5
« on: May 24, 2016, 06:57:55 PM »
HI all,
I'm using Photoscan as part of a project to trace irrigation systems in parts of Spain. The coverage is very irregular - - we were using a quadcopter with shaky battery life over obstructed terrain. We also only have a few GCPs -- about 6 or 7 we can use, fairly well spread out. The GCPs were acquired via DGPS so those are pretty accurate, but if the final model is a few dozen cm or even a meter off in some places it's not the end of the world, since this is more for survey than mapping.
(photo coverage map attached--coverage area is about 0.25 sq. km. Yellow pins/circles are GCPs)
I'm doing this on a (reasonably good) laptop--an Acer ASPIRE V15 Nitro--16MB ram (going up to 32) and an NVIDIA GeForce GTX 960M.
Photos are 12 mpx, taken with a fisheye lens (may have been a GoPro but will have to check)
HEre are my questions:
1.) Is it even feasible to run a project of this size (3500 images) on my laptop without breaking it into chunks? Say using ground control pair preselection, medium resolution point cloud/models?
If Not, and I need to break it into chunks:
2.) Can different chunks geographically overlap? We sometimes re-flew the same area to improve coverage, and it's difficult to really break this up where each chunk is a different region.
3.) Since there are only a few GCPs, and they are very spread out, there's no way to have small chunks (say 300 images or less) have more than 1 or 2 of them. Can ground-control based pair preselection still be used even with only a couple of markers, to then be merged later?
4.) Somehow, when I ran the first step (alignment) on all the images at lowest quality, it worked ok (about half aligned, which is about what I expected), but when I ran it again on medium alignment, it failed--only aligning 5 out of 3500 photos. ANy reason why this might be the case?
If anyone has thoughts on these Q's, or ideas for workflow, I'd really appreciate it!
--Steve