the influence of 'projection accuracy'

[jamesm]

I have a project of ~400 UAV photos that uses ~270 GCP for georeferencing (only) – i.e. no GPS camera position info etc. I have read in earlier posts that the “Projection accuracy” parameter is only supposed to have an influence if GCP are active in the bundle adjustment. However, I am seeing changes even when GCP are not active.

With the cameras initially aligned:

Set projection accuracy to 0.1, uncheck all GCP, optimise, recheck all GCP, update :  total error on GCP = 0.50 m

Then:

Set projection accuracy to 1.0, uncheck all GCP, optimise, recheck all GCP, update :  total error on GCP = 0.33 m

The different final error values indicate that the project shape must have changed – which is confirmed by visualising the differences. Similar variation is seen whether or not camera parameters are free in the ‘optimise’ step. Can anybody shed any light on this? Are the markers and their observations (projections) still being used in the bundle adjustment, just with them not linked or constrained to their object space coordinates? Of course, we will eventually be using the GCP in the bundle adjustment, but I'd like to understand what is going on here too.

Many thanks,
Mike

[jamesm]

Dear Alexey,
Any chance of a comment from you on this? Much appreciated.
All the best,
Mike

[FLORIANO]

I am interested in this question too.
Could someone answer us, please?
Thanks a lot.

[Paulo]

I thnk it would be great to have in the documentation more explanation on the various parameters in Reference settings and their influence on final alignent adjustment.

For example in pre release 1.2 we now have 2 accuracy image measuremnt parameters:

- marker accuracy default at 0.1 pix
- tie point accuracy default at 4 pix.

If I understand this correctly than it means than in adjustment image marker measurement  is considred 40 times more accurate than tie point matching measurement...

Why would this be so? I would think computer image matching to be more precise than manual image marker measurement....

Any enlightment is welcome!

 

[mauroB]

Dear all,

It has been long time from the intiial post.

Any news? Based on my personal tests, I think that the markers (even if unchecked during the optimization step) actually work as tie points, by affecting the BA results according to the projection accuracy value.

Kind Regards
MB