good error point map and a bad dem

[lucia.and...@gmail.com]

Hello, 

I was wondering if anyone could help me with this problem:

I am working on creating a DEM from historical images. I have generated match points from stereo triangulation, for the filtering, I have set --rm-threshold 0.9 and --rm-min-matches 85, hoping to remove bad matches. The error point map after the first pass looks good (with a range of 0-0.5). However, the elevation model shows some misalignment.

The second pass of BA does the reverse, creating DEM with seemingly good alignment to the reference but the error is very high, at some places. 

Would anyone have any idea what the source of this could be and how to solve it? I suspected bad matches, but the matches look alright and different BA options also did not help. 

Many thanks, 

Lucia

[Oleg Alexandrov]

You can try to see where your newly produced DEM disagrees with reference, and also inspect the intersection error image (triangulation error) produced with point2dem. If both of these are more problematic at the boundary, that may mean issues with intrinsics, which is focal length, optical center, and distortion. 

If the matches are reasonably well distributed, likely they are not the issue. 

--
You received this message because you are subscribed to the Google Groups "Ames Stereo Pipeline Support" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ames-stereo-pipeline...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/be7a1cfc-1974-47ee-96a8-ac3135b8e54fn%40googlegroups.com.

[Oleg Alexandrov]

To add, historical images usually have unreliable intrinsics. If that's the case, we have quite some reading on how to fix that at https://stereopipeline.readthedocs.io/en/latest/bundle_adjustment.html#refining-the-intrinsics-per-sensor (and higher in the same document). This can be detailed work but was shown to work well (unless optical bar cameras are used).

[lucia.and...@gmail.com]

Thank you, 

I have tried but none of the techniques helped. I had to start from the beginning which seemed to improve the results.

Thanks you!

Lucia

[Oleg Alexandrov]

Going through techniques in quick succession is likely going to not yield quick results, indeed.

Historical images can be tricky. I hope you don't deal with the panoramic ones, which are long, and then very wide at end points. Even for frame cameras, high focal length makes the problem poorly behaved. Need to have a good convergence angle too and good illumination.

In short, such work requires quite a bit of pondering, examination, and understanding of the specific setup. Some folks have a full-time job just dealing with such things. :)

If still stuck, you're welcome to write to me offline with more specifics.

To view this discussion on the web visit https://groups.google.com/d/msgid/ames-stereo-pipeline-support/02c08720-8ab2-487d-b5a5-cf90aab052c3n%40googlegroups.com.