Holes in point cloud but not orthomosaic

[sfrederickson]

HI there,

I am processing around 8000 scanned images and trying to make an orthomosaic, DEM, and point cloud.

After aligning the photos and creating the point cloud, there seems to be many holes in it. The orthomosaic is fine, but the point cloud is very patchy, especially when zoomed in. I'm attaching a screenshot to this post.

For the alignment, I used a 40,000 key point limit, and a 4,000-tie point limit. I could potentially restart the alignment, but it took 22 days to complete.

Any help would be appreciated!

Best,

Sam Frederickson

[jbind]

What is the surface material/texture in the missing areas? Especially if it's water, this isn't unexpected for the pointcloud since there are no points that can be reconstructed...
The orthophoto (and DEM) don't have holes there, since these areas will be interpolated into a (relatively) flat surface.
Re-alignment won't change that.

[sfrederickson]

Hey there,

The surface material/texture is land and pretty similar to its surroundings. I am re-running the dense cloud process so hopefully that will yield some better results.

These images were taken by a plane in the 60s and have some interesting flight paths, could that be part of the issue?

[jbind]

Ok, so it shouldn't be the surface texture, then. But if they're taken in a regular aerial survey in the 60s (when often, the quality of imagery was very good for bw images), the overlap might be too limited to process really good SfM results. The cameras were calibrated to a very high standard and overlap was usually much less than what you'd use for a UAV survey. We have successfully processed historical aerial imagery through Metashape, but you're dealing with very different source data. Also, you are dealing with digitised analogue imagery. There's a whole can of worms that can come through the digitilisation process... So, in summary, it could be multiple factors. Maybe somebody else can offer more advice - my experience with historical aerial photos is limited to 3-4 areas we reconstructed.