"ground control" pane - "view estimated" button

[jan]

What is the purpose of the "view estimated" button in the "ground control" pane.

I want to use it to retrieve coordinates of some objects by putting flags (markers) on these objects in a georeferenced model in agisoft.

Am I getting precise object coordinates with this method?

regards,
Jan

[Alexey Pasumansky]

Hello Jan,

Estimated values gives you the coordinates of cameras/markers (or scale bar length) that were calculated by PhotoScan. If you want to get coordinates of any point on the model you need to create the marker instance and specify its projections on every photo the point of interest is visible on. The estimated coordinates for this marker will give you the coordinates of the point.
The accuracy will depend on the alignment accuracy and the overall georeferencing accuracy for the model.

[jan]

Hello Alexey,

thank you, this is the meaning I was assuming.

I though have very bad results using this feature (see below):

I have made 2 models of the same area (about 250 m long x 125 m wide), but with pictures taken at different times. Model_1 is based on pictures taken in the year 2010, and Model_2 is based on pictures taken in 2012.

There are 4 aerial targets visible on the 2012 pictures, about 50 to 100 meter distance from each other. Their coordinates are known precisely (as measured with a total station).

There were however no well known aerial targets on the 2010 pictures. Therefore, I looked for 8 objects which are visible on both (2010 and 2012) models, and which didn't move during this period of time. I estimated the coordinates of these 8 objects by placing 8 markers on these objects in the 2012 Model. Via "View Estimated" in the Ground Control panel, I could then retrieve these estimated coordinates. I then used these 8 coordinates as source coordinates on the corresponding marked objects in the 2010 model. After this, I optimized the photo alignment, hereby hoping to make the model of 2010 also georeferenced, and congruent with the 2012 model. However, the result is unacceptable, with errors of more than 10 meters. It is much worse than before the optimization.

What could be the reason for this bad result? 

kind regards,
Jan

[Alexey Pasumansky]

Hello Jan,

The behavior of the errors after optimization is strange. Providing that the objects remain on their places and markers projections were defined correctly there should not be such problems.

What is the marker accuracy parameter used in Ground Control pane settings dialog? What version/build are you using?
Could you please try installing the latest 0.9.1 build (if you are using this version) or try to optimize the alignment with 1mm marker accuracy.

[jan]

Hi Alexey,

I'm using 0.9.0 build 1586 (64 bit). The marker accuracy parameter was 0, I've set it to 0.001 meter as you suggested. This seems to give much better results as the model is more or less flat again after the optimization (in contrast with before). There are however still large errors according to the info in the ground control pane (1 error even above 10 meters), but this isn't obvious in the point cloud at first sight (strange). I will let you know when the geometry is built, if the result is satisfactory or not.

many thanks!
Jan 

[jan]

Hello Alexey,

I'm sorry, I made a mistake. Of the 8 objects visible on both models, I accidentally misplaced one marker on the model from which I estimated the coordinates (model 2012). This will probably be the reason for the bad optimization in model 2010.

A new question however arises from this:

Is it ok to again (but now with the right marker coordinates) optimize starting from the point cloud which was badly optimized before? Or is it necessary to start from scratch (i.e. start a new alignment from photos)?
In other words, does the quality of the optimization depend on the quality (i.e the deviation from the "real coordinates") of the current point cloud to be optimized?

thanks,
Jan

[Alexey Pasumansky]

Hello Jan,

That depends on the overall accuracy after the optimization with the incorrect data.
In case errors on markers, reprojection error are not high and sparse point cloud is not broken - you can try to optimize even this data and check if the marker errors are reduced.