Alignment strategy for thousands of historic aerial photos

[lefsky]

I've created multiple orthomosaics using 1000's of historic aerial photos. In all cases, I've been unable to get the entire collection of photos to align at one time, even when I have the coordinates of photo centers.

The approach I've developed is to align different subsets of the photos. A simple example would be to align the northerly third of photos, the southern third of the photos, and a central section of the photos that overlaps the other two. I would then do the same horizontally. I then have 9 overlapping subsets that can be aligned to each using "Align Chunks" using the camera option.

In a post regarding memory management, Alexey indicated that the cameras should be aligned in a single chunk (which requires less memory than the build dense point cloud step) and then, if memory is an issue, split into multiple chunks to build dense clouds. Those point clouds can then be merged together with Merge Chunks.

I have two specific questions:

1. What are the drawbacks to the method I use (aligning multiple overlapping subsets of cameras)

2. Because each of the cameras is represented in more than one chunk, the time involved in generating dense clouds for each chunk (and then merging the chunks) exceeds that of first merging the chunks and then building the dense cloud from the merged chunks. Is there a reason to prefer the former approach (building dense clouds for each chunk and then merging)?

Thanks in advance for any help the community can provide.