POLITECNICO DI TORINO

DIATI

Photogrammetry - part 2
Interior orientation, camera and exterior orientation

MARCO PIRAS
GEOMATICS: 01RVUMX
 Photogrammetry part 2
Summary

Interior orientation

Camera classification
Calibration of «traditional» camera
Aerial metric camera

Semi-metric camera, professional camera and massa market camera

Digital images
Indirect data acquisition

Direct data acquisition

Calibration of digital camera
 Steps of photogrammetric process
Principles of Photogrammetry

The Photogrammetry allows: O

- If at least two frames are given
- And measuring the image
coordinates (x1, 1) and (x2, 2)
about the homologous points

I1
To estimate the 1
coordinates of the object
(X, Y, Z) of points in the c x1 I2 2
object space
Image 1 x2
C1 c

C2 Image 2
 Interior orientation
Under these conditions:
• It is possible to measure the image coordinates:
- Definition of the image reference system
- To be able to measure the image coordinates (point on the frame)

• the position of the projection center is known (internal orientation)

• the position of the projection center and its attitude are known in the object
reference system (exterior orientation)

Considering:

The «ideal» central perspective is not satisfied due to many operative conditions as
errors and distortions

The camera are different, in particulare considering errors and distortions.

 Interior orientation
Image coordinate system

z PP = principal point
k j (Initially, we can consider the
 principal point (PP) as projection
C (X0,Y0,Z0)
x of the center of acquisition (C) on
w the plane of frame. This is a first
-c  approximation)

PP I
x

Z
Y
P
X
 Interior orientation
Image coordinate system
 Interior orientation
Image coordinate system

The parameters are: z

• PP (x0, 0) coordinates


• c principle distance 0
PP

x CF x0
x


-c 
x0

PP I
x
x
CF

0
 Interior orientation
Definition of the image coordinate system
 

CF CF
x x

Zeiss Wild
 Interior orientation
Definition of the image coordinate system

Classification of the used camera in photogrammetry

• metric camera : Interior orientation is known (by calibration) and

stable during the time
• semi-metric camera: Interior orientation is known (by calibration)
and it can change during the time
• professional camera (and mass market): Interior orientation is
unknown and variable during the time
 Interior orientation
Differences between models

z Theoretically, C, P, I shoul be aligned,

respecting the rigorous laws of the
 central perspective.
C
x Practically, the point P is in I’ (and not in
I), because the image is affected by
 several distortion.
I’ (x ‘, ’) The differences (Dx, D) have to be
limited, in a specific range
x
I (x , ) Knowing the deformation law or its
(Dx, D) approximation (Dx, D) , it is possible to
calculate(x ‘, ’), in order to obtain a set
of coordinates more similar to the
Z «theoretical» ones (x , )
Y
P
X
 Digital images
In digital photogrammetry, the primary data Definition
is defined as digital image, in which the i

photographic content (radiometry) is stored

as numbers.
This representation (raster) is obtained j

subdividing the image into elementary

elements of finite dimensions (pixels, not
more divisible) and associating to each of
them the number (or numbers) that
gij
represents the radiometry of the portion of
image contained .
Each pixel can be seen as an element of a
matrix and therefore be univocally  1 9 13 ... 124 
identified by two integers that represent the  2 34 25 ... 100 
 
row and column position of the elementary 50 20 32 ... 90 
element within the matrix.  
 ... ... ... ... ... 
The pixel has a fixed position a priori that can 60 56 52 ... 240
change during the time.
 Digital images
Conventional radiometric definition (B/W)
If the digital image represent an object composed by two colors (a black-white
portrait), the radiometric information is defined only by two integer numbers: 0 =
white 1= black
The data storage requires only 1 bit for each pixel.
If the digital image represents an object in
gray scale (e.g black-and-white image),
radiometry can be expressed with an
integer variable between 0 (black) and 255
(white). Intermediate values at these two
extremes represent the various gray scale.
The black-white interval is divide into 256
parts for two main considerations:
- the human eye can distinguish about 80
different grays
- an integer between 0 and 255 can be
stored in 1 byte
 Digital images
Conventional radiometric definition (color)
If the image has to represent an coloured object, there are two possible approaches
to define the radiometry:

True color Image (RGB)

Each color is described as the sum of
three bands corresponding to the
main colors (red, green and blue)
Each band is represented by 256 values
from 0 (= no color) to 255 (= color
saturation).
The radiometry of a pixel is
represented by three integers which
describe the saturations of the three
main bands; this representation
requires 3 bytes for each pixel
 Digital images
FORMATS
The numerical data that represent the digital image must be memorized in
order and according to an organization that allows to faithfully reconstruct
the image. The set of rules that manage the storage is called FORMAT.

The format must:

- ensure minimum memory size;
- be able to have a compression format
- be a public domain
- allow to store all radiometric scales

The formats more used in photogrammetry are TIFF and JPEG

TIFF format (Aldus and Microsoft) allows to store more images (as pyramids) in
one file .
Moreover, it allows to use a compression which not reduce the information
and quality (LZW). In RGB image, this compression brings to reduce at 50% of
memory size.
 Digital images
FORMATS
JPEG ,JPEG 2000 and ECW can store the images, in two differerent
methods:
- LOSSLESS , without decreasing the image information, as LZW in TIFF
format
- LOSSY , some information is lost but
without a remarkable reduction of the
quality. The images are based on DCT o
DWT transformation.

Geometrical degradation is shown as:

1. Blurriness effect on the radiometric edge;
2. Object shifting in the image
3. Aliasing and blocking

These effects could be both local or global, and their values depend
on the lightness intensity variation and compression ratio
 Digital images
Geometric degradation vs compression ratios
Geometric degradation can be represented by standard deviation differences in the
markers positions between the compressed image and the uncompressed original
image. These positions are expressed in image coordinates as pixel units.

The trend is almost

linear with low
compression ratios,
while it is very strong
variation (that is not
acceptable in
photogrammetry) with
compression values in
the range 1/7 - 1/10.
 Digital images
Geometric degradation vs compression ratios
Geometric degradation has been also verified in photogrammetric plotting. The same points
were plotted within a model based on uncompressed images (reference data) and in the
various models composed by images with increasing compression ratios. The differences can
be represented as standard deviation, in object coordinates.

Using JPEG, the maximum level of compression is = 7. ECW or JPEG2000 format can be
adopted up to a compression level = 30.
 Digital images
Metric content
It is necessary to define a
Coordinate system of the Dx/2 Dx
image (x,) , which allows i x
to combine each pixel to
D/2
the real coordinates
In this coordinate system, the
pixel has a integer size j
(Dx, D) and its position   D * j
is defined on the center of D
the cell j = fix( D0

It is possible to define a
biunivocal relationship
between the pixel position
in the image (i,j) the the
x  Dx  i
interior coordinates (x,).
i = fix( xDx0 gij

 Digital images
Acquisition step
The traditional measure of coordinates (analytical plotter) is replaced by the
identification of the pixel in the image matrix (digital plotter).
This position is defined as real coordinates considering the equation previously
mentioned.
The position of the pixel is fixed and it is the acquisition system which
associates the corresponding radiometric value to a determined pixel

Using a digital image, no coordinate measurements are required: it

is sufficient to identify the correct pixel in the image matrix.

All measurements (the position of the pixels) have already been

made by the acquisition system (which is the real measuring
instrument)
 Acquisition of digital images
Digital images can be acquired in two different methods:
Direct acquisition: the image is created by digital camera on the digital support. The
camera could be metric or not-metric.
Indirect acquisition: Image is created making a scan of a traditional image
(hardcopy). In this case, it is necessary to use special scanner (photogrammetric
or DTP), in order to reduce the distortions.

Read out

Read out

A/D converter
255

0
0
 Acquisition of digital images
Direct acquisition
How does a digital camera works ?

The image storage is based on Electronic principles. The support where the
image is stored is called «digital sensor»

The digital sensor is composed by a matrix of elements (pixels), where each of which
is sensitive to a primary color. Each pixel consists of a capacitor able to record the
amount of energy (light intensity) incident and to convert it into a integer value
 Acquisition of digital images
Direct acquisition
Charge Coupled Device (CCD)
The CCD sensor is composed by a matrix of photodiodes
that transform the energy of incident light into an electric
charge. It is also called a "charge transfer" device because
the pixels that compose it also perform the function of
transferring it sequentially from one to the other.

The "captured" light energy is

transformed into an electric
charge which is transferred along
the columns until it reaches the
shift register: it is transformed
into d.d.p (digital development
processing) and sent to an
analog-digital converter
 Acquisition of digital images
Complementary Metal Oxide Semiconductor (CMOS) Direct acquisition

It is similar to CCD sensor. In

this case, the electronic
conversion operations of the
signal from charge to d.d.p. is
realized in each photodiode.

No charge scrolling is required, but simply read the

value of d.d.p. of the pixel. These sensors are much
faster.
 Interior orientation
In photogrammetric cameras, the lens is a Metric camera
complex lens system to limit aberrations.

The diaphragm (aperture) is not at the

center of the lens, so the first problem
consists in identifying the projection
center.

Optical
Asse axis
ottico

Diaframma
Diaphragm
 Calibration
Metric camera
Principal optical
piani ottici planes
principali Image’s plane
piano dell'immagine
Objectspazio
spaceoggetto H H'
Image
spazio space
immagine P’ punto
point in
P'
image space
immagine

Optical axis  '’


asse ottico N
N=O PP punto
PP
  N’
N'=O' principale
principal
raggio centrale point
Object point
punto
oggettoPP
s ee s'=c
distanza oggetto distanza immagine
Object distance Image distance
In each optical system there are two principal planes that intersect the optical axis in N
and N ', called nodal points (external and internal)
N can be considered the center of projection in the object space and N ‘ in the image
space
With respect to the simplified schema (lens as point), the two stars of directions,
external and internal, are separated by the segment "e", but the parallelism ensures
the maintenance of directions
Calibration of metric camera
Correction of lens distortions
 Interior orientation
Correction of the distortion
It is possible to correct the distortion by:
• a physical model when an analytical relationships of the distortion effect
(systematic aspect) can be estimated
• a mathematical model when the distortion effect depends on many factors
which are not estimated with precision (pseudo-accidental relationship).

The main perturbative effects of the theoretical model are :

CAMERA: principal distance physical model

fiducial center and PP

LENS: LENS distortion physical model

IMAGE: stretching of the film mathematical model

temperature, humidity and conservation
no planarity of the film
sensors geometry
 Acquisition of digital images
The geometry of the film support In the digital sensors
The film support has several deformations: The sensor is produced on a silicon support
- no planarity of the film and it is quite stable during the time in
- mechanical stress (stretching ) geometric terms. It does not present any
- chemicals stress (development). significant planarity problems for structural
reasons or chemical stresses.
It can have some mechanical stress (e.g.
sensor cleaning devices). If this camera is
The sensor could be considered used for photogrammetric purposes this
geometrically stable during the time function must be disabled.

There are some effects due to the sensor geometry.

Not suqared
1) Square pixel pixel
2) Lines Lines not
perpendicular perpendicular to
to the columns the columns

Theory Real case

 Calibration of metric camera
Lens distortions
The incident ray passes through the lens with an incident angle equal to  and it
has a final angle equal to ’  
Lens of the metric camera have a spherical symmetry perfectly defined: in this case,
the distortion has mainly a radial component.

Dr = f( r ) r '  c tan   Dr 

piani ottici principali piano dell'immagine I’

H H' I’
spazio oggetto spazio immagine
P'I punto
immagine r’ Dr
I
asse ottico N=O 
 '  ’ PP punto
 N'=O' principale
raggio centrale x
punto
oggetto P
s e s'=c
distanza oggetto distanza immagine
 Calibration of metric camera
D[m]
Correction of lens distortions C

D max Dmin
5
0  [mm] c

-5 30 60 90 120 150
kc

r [mm] 10 20 30 40 50 … 160 165

Table description
Dr [m] +2 +5 +6 +3 +1 -2 -1

The curve of distortion can be described by a polynomial function,

with an odd degree in r
 Dr  a1r 3  a2 r 5  a3 r 7  ...
x  x0
Dx1  Dr  (x  x 0 ).(k1r 2  k 2 r 4  k3 r 6  ...)
I’ r
 0
I D1  Dr  (   0 ).(k1r 2  k 2 r 4  k3 r 6  ...)
r
r Dr
0 PP Modified collinearity equation
CF x0
x r ( X  X 0   r21 (Y Y 0   r31 (Z  Z 0  Z
x  x 0  Dx1  c 11  x 0  Dx1  c x
r13 ( X  X 0   r23 (Y Y 0   r33 (Z  Z 0  N
r ( X  X 0   r22 (Y Y 0   r32 (Z  Z 0  Zy
  0  D1  c 12  0  D1  c
r13 ( X  X 0   r23 (Y Y 0   r33 (Z  Z 0  N
 Calibration of «commercial» camera
Characteristics Correction of lens distortions

They are the «ordinary» commercial digital camera, with

good lens but not with particular quality for
photogrammetric applications.

Focus is continous.
Non dispongono di dispositivi di spianamento pellicola
Il formato della pellicola è 24x36 mm

It has «NO SENSE» to calibrate the system based on

camera-lens.
Interior orientation and distortions are unknown and
variable during the time

Autocalibration procedure is often adopted.

 Calibration of «commercial» camera

x  x0
Dx1  Dr  (x  x 0 ).(k1r 2  k 2 r 4  k3 r 6  ...)
r D I’ Dt
 0
D1  Dr  (   0 ).(k1r 2  k 2 r 4  k3 r 6  ...) I
r PP
Dr
r
In addition to radial distortions, there are:
x
Tangential distortions
 
Dx 2  p1 (r 2  2x 2 )  2 p2x (1  p3 r 2 )
D 2   p (r
2
2
 2 2 )  2 p x (1  p r
1 3
2
) Autocalibration will be discussed later.
Modified collinearity equations
r11 ( X  X 0   r21 (Y Y 0   r31 (Z  Z 0  Zx
x  x 0  Dx1  Dx 2  c  x 0  Dx1  Dx 2  c
r13 ( X  X 0   r23 (Y Y 0   r33 (Z  Z 0  N
r12 ( X  X 0   r22 (Y Y 0   r32 (Z  Z 0  Zy
  0  D1  D 2  c  0  D1  D 2  c
r13 ( X  X 0   r23 (Y Y 0   r33 (Z  Z 0  N
 Calibration of digital camera
Radial distortions
x  x0
Dx1  Dr  (x  x 0 ).(k1r 2  k 2 r 4  k3 r 6  ...)
r
 0 x
D1  Dr  (   0 ).(k1r 2  k 2 r 4  k3 r 6  ...)
r
x0 0
Tangential distortions PP
 
Dx 2  p1 (r 2  2x 2 )  2 p2x (1  p3 r 2 )
D 2   p (r 2
 2 2 )  2 p x (1  p r 2
)
2 1 3

Distortion of the sensors
Dx 3  c1 (x  x 0 )  c2 (   0 )
Modified collinearity equations

r11 ( X  X 0   r21 (Y Y 0   r31 (Z  Z 0  Z

x  x 0  Dx1  Dx 2  Dx 3  c  x 0  Dx1  Dx 2  Dx 3  c x
r13 ( X  X 0   r23 (Y Y 0   r33 (Z  Z 0  N
r12 ( X  X 0   r22 (Y Y 0   r32 (Z  Z 0  Zy
  0  D1  D 2  c  0  D1  D 2  c
(  ( 
r13 X  X 0  r23 Y Y 0  r33 Z  Z 0(  N
 Calibration of digital camera
Theoretical model of sensor distortion
This distortion is constant during the time c1
and depends only on the construction of
the sensor. It is defined through a
particular affine transformation

Dx 3  c1 (x  x 0 )  c2 (  0 )
D3  0
C2 (/C1)
We will see when using this correction, coupled with
the corrections due to the radial and tangential
distortion
Modified Collinearity equation
r11 ( X  X 0   r21 (Y Y 0   r31 (Z  Z 0  Z
x  x 0  Dx1  Dx 2  Dx 3  c  x 0  Dx1  Dx 2  Dx 3  c x
r13 ( X  X 0   r23 (Y Y 0   r33 (Z  Z 0  N
r12 ( X  X 0   r22 (Y Y 0   r32 (Z  Z 0  Zy
  0  D1  D 2  c  0  D1  D 2  c
(  ( 
r13 X  X 0  r23 Y Y 0  r33 Z  Z 0(  N
 Digital camera classification

Metric camera
The interior orientation is known and constant during the time.
The optical lens have an high quality (similar to the traditional
metric cameras). Calibration is realized in an optical bench and
by specific parameters. Tangential and similar distortions are
not considered. Often radial distortions are also corrected “a
priori”, generating images that are substantially free of
distortion.

Semi-metric Camera The interior orientation is known, but

variable (t). Given the stability of the sensor, these parameters
depend only on the focus (with discrete steps): in the case of
constant focusing, they are constant (t). The lens are similar to
traditional cameras and the calibration are realized on an
optical bench. Radial distortions and sensor distortions are
always defined.
 Digital camera classification
Professional Camera
They have an interior orientation that is unknown and variable
during the time. Given the stability of the sensor, these
parameters depend only on continuous focusing: in the case of
constant focusing they are constant over time. Using analytical
techniques so called "autocalibration“, it is possible to estimate
the inner orientation parameters, which can be considered
characteristic of the camera for a defined focal length. Radial,
tangential and sensor-related distortions are used.

Mass Market camera

They have an inner orientation that is unknown and variable.
They have a system that retract the optics in a protected position
when the camera is turned off. This movement modifies the
optical definition and inner orientation parameters and the
distortions changes every “switch on”. If the camera does not
switch off and the focus is not changed, these parameters can
be considered constant: it is possible to estimate them using
autocalibration.
Radial, tangential and sensor-related distortions are used.
 Aerial digital camera
General properties
The geometry is quite different (principal distance, sensor
size, et.) therefore it is better to consider the GSD (Ground
Sample Distance)
Some sensors are able to work in INFRARED band
Image shifting is continously corrected TDI (Time Delay and
Integration)

Frame
RECTANGURAL MULTIPLE SENSORS
DMC (Digital Modular Camera) di Z/I;
Ultracam D and X by Microsoft Vexcel;
DiMAC (Digital Modular Aerial Camera) by DiMAC System;
Pushbroom
LINEAR MULTIPLE SENSORS
ADS 40 (Aerial Digital Scanner, I and II generation) e ADS 80
by Leica Geosystem;
3-DAS-1 and 3DAS-2 (Digital Aerial Scanner) by Wehrli &
Associates/Geosystem;
JAS 150 by Jena Optronic AG.
 Aerial digital camera- GSD

Two lines of pixels (CCD)

Two lines of pixels (ground)

 Aerial digital camera
DMC by Z/I
4 panchromatic sensors (12
bit/pixel) with 4096 x 7168
pixel (12 μm) and focal length
= 120 mm . Image = 13824 x
7680, where the distortions
(all) are corrected

4 sensors (RGB and IR) (12 bit/pixel) with 2000 x 3000 pixel
(12 μm) and focal length = 28 mm
All information are merged (pansharpening) generating a
final image with a resolution = 13824 x 7680 (RGB and IR)
where all distortions are corrected
 Flight planning
l B Vocabulary

Am
A

L
(flat terrain)

B = air base l = image size

H = relative altitude (elevation) A = strips inter-distance (-> %sidelap)
L = size of the area (->%overlap) Am = area of the model
 Aerial digital camera
DMC by Z/I
Flight planning with a
relative altitude = 2000 m

H Sm GSD LT LL B (40 %LL) Parameters

[m] (1:) [m] [m] [m] [m] with
500 4167 0,05 691 384 154 different
1000 8333 0,10 1382 768 307 altitudes
2000 16667 0,20 2765 1536 614
3000 25000 0,30 4147 2304 922
4000 33333 0,40 5530 3072 1229
5000 41667 0,50 6912 3840 1536
6000 50000 0,60 8294 4608 1843
 Aerial digital camera
UltraCam by Vexcel/Microsoft
4 panchromatic sensors
4992 x 3328 (9 μm) and a
focal length = 100 mm. Final
image 14430 x 9420 where
all distortions are corrected.
 Aerial digital camera
UltraCam by Vexcel/Microsoft

Flight geometry
at 2000 m

H Sm GSD LT LL B (40 %LL)

[m]\ (1:) [m] [m] [m] [m]
500 5000 0,04 519 339 136
1000 10000 0,09 1035 675 270
2000 20000 0,18 2070 1350 540
Geometry
can change
3000 30000 0,27 3105 2025 810
with
4000 40000 0,36 4140 2700 1080 altitude
5000 50000 0,45 5175 3375 1350
6000 60000 0,54 6210 4050 1620
 Aerial digital camera
ADS40 by Leica Geosystems
Flight direction

The analytical
model has to be
corrected!

The system is composed by 2 CCD

sensors, with 12000 pixel (6.5 mm)
globally 24000 pixel,( 3.25 mm)
There are 4 sensors (RGB and IR)
for each line nadir and backward,
and one in foreward line
 Aerial digital camera
ADS40 by Leica Geosystems

16° backward
26° foreward

H Sm GSD LT
[m] (1:) [m] [m]
500 7966 0,05 621
1000 15931 0,10 1243
2000 31862 0,21 2485
3000 47794 0,31 3728
4000 63725 0,41 4971
5000 79656 0,52 6213
6000 95587 0,62 7456
 Semi-metric digital camera

ROLLEI d30 Metric

One of the first calibrated digital

camera.

sensor: 2552 x 1920 pixel

size 2/3”
Focal length: 10-30 mm (such as a 40-
120 mm for camera standard)
Focus with 2 fixed calibrated steps
 Semi-metric digital camera

ROLLEI 6008 Phase One

sensors: 5440*4080 pixel

48.96 * 36.72 mm
Focal length: 40 mm
Pixel size: 9 μm
 Professional digital camera

Kodak DCS pro 14n Canon EOS 5D

CMOS with 13.89 MPx(4560 x 3048), di CMOS with 12.8 Mpx (4368x2912), with a
sensor size 24 x 36 mm. sensor size =24 x 36 mm.
Interchangeable lens Interchangeable lens
 Professional digital camera

Sony Nex 5K Panasonic DMC-GF3

Sensor Live MOS

CMOS (APS-C) 12.1 Mpx
14.2 Mpx (4592 x 3056) Size 17.3 x 13 mm.
Size 23,5 x 15,6 mm Interchangeable lens
Interchangeable lens
 Mass market digital camera
Ricoh GX200
CCD with 12 Mpixel (4000x3000)
size 1/1.7”
Focal length 5,1 - 15,3 mm
(equivalent focal length at 35 mm: from 24 to 72 mm).

Canon IXUS 980 IS

CCD with 14.7 Mpixel (4416 x 3312)
Size 1/1.7”
Focal length = 6,1 - 29.0 mm
(equivalent focal length at 35 mm: from 36 to 133 mm).

Nokia N86 8M
CMOS with 8 Mpixel (3280 x 2464)
Focal length = 4,61 mm
 Exterior orientation
Summary

Exterior orientation using one frame only

Exterior orientation using two frames and one step

Exterior orientation using two frames and two steps
Relative orientation

Absolute orientation
 Exterior orientation
Plotting
The plotting requires that the geometry of the two frames is
reconstructed by means an analytical projection. The problem is
easier if you know the parameters of inner and exterior orientation of
the frames

P2

P1

Frame 1 (Sx) Frame 2 (Dx)

 Exterior orientation
The unknown are the exterior orientation parameters:
Analytical procedures
O1 B O2 

(X0, Y0, Z0, w, f, k)j j=1…num. fot, 1 2

c x
x1 2

There are several approaches:

P2

Independent orientation with 1 frame Z

P3
(j=1) P4
Y P1
O2 2 x2
Simultaneaus orientation with 2 X

O1 B O2 
frames and 1 step (j=1-2) b
c
1

x1
2

x
2

O 1 1 x1
Simultaneaus orientation with 2 c
frames and 2 steps(j=1-2) : P2
P3
x P2
Z
P4

Relative orientation z
P1


P4
P3

y Y P1

X
Absolute orientation


Photogrammetric triangulation Z
P2

(Simultaneaus orientation with all P3

frames which compose the block, j=1-n) Y P1 P4

z
y
x
X
 Independent orientation with 1 frame
O1 B O2 
Unknown parameters are 6: 1 (X0, Y0, Z0, w, f, k)1
2
c
x1
r11 ( X  X 0   r21 (Y Y 0   r31 (Z  Z 0 
x  x0  c
r13 ( X  X 0   r23 (Y Y 0   r33 (Z  Z 0 
r12 ( X  X 0   r22 (Y Y 0   r32 (Z  Z 0 
  0  c
r13 ( X  X 0   r23 (Y Y 0   r33 (Z  Z 0  P2
Z
Some ground control points (points with knonw
P3
coordinates) (Xi, Yi, Zi), are defined in the image
coodinates(xi, i) P4
Y P1
For a generic point i, it is possible to write the X
following collinearity equations:
r11 ( X i  X 0   r21 (Yi Y 0   r31 (Z i  Z 0 
xi  x0  c
r13 ( X i  X 0   r23 (Yi Y 0   r33 (Z i  Z 0  xi = f(x0 , c, X0 , Y0,, Z0 , w , f , k , Xi , Yi,, Zi)
r12 ( X i  X 0   r22 (Yi Y 0   r32 (Z i  Z 0  i = f(0 , c, X0 , Y0,, Z0 , w , f , k , Xi , Yi,, Zi)
i   0  c
r13 ( X i  X 0   r23 (Yi Y 0   r33 (Z i  Z 0 
 Independent orientation with 1 frame
It is necessary to have at least 3 ground
control points (GCPs). It is possible to
xi = f(x0 , c, X0 , Y0,, Z0 , w , f , k , Xi , Yi,, Zi)
write 6 equations, with 6 unknown
(underlined parameters). Usually, 4-6 i = f(0 , c, X0 , Y0,, Z0 , w , f , k , Xi , Yi,, Zi)
GCPs are used, which are well distributed
in the image. O1 B O2 
1 2

Collinearity equation are not linear (in c x

x1 2
the unknowns) and therefore a
linearization around approximate exterior
orientation parameters is required.
The approximate values can be
P2
obtained: Z
- from the flight planning (aerial image) P3
- project of shooting (terrestrial case) P4
Y P1
using a linear approximate analytical X
model (DLT)
55
 Independent orientation with 1 frame
Linearization of collinearity eq. (x)
r (X  X 0   r21 (Y Y 0   r31 (Z  Z0  X 0  X 0  dX 0 Y0  Y0  dY0 Z 0  Z 0  dZ 0
0 0 0
Z
x  x0  c 11  x0  c x
r13 (X  X 0   r23 (Y Y 0   r33 (Z  Z0  N w  w 0  dw j  j 0  dj k  k 0  dk
X  X 0  dX Y  Y 0  dY Z  Z 0  dZ
x
  2 (r13 Z x  r11 N   a2
c
X 0 x
  2 (Nr11  Z x r13   a8
N c
X N
x x
  2 (r23 Z x  r21 N   a3
c
  2 (Nr21  Z x r23   a9
c
Y0 N Y N
x x
  2 ( Nr31  Z x r33   a10
c
  2 (r33 Z x  r31 N   a4
c
Z 0 N Z N
x c  
  (Y  Y0 r33  (Z  Z 0 r23  x  (Y  Y0 r31  (Z  Z 0 r21   a5
Z
w N N 
x c 
 (Z x cos k  Z y senk  x  N cos k   a6
Z 
f N  N  tx  x  x 0
x c
  Z y  a7
k N
 tx  a2 dX 0  a3dY0  a4 dZ 0  a5 dw  a6 dj  a7 dk  a8 dX  a9 dY  a10 dZ  0

56
 Independent orientation with 1 frame
Linearization of collinearity eq. ()
X 0  X 0  dX 0 Y0  Y0  dY0 Z 0  Z 0  dZ 0
0 0 0
r12 ( X  X 0   r22 (Y Y 0   r32 (Z  Z 0  Zy
  0  c  0  c
r13 ( X  X 0   r23 (Y Y 0   r33 (Z  Z 0  N w  w 0  dw j  j 0  dj k  k 0  dk
X  X 0  dX Y  Y 0  dY Z  Z 0  dZ

  2 (r13 Z y  r12 N   b2
c

  2 (Nr12  Z y r13   b8
X 0 N c
X N

  2 (r23 Z y  r22 N   b3
c

  2 (Nr22  Z y r23   b9
c
Y0 N
Y N

  2 (r33 Z y  r32 N   b4 
c
  2 (Nr32  Z y r33   b10
c
Z 0 N Z N
 c  Zy 
  (Y  Y0 r33  (Z  Z 0 r23   (Y  Y0 r32  (Z  Z 0 r22   b5
w N N 
 c  
 (Z x cos k  Z y senk   Nsenk   b6
Zy
f N  N 
t    0
 c
  Z x  b7
k N
 t  b2 dX 0  b3dY0  b4 dZ 0  b5 dw  b6 dj  b7 dk  b8 dX  b9 dY  b10 dZ  0
 Independent orientation with 1 frame
This procedure, called pyramid vertex, O2 
O1 B
has the following disadvantages: 1 2
c x
1. The concept of ray intersection is x1
not considered
2. At least 3 GCPs are required, but
usually more GCPs are used,
especially on the board of the image
P2
A good solution is only obtained in the
Z
area described by the GCPs. The
extrapolation could be realized but only P3
with a limited area (5-10%).
P4
Y P1
X
 Independent orientation with 2 frames and 1 step

There are 12 unknowns: (X0, Y0, Z0, w, f, k)j j=1,2 O1

1
B O2 
2
c x
GCPs and tie point are used. x1 2

For each GCP, it is possible to write 4 collinearity

equations (2 for each frame)

P2
Z

r11 j (X i  X 0 j   r21 j (Yi Y 0 j   r31 j (Z i  Z 0 j  P3

x ij  x 0  c
r13 j (X i  X 0 j   r23 j (Yi Y 0 j   r33 j (Z i  Z 0 j  P4

(  (  ( 
Y P1
r12 j X i  X 0 j  r22 j Yi Y 0 j  r32 j Z i  Z 0 j X
ij   0  c
r13 j (X i  X0 j   r (Y Y   r (Z
23 j i 0j 33 j i  Z0 j 
x i1  f (x 0 , c, X 01 , Y01 , Z 01 , ω1 , j1 , k 1 , X i , Yi , Z i )
i1  f ( 0 , c, X 01 , Y01 , Z 01 , ω1 , j1 , k 1 , X i , Yi , Z i )
x i 2  f (x 0 , c, X 02 , Y02 , Z 02 , ω 2 , j 2 , k 2 , X i , Yi , Z i )
i 2  f ( 0 , c, X 02 , Y02 , Z 02 , ω 2 , j 2 , k 2 , X i , Yi , Z i ) 59
 Independent orientation with 2 frames and 1 step
It is possible to use some tie points, which are
points without known coordinates, but well
detectable in the image.
O1 B O2 
1 2
For each TIE POINT, it is possible to define the c x
unknown obkect coordinates (Xi, Yi, Zi), and to x1
measure the image coordinates (xij, ij) j=1,2
For each tie point, it is possible to write 4
collinearity equations, adding 3 new unknowns
(cyan boxes): P2
Z

x i1  f (x 0 , c, X 01 , Y01 , Z 01 , ω1 , j1 , k 1 , X i , Yi , Z i ) P3
i1  f ( 0 , c, X 01 , Y01 , Z 01 , ω1 , j1 , k 1 , X i , Yi , Z i ) P4
Y P1
x i 2  f (x 0 , c, X 02 , Y02 , Z 02 , ω 2 , j 2 , k 2 , X i , Yi , Z i ) X
i 2  f ( 0 , c, X 02 , Y02 , Z 02 , ω 2 , j 2 , k 2 , X i , Yi , Z i )
1 GCP  4 equations, 0 new unknown
1TIE POINT  4 equations, 3 new unknowns
 Independent orientation with 2 frames and 1 step

Tie points are an additional constrain, which allow to improve

the robustness of the system, on statistical point of view
Linearization is required (as before)
The solution is estimated with a single step
This approach is the fundamental for the bundle block
adjustment, adopted in the photogrammetric triangulation.
This approach estimates good solution only in internal area of
the GCPs. Extrapolation is possible, but only in limited region.
 Independent orientation with 2 frames and 2 steps
Scale of the model

O1 b O2 O1 b/m O2
1/m

Model in scale 1:m

Model in scale 1:1

Independent orientation with 2 frames and 2 steps

1 –relative
orientation

model

2 –absolute
Z orientation

terrain
X
O (0,0,0)
 Independent orientation with 2 frames and 2 steps
The aim is create a model, allowing the
intersection between the infinite couples of rays.

It is possible to realize this condition:

- Blocking 1 frame and moving the frame 2
There are 6 DOF (degree of freedom):
3 translations bx, by, bz
3 rotations w2, j2, k2 O bx O2
1
- But the value bx defines only the scale of the model

k2
j2

Frame 1

w2
bz O Relative orientation is
bx by2 based on 5 parameters
(unknowns) :
by, bz, w2, j2, k2
 Independent orientation with 2 frames and 2 steps
asymmetric
k2
j2

Frame 1

w2
bz O2
bx by

The model coordinate system is coincident with the coordinate system in the
frame 1
 Independent orientation with 2 frames and 2 steps
symmetric
The center are blocked, only rotations are allowed.

Coordinates system has origin in O1, and x axis which passes in O2 …

k1 z
k2
Frame 1 j2
j1
y

w1=0
x
O1 O2 w2
The relative orientation depends on
(angular parameters):
j1, k1, w2, j2, k2
 Independent orientation with 2 frames and 2 steps
Analytical solution
z Two straight lines have intersection only if
they have the complanarity conditions
This condition could be described as this
y multiplication
O1 xb O2 b, p , p
1i 2i

p2i
p1i
b
^ p2i x p1i = 0
i = 1 … ≥5
Pi

c 2 x11  12  x 2 2  22 

p   db y  db z  df1   c  dw1  x1dk1  df 2   c  dw2  x 2 dk 2
h h c  c c  c 
67
 L’orientamento di due fotogrammi
Asymmetric relative orientation
Starting from the collinearity eq., it is possible to define the parallax equation of height.

c 2 x 2 2  22 
p   db y  db z  df 2   c  dw2  x 2 dk 2
h h c  c 
With p = 0 and writing at least 5 of these equations for 5 homologous points, it is
possible to estimates (by, bz, w2, f2, k2)

Symmetric relative orientation

x11 x 2 2  22 
p   x1dk1  x 2 dk 2  df1  df 2   c  dw2
c c  c 
With p = 0 and writing at least 5 of these equations for 5 homologous points, it is
possible to estimates (f1, k1, w2, f2, k2

If the equations are more than 5, the solution is estimated by LS. The residuals are
the «residual parallaxes», which describe the quality of the orientation. In the
relative orientation, GCPs are not required.
68
 Independent orientation with 2 frames and 2 steps
Absolute orientation
The relationship between model x, y, z and object X, Y, Z ,
can be described as (isotropic scale factor):
O2 2 x2

 X   Xu   x
b
     
O 1 1 x1  Y    Yu   m  R   y
Z Z  z
c
P3
   u  
P2 x
P4

z
 Where Xu, Yu, Zu are the object coordinates of
P1
 y the origin in the image coordinate system (x,y,z)
m scale factor
 R rotation matrix with ,,
P2
Z
P3
These seven parameters are called absolute
Y P1 P4 orientation parameters.
X
 Independent orientation with 2 frames and 2 steps
In order to estimate these seven
parameters, it is possible to write:
Analytical solution
• three equations for each 3D
points
 X   Xu   x
• two equations for each 2D      
point (horizontal components)
 Y    Yu   m  R   y
Z Z  z
• one equation for each
   u  
altimetric point  X   Xu   x
     
It is fundamental that the points  Y    Yu   m  R   y
are not aligned. Z Z  z
   u  
 X   Xu   x
Some combinations:      
 Y    Yu   m  R   y
2 (3DP) + 1V Z Z  z
   u  
2 (2DP) + 1V
….
 Independent orientation with 2 frames and 2 steps
Analytical solution
 X   X u   x
 Y   Y  m R  y
 Z   u   
   Zu  z
X  x = X0   = d,  = d,  = d, m = 1 + dm, Xu = dXu

 cos f cos k  cos fsenk senf 

 cos wsenk  senwsenf cos k cos w cos k  senwsenfsenk  senw cos f 
 
 senwsenk  cos wsenf cos k senw cos k  cos wsenfsenk cos w cos f 

 1  d d  1 + 𝑑𝑚 −𝑑𝐾 = −𝑑𝐾 − 𝑑𝐾𝑑𝑚~ − 𝑑𝐾

 
dR   d 1  d 
  d d 1 

1  dm  d d   dm  d d 
   
m  R  (1  dm dR   d 1  dm  d   I   d dm  d 
  d d 1  dm    d d dm 
 
Independent orientation with 2 frames and 2 steps
Analytical solution

X = dXu + X0dm + Z0d - Y0d + X0

Y= dYu + Y0dm - Z0d + X0d + Y0

Z= dZu + Z0dm + Y0d - X0d + Z0

 Flight planning
l B Vocabulary

Am
A

L
(flat terrain)

B = air base l = image size

H = relative altitude (elevation) A = strips inter-distance (-> %sidelap)
L = size of the area (->%overlap) Am = area of the model
 Photogrammetric triangulation
Photogrammetric triangulation allows to be “free”
1 3 8
from the need to have at least three GCPs for
each stereoscopic model, to be detected with 5 7 9
Strip 1 11 12 13 14 15 16
terrestrial or satellite methods. 4

This technique allows: 6 10

2
- the simultaneous determination of the exterior
orientation parameters of a block of frames; Strip 2 21 22 23 24 25 26

- the possible determination of unknown inner

orientation parameters (self-calibration in the
case of using the projective star model)
Strip 3 31 32 33 34 35 36
- the determination of the coordinates of a
series of points. These points can be, natural or
artificial points.
Punto d'appoggio completo 23 Punto nadirale con
n° del fotogramma
Punto d'appoggio altimetrico
Punto incognito o di legame
 Photogrammetric triangulation
Photogrammetric triangulation can be made using two different approaches:

1) Bundle block adjustment based on the direct use

of the equations of collinearity: this method allows to
obtain the maximum possible precision and is the z
y

only one that allows to use frames of different nature x

(metric, semi-metric and non-metric) allowing the

determination of the parameters of interior and
exterior orientation unknown ;
2) Independent model, which is based on the
analytical absolute orientation, considering all stereo
models which compose the block, already realized.
Punti di appoggio
altimetrici
Punti di appoggio
completi
Punti di legame

In this way, it is possible to reduce the number of Ground control points