Unmanned Aircraft Systems

Data Post-Processing

Section 2 – MicaSense 5-band MultiSpectral Imagery

USGS National UAS Project Office – March 2016

Synopsis
In this introductory training class, we will explore how to handle image data captured from an
unmanned aerial vehicle equipped with an on-board camera or sensor. Utilizing Computer Vision –
Structure-from-Motion techniques that estimates three-dimensional information from two-
dimensional images. Using real world data captured from a UAS, we will illustrate how it is possible
to generate georeferenced point clouds, elevation models and mosaiced image bases for mapping
and geographic information system data layer creation.

Requirements
- A registered version of Agisoft PhotoScan Version 1.2.3
- Access the data files noted below
- No previous experience with PhotoScan is necessary

Workflow
The following step-by-step instructions are intended to familiarize participants with the relevant
components of PhotoScan. A short description is given, followed by a specific “cookbook” of
instructions for how to process a dataset from beginning to end.

Data
Three separate real world datasets are provided for Exercises to see how actual collected data is
processed into workable GIS data layers.

Class Outline
- Import images collected from a UAS
- Align the images and remove or adjust errors to create a sparse point cloud
- Create a dense point cloud
- Create a mesh or digital surface model
- Create image texture
- Output the products for use in GIS

2
MicaSense RedEdge 5-band image captures to mosaic and rectify using Agisoft PhotoScan
Agisoft PhotoScan Version 1.2.3 (Build 2331 – 64bit)

1.) Adding Photos

Procedure Description: Images are loaded to begin the mosaic and rectification process. Images that
contain GPS embedded coordinate data available directly from the camera or captured from the UAS,
allows for initial referencing of the images to the ground. Images can be from different flights, altitudes
and folders with standard image formats supported such as .jpg, .png, .tif, etc.

• Workflow… Add Photos … Select all the photos (each image capture should have 5 separate
files (.tif format)…Open.
• Select ‘Create multispectral cameras from files as bands’… OK

• Once photos are added, they will appear as a ‘Chunk’ in the Workspace panel with the number
of photos that were added. Thumbnails of the photos will appear in the Photos panel.

• [Optional] - In the ‘Workspace’ panel, right click on the ‘Chunk’… Set Master Channel…select
the master channel (or band) you prefer to use for the PhotoScan image correlation process,
or use the default.

3
Setting the Coordinate System for Image Referencing
• On the ‘Reference’ panel … Select the ‘Settings’ icon … check to make sure the coordinate
system is set to what the camera or UAS GPS was using while collecting the photos (i.e..
Geographic Coordinate System, WGS84). Note: Keep the accuracy settings to the default
values.

Checking the Camera Calibration

• Select Tools from the main menu…Camera Calibration
• Basic information is extracted from the EXIF (image header info) such as pixel size, focal length
and resolution

4
2.) Aligning Photos

Procedure Description: To begin the photo alignment process, three steps are performed to tie the
images together in an automated image correlation process in order to create a sparse point cloud. This
initial phase is also used to select the best images to use for the model.

Note: Bad images can initially be removed before running the align photo process by selecting them,
right clicking and removing the cameras. This can save considerable time in the alignment process.

• Workflow… Align Photos…

• Settings:
• Accuracy : highest = full image resolution lowest = down sampled resolution
• Pair preselection : use generic or disabled if image locations are not known
use reference if images have known reference info in EXIF or log file
• Advanced:
Key point limit (point of interest) = 60000
Tie point limit (pts matched on 2 or more photos) = 0

• PhotoScan processes three steps in order to create a sparse point cloud; detecting points,
selecting pairs of images, and matching points.

5
Eliminating bad photos
• Photos that did not align are shown graphically by the ‘dots’ and also in the Reference Panel with
no error measurements calculated. These photos can be removed by selecting with the
Rectangle Selection tool, or by highlighting in the Reference Panel…right click…remove
cameras…yes.

6
• Photos that don’t add value to the model – such as take off and landing times from the UAS - can
also be removed. The Rectangle Selection tool can be used to select…right click in the Reference
panel on the highlighted image…remove camera…yes.

• A realignment of the photos is necessary when images are removed. Workflow…Align

Photos…(use the same settings as before or increase accuracy to a higher level)…OK

7
3.) Place Markers (Adding Ground Control)

Procedure Description: Markers, or ground control points can be added to better improve the model and
georeferencing accuracy of the final data outputs.

• Will not be covered due to time constraint

8
Optimizing the Photo Alignment
Procedure Description: Optimizing allows for a higher accuracy in the camera parameters and corrects
distortions caused by the camera lens – often seen in a bowl shaped point cloud. This function is
performing a photogrammetric bundle adjustment.

Optimizing the Photo Alignment

On the ‘Reference’ panel … Select the ‘Settings’ icon … (use settings below if it does not
default to these values).

• Select the Optimize Cameras either from the ‘Tools’ tab on the ‘Main Menu’ or from the icon on
the ‘Reference’ panel Use the default values or check parameters as shown below…OK

f – camera focal length (x,y)

cx, cy – center of lens or principal point
Other values – radial distortion coefficients

• After Optimizing, check the ‘Console’ window and look for the RMSE. Ideally it should be
between .13 --> .17. It is also good to watch the Projection and Error columns in the
Reference Panel. The limit is to not go less than 200 on Projections and less than 1 meter
on the overall error.

9
4.) Gradual Selection

Procedure Description: The Gradual Selection procedure will be used several times in order to improve
the geometry of the overall model.

Reconstruction Uncertainty
• Select ‘Edit’ from the main menu…Gradual Selection…Reconstruction uncertainty (from the
pulldown menu)… the goal is to reach a Level = 10 although if too many points are selected this
may not be possible (below 50 is highly recommended)…OK.

• Selected points will show up as pink. To delete those points, select the from the main menu.
After deleting points, another optimization is needed . Select the optimize icon from the
Reference panel Use the same setting as before.

• The Reconstruction uncertainty procedure should be run 2 times.

10
Projection Accuracy
• Select ‘Edit’ from the main menu…Gradual Selection…Projection accuracy (from the pulldown
menu)… goal is to reach a Level = 2-3… OK.

• Selected points will show up as pink. To delete those points, select the from the main menu.
After deleting points, another optimization is needed . Select the optimize icon from the
Reference panel

• The Project Accuracy procedure should be run 2 times.

11
Reprojection Error
• Select ‘Edit’ from the main menu…Gradual Selection…Reprojection Error (from the pulldown
menu)… goal is to reach a Level = 10% of the points. Note: in the lower left-hand corner, try to
get close to 10% of the point selected (i.e.. 9,812 out of 99,092 selected to be deleted)…OK.

• Selected points will show up as pink. To delete those points, select the from the main menu.
After deleting points, another optimization is needed but, before doing so, the tie point accuracy
can now also be tightened.

• Select the icon from the Reference Panel, and enter the desired tie point accuracy value.
Use .1 if the images are very clear, .3 – 1.0 if they are not as crisp.

• Select the optimize icon from the Reference panel and check all the remaining distortion
parameters…OK.

• The Reprojection Error procedure only needs to be run 1 time.

12
5.) Build the Dense Point Cloud

Procedure Description: A dense point cloud can be now be derived from the better estimated camera
positions calculating several x,y,z points as well as assigning color values to accurately create the model.

• Workflow…Build Dense Cloud…(Note: Higher the quality the more intensive processing and time
needed to derive. Depth filtering set to Aggressive is used for the most complex detail.)…OK

• Select the icon on the main menu if the dense point cloud does not display (usually defaults
to a sparse point cloud display.

13
6.) Build Mesh

Procedure Description: From the dense point cloud, a polygon mesh model can be generated.

• Workflow…Build Mesh…OK
(Note: Surface type = Height Field for vertical photography, Arbitrary is used for oblique models)

• Select the icon on the main menu to display the mesh (shaded, solid or wireframe).

14
7.) Build Texture

Procedure Description: Texture in the form of image overlay can be generated to be able to inspect the
model before exporting the orthophoto mosaic.

• Workflow…Build Texture…OK
(Note: If ‘Enable color correction’ is used, the time to generate the texture may be increased)

• Select the icon on the main menu to display the orthophoto (texture).

15
8.) Build DEM

Procedure Description: A digital elevation model can be generated from the model into a desired
coordinate system and projection.

• PhotoScan will want you to save the project at this point. File…Save as…Filename.psx.
Workflow…Build DEM…select the coordinate system of the output (defaults to the model
setup)…OK. (Note: Source data can be either Dense Cloud or Mesh. Dense cloud is used for
better accuracy.)

• After the DEM is generated it can be displayed in the Ortho Display by double clicking the layer
in the Workspace Panel under the ‘Chunk’.

16
9.) Build Orthomosaic

Procedure Description: A digital orthomosaic can be generated from the model into a desired coordinate
system and projection.

• Workflow…Build Orthomosaic…OK
(Note: Reprojection of the image can be done at this point or during the export of the orhtoimage.)

• After the Orthomosaic is generated it can be displayed in the Ortho Display by double clicking
the layer in the Workspace Panel under the ‘Chunk’.

17
10.) Exporting Products

Procedure Description: The various products generated through the modeling process can all be exported
into standard formats for use in display or GIS data layers.

• From the Main Menu Select File…Export (Points, Model, Orthomosaic or DEM). Another option is
to export by right clicking on the layer under the ‘Chunk’ in the Workspace Panel.

• Exported layers can then be used in standard GIS software (i.e.. Global Mapper below)

Orthophot mosaic (5-band) Digital Elevation Model DEM with Contours