Force alignment between certain images during optimization

[vik748]

Hi,
I am trying to run an alignment on a set of 4000 images taken while going around a large ice-berg. If I leave the project to align automatically, it fails miserably.  However if I start from one end I am able to align and get a general shape of the path.  The challenge is that when the loop is closed the overlapping images don't seem to register that they are looking at the same thing and make a separate path.
I have been able to minimize this effect by starting from the overlapping area, which lets both runs of that area align well.  Then sequentially aligning from both sides till they meet at the opposite end.  However, there is still a small mis-alignment at the end which I am unable to get rid off.  Here is what I have tried:
1. Multiple runs of optimize cameras while changing / holding different parameters.
2. Filtering the point cloud to remove 'bad' points or those with high uncertainty.

Please see attached screenshots for an idea on what is going on.

So, in the area where I have mis-alignment, the images are taken right next to each other and have considerable overlap.  Is there a way to force Metashape to keep them correctly aligned and adjust the others accordingly?

Thanks,
Vik

[Mak11]

Vik,

1) Did you try to optimize camera alignement afterwards (Tools->Optimize cameras)
2) Did you try to run the alignment with Generic Preselection OFF (alignment will take longer to compute but you often get better results)
3) An iceberg being mostly uniform in terms of features (colors, shape etc) is the not best case scenario for alignment.

Mak

[vik748]

Mark, answers to your questions:
1. Yes multiple times
2. Yes, this is how I can ensure I get cross matches between the images in the beginning and the end.  Generic pre-selection is able to pick it up though.
3. I use adaptive histogram equilization which enhances features, so the matching is actually quite good.  Here is the matching between the two consecutive images which are being mis-aligned.

[Mak11]

Vik,

Alexey from Agisoft will probably give you more answers (and help than me) once his sees this thread later today, but their are more Red matching lines that Blue ones which isn't good.

To quote him from an earlier thread:

Blue lines are connecting valid (used) matching points between the selected pair of images, and red lines are connecting invalid (unused) matching points.

The reason of the very few number of matching points detected could be related to the processing settings used or to the quality of the images (high noise, image blur and etc.)

Mak