133, 134, 181
flight map, 45 marking, 93
flight plan, 2, 41 ground control points (GCP), 91, 95
flight strip, photogrammetric, 41 ground control points, multiple, 82
floating mark, 8, 11, 111, 131 GRS80 (ellipsoid), 27, 173
floor, forest, 144
fluid flow, 129 H
FM International Oy FINNMAP (mapping half-angle formula, 157
company), 4 hand cursor, 122
focal length, 27, 31, 32, 35, 41, 46 handwheel, 114
calibrated, 36, 77 Heerbrugg, Switzerland, 38
footwheel, 114 height
Frauenkirche (Munich), 161 above sea level, 26, 129
free fall, acceleration of, 182 normal, 26, 129
orthometric, 26, 129
G height colour, 130
Gauss, Carl Friedrich, 11 height contour, 130
Gaussian blur, 126 Helmert transformation

Ó Š.î á
198 ABCDEFGHIJKLMNOPQRSTUVWXYZ I NDEX

three-dimensional, 167, 171 land-use planning, 151

two-dimensional, 173 laser scanning, 3, 7
Helsinki, Finland, 3, 18, 38 latitude
homologuous point, 125, 132–135, 138 astronomical, 159, 160, 166
horizon camera, 9 geodetic, 166
Laussedat, Aimé, 8
I least squares, theory of, 11
i.i.d., 16 least-squares adjustment, 174
image co-ordinates, 65, 74 least-squares estimation, 68, 76
image correlation, 97 least-squares matching, 125, 132, 136, 138
image pyramid, 123 least-squares method, 14
ImageMagick (software), 58 least-squares solution, 83
imagery ordinary, 85
aerial, 8 Legendre, Adrien-Marie, 11
photographic, 2 Leica Photogrammetry Suite
satellite, 7 (LPS), 124
independence, statistical, 13 Leica RC30 (camera), 40
inertial measurement unit (IMU), 9, 28, levelling network, 154
29, 58, 179 levelling, precise, 129, 151
inertial navigation system (INS), 57 Levenberg-Marquardt algorithm, 95
infrastructure, 129, 151 Light Detection And Ranging (LiDAR),
initial approximate values, exterior 146
orientation, 85 linearization, 14, 80, 98, 164
Interferometric Synthetic-Aperture local astronomical co-ordinates, 155
Radar (InSAR), 143 longitude
interferometry, 143 astronomical, 158–160, 166
interior orientation, 2, 65–69, 74, 77, 95, geodetic, 166
145
Interior Orientation module (e-foto), 56, 69, M
73, 85 machine vision, 132
International Association of Geodesy map
(IAG), 154, 169 colour, 7
International Earth Rotation and greytone, 7
Reference Systems Service topographic, 9
(IERS), 171 map co-ordinate system, 29
International Ellipsoid of 1924 map projection, 26, 29
(Hayford), 169, 177 mapping
International Terrestrial aerial, 5
Reference Frame (ITRF), 28, 50, 154, 162, matrix of Jacobi, 15
171, 172 measurement class, 149, 150
International Terrestrial Reference meridian
System (ITRS), 93, 154 central, 29
irrigation, 129, 151, 152 local, 29
ITRF2008, 162 meridian convergence, 29
microseismicity, 152
K mirroring, 188
Kalman filter, 28 mobile telephony, 129
kappa κK (Greek letter), 28 Monteverdi/OTB (software), 17
kite, 8 mortgage, 149
KKJ (Finnish national map grid mu µM (Greek letter), 168
co-ordinate system), 60, 170, 172, 175 multipath, GNSS, 179
Multiplex (stereo restitution
L instrument), 8
lambda λΛ (Greek letter), 28, 158 Munich, Germany, 161
Landsat 7, 34
land-use inventory, 151 N
Nenonen, Vilho, 9

Ó Š.î á
I NDEX ABCDEFGHIJKLMNOPQRSTUVWXYZ
199
Nepal, 45 photography, aerial, 9
North, East, Up (NEU) system, 156, 167 phototriangulation, 58, 95
Phototriangulation module (e-foto), 104, 113,
O 114, 121, 133
observation equations, 12–14, 16 pitch, 29
aerotriangulation, 95, 97, 101 pixel, 7
camera calibration, 36 pixel array, 7
exterior orientation, 76, 78, 81 plate tectonics, 171
initial values for exterior orientation, plumbline, 160, 161, 182
84 direction, 166
linearized, 82 plumbline deflection, 160, 166
observations regional, 161
linearized, 100 point of interest, 126
observations, vector of, 11, 13, 84 port (nautical), 88
omega ωΩ (Greek letter), 28 precise ephemeris, 93, 172
operator, of aerial camera, 77, 88 principal point, 35, 36, 65–67
optical centre, 35 proj4 (software), 17, 175
optics, 8 project definition file, 49, 62, 69, 85, 91
orbit determination Project Manager module (e-foto), 49, 56, 62,
satellite, 162 69, 85, 86, 89, 101, 102
orientation pushbroom sensor, 31, 34
absolute, 110 Pythagoras theorem, 27
relative, 97, 109
orthogonality Q
of rotation matrices, 187 quadratic form, 14
orthographic projection, 145 quality control, of networks, 14
orthometric height, 166 quality, of least-squares solution, 76
orthophoto map, 1, 2, 142 Quantum GIS (software), 17
orthophoto mapping, 145, 146 quotient rule, of differentation, 81
orthophoto production, 141
ortho-rectification, 2, 7, 142, 143, 147, 148 R
overdetermination radial distortions, 35
absolute orientation, 110 radiometry, 7
exterior orientation, 76, 91 rainfall, 129
overlap raster file, 129
longitudinal, 41, 43, 53 raster format, 131
of images, 41 real estate, 149
transversal, 41, 53 realization, of a co-ordinate reference
system, 154
P real-time kinematic (RTK) GPS, 150
parallax, 110, 111, 118 rectangular co-ordinates, 153, 185
partial derivative, computation, 81 rectangular geometry, 27
Pearson correlation, 136, 138 Red-Green-Blue (RGB),
permutation, 188 visual primary colours, 33
personal computer, 1, 10, 17 redundancy, 97
perspective, 8 reference ellipsoid, 28, 29, 159
perspective drawing, 8 reference frame, 2
perspective image, 131 geocentric, 28
phi ϕΦ (Greek letter), 28, 159 reference system
photogrammetry, 1, 5, 7 vertical, 151
aerial, 7, 10, 12 refractive index, 21
digital, 1, 17 Region Growing step, 137
history, 3, 7 relief shading, 130
numerical, 11 remote sensing, 7
soft-copy, 32 resection, spatial, 75, 85
terrestrial, 7 reservoir dam, 129, 151, 152

Ó Š.î á
200 ABCDEFGHIJKLMNOPQRSTUVWXYZ I NDEX

residual, 68 stereo restitution, 8, 109, 146

residuals, vector of, 13 stereo restitution instrument, 109, 131
Ringroad II (Helsinki), 3 stereo vision, 8, 109, 110, 114
Rio de Janeiro, Brazil, 3, 18, 59, 119, 136, Stereoplotter module (e-foto), 113, 114, 141
146 stereoscopy, 8
robustness, of a solution, 92 stochastic quantity, 12
roll, 29 Stolze, Franz, 8
root-mean-square error (RMSE), 69, 73, stratification, of the atmosphere, 21
74, 106 structural integrity, of reservoir dam, 152
rotation angle, 75 synthetic aperture radar (SAR), 141
rotation axis
of the Earth, 153, 154 T
rotation matrix, 173, 176, 185, 188, 189 tangent plane, 27
of camera, 78, 79 tau τT (Greek letter), 23
two-dimensional, 186 Taylor expansion, 15
terrain co-ordinates, 2, 75, 77, 86, 88
S camera-rotated, 78
satellite positioning, 129, 171 terrain height, 78, 132, 143, 144, 152, 183
scale factor, 78, 168 average, 85
Scale Invariant Feature Transform terrain slope, 125, 131
(SIFT), 125, 132 theodolite, 154
scale, image, 41 tide gauge, 154
scanner tie point, 14, 40, 47, 95, 96, 100, 101, 124
photogrammetric image, 31 tie-point generation, automatic, 40
Schermerhorn, Willem, 9 Tikhonov regularization, 95
science museum, 10 tilt angle, 75
seed editor, 134–136 timing, of photography, 45
seed point, 132–134, 138 topocentric co-ordinates, 155
sewage, 129 rectangular, 156
Shuttle Radar Topography Mission spherical, 156
(SRTM), 178 traffic, 129
sigma σΣ (Greek letter), 12, 16 translation (parallel shift), 167
signal, 93 triangulated irregular network (TIN), 130
signalization, 92, 150 tropics, 46
similarity transformation, 67, 168
simplification, of U
observation equations, 80, 82 Ubuntu Linux, 5, 18, 58, 63
Snellius, Willebrord, 21 UltraCam Eagle (camera), 32
Snell’s law, 22 uncertainty
solution of calibration measurements, 35
least squares, 16 of measurement, 12
solution, least-squares, 14 of observations, 12, 16
Spatial Resection module (e-foto), 114 of transformation parameters, 170
exiting, 89 understorey, forest, 144
spherical co-ordinates, 153 unit matrix, 13, 169
spirit level, 154 unit weight
stabilized platform, 182 standard error of, 13, 16, 69, 73, 89
Standard Atmosphere 1976, 25 variance of, 69, 76
standard deviation, 12 Universal Transverse Mercator (UTM,
standard error, 12 map projection), 27, 51, 59, 60,
starboard (nautical), 77, 88 93, 173, 177
State University at Rio de Janeiro (Brazil), unknowns, vector of, 11, 14, 84
2, 17 linearized, 98
stereo model, photogrammetric, 8, 110, unmanned aerial vehicle (UAV), 1, 34, 40
145 UTM zone, 51
stereo plotter, 11

Ó Š.î á
I NDEX ABCDEFGHIJKLMNOPQRSTUVWXYZ
201
V
variance matrix, 13
of unknown estimates, 16
verification point, 179
verification surface, 179
vertical photogrammetry, 80
vi (text editor), 60
vim (text editor), 63
Vinci, Leonardo da, 7
Väisälä, Yrjö, 9

W
wavelet, 126
weather
uncertainty, 46
weight, 14
weight coefficient matrix, 13
weight matrix, 14
whiskbroom sensor, 31, 34, 143, 179
Windows, 5, 18, 58
work station, photogrammetric, 2, 10
workflow, photogrammetric, 10, 17, 33, 144
World Bank, 145
World Geodetic System 1984 (WGS84), 93

X
xi ξΞ (Greek letter), 65, 160

Y
yaw, 29

Z
zenith angle, 156
zeta ζZ (Greek letter), 156
zoning, 1, 149

Ó Š.î á