Hello all,
thanks a lot for your suggestions.
I have checked and indeed there are almost no tie points on the overlap area - not because of low overlap, but because all such tie points (except maybe 10 or so) were filtered out during the point cloud optimization. So that overlapping photos from two different flights have usually ~2000 matches with each other, but all marked as invalid.
I guess it makes sense that a tie point between two photos at different altitudes (and with some time delay) could have worse quality than between two consecutive photos in a same flight, so these get all filtered out.
With a less aggressive optimization, many more tie points are left in the overlap areas (I will have the DEM ready tomorrow to check the result).
So now I am wondering how this compares to the processing Paulo suggested: for an accurate relative alignment, is it better to place 3 markers in the overlap area, or let Metashape use its own tie points? I am not sure that I can manually place the markers with an RMS < 0.3 px (as is the accuracy of the sparse cloud after the more gentle optimization).
To share the data set, I have to ask permission, I'll let you know.
Thanks a lot!