Using camera external orientation

[patcarbon]

Hi, sorry if some of these questions are obvious but I'm new with photoscan.

So I want to use a photo dataset generated by a UAV (usually between 100-300 pics).  The UAV produces camera position and orientation so the obvious thing to do is to use those instead of ground control points.  A few things that weren't clear from the manual:

1- What is the naming convention for the XML files?  How can I produce these in a way that photoscan will easily import a few hundred and correctly link each file to the correct photo?
2-How does photoscan handle error in the external orientation?  Is there a least squares adjustment?  Does it warp the DTM to fit external orientation (EO) or does it adjust the EO and preserve the DTM?
3-Where in the workflow should I import the EO? Before or after the alignment?

Many thanks
Patrice

[Alexey Pasumansky]

Hello Patrice,

1) It will be easier if you use tab-separated .txt file to import camera positions in the following format:
       image_file_name (with extension)    x-coordinate    y-coordinate    height

2) PhotoScan uses least squares adjustment. During adjustment the reconstructed scene data including model geometry and camera positions is transformed so that the RMS error between estimated and measured camera positions is minimized. The transformation used is a 7 parameter similarity transform (translation + rotation + scaling).

3) Setting the ground control has to be done before aligning photos stage if you wish to select "Ground control" option as pair preselection parameter in Align Photos dialog. But anyway it is necessary to "Update" Ground Control after aligning photos is done.

[Oli63]

Does it bring any advantage in the aligning process (speed or quality) if I add more data in the ground control than x, y and height and if so, how is it done? I assume, I need to define the variables in the first line of the text file, separated with tabs.
In our case, the UAV would give at least turning rates and attitude of the UAV in all three dimensions.

[Alexey Pasumansky]

Hello Oliver,

For large aerial data sets we recommend to use Ground Control pre-selection since it uses only a group of neighbouring photos in matching process, therefore significantly decreasing processing time.

The input format could be customized in version 0.8.5 (in respect of column separator, order and numbers of columns) if using "Character separated values" file-type when loading data into Ground Control pane.
It is possible to import X,Y,Z data and also Yaw, Pitch, Roll, but the latter three values are not obligatory.

[Oli63]

Thank you Alexey for the quick response.
With other words: X,Y and Z only help in finding the right matches of neighbouring photos in the alignment process but nothing more? Do these informations NOT help in improving or speeding up the geometry process? Aren't roll, pitch and yaw useless in this case?

[Alexey Pasumansky]

Hello Oliver,

Source camera coordinates are also used in optimization process (some times it is recommended to perform optimization only based on cameras before involving markers in process - usually it is required if the point cloud model has great "bowl-effect").
And during build geometry stage only estimated camera positions are used.

Source yaw, pitch, roll data is not used yet.