Agisoft Metashape

Agisoft Metashape => General => Topic started by: andyroo on April 14, 2014, 09:34:38 PM

Title: Broken Model - interesting results - build 1847
Post by: andyroo on April 14, 2014, 09:34:38 PM
I just processed my second flight where I deployed two cameras. Canon D10 is shooting 6.2mm (28mm at 35mm equivalent) and a canon EOS M shooting at 40mm. There's a screenshot of the broken alignment attached.

I also upgraded to build 1847 just to compound the confusion. Loaded camera models for both cameras from my previous successful surface model construction and got an interesting result - the whole model aligned, but there is a break in the ground plane and the terrain shoots off into the sky at about a 30° angle.

I would understand this better if there was an overlap issue with my 28mm photos, but the plane flies at a constant elevation above ground and gives me a good (about 80%) along-track overlap, and the flight lines give me about 70% sidelap. So the only thing I can figure is that the addition of the 40mm photos (0 overlap, random sidelap) screwed up the surface reconstruction. Although it worked fine last time. I'll test this theory tonight, but I've done about 40 flights on this same area, and haven't had this happen before, so it seems like it's either the additional camera or the latest build.

I think I can probably fix it by adding all of the GCPs but that will probably be a pain with the offset terrain. I think it will probably also fix if I re-run the alignment without the EOS M imagery, but I don't know a way to then run a second alignment to only add the EOS M imagery. I am wondering if it might also make a difference in which order I load the images into the project. Looking for any advice.

Also just FYI, My goal w/ the 40mm imagery is twofold. First I am testing the camera to try to get it taking pictures reliably and rapidly so I can replace the 28mm with it. Right now I can only maintain focus if I stay in autofocus mode because the temp change from the ground apparently messes w the pre-focus to infinity. Unfortunately with autofocus it only takes pics about every 5 seconds which is too slow. Second purpose is to  increase ground resolution. Even though I don't have good overlap, that's provided by the D10, and the EOS M effectively covers about 80-90% of the survey area (based on my guesses from registering about 150 GCPs on every image). It worked great on the last flight, but This one seems to be an issue.
Title: Re: Broken Model - interesting results - build 1847
Post by: Wishgranter on April 14, 2014, 11:23:29 PM
from the photo it seems that you fly over trees, where the break happen.... trees a re problematic sometimes..... try  sort the photos so have clean AND proper overlap without too much photos near other and as i see it the photos are not precisely separated.. problem with sync ?? that mean between some of them could be bigger distance - align problems...   
Title: Re: Broken Model - interesting results - build 1847
Post by: andyroo on April 15, 2014, 01:17:57 AM
Hi Wishgranter.

Thanks for the quick reply. I was able to reconstruct the imagery without problems by removing the higher-resolution camera pictures from the EOS M. I think I will be able to reconstruct the whole model without problems just by removing a few EOS M pictures in the problem area. I think it is interesting to see this problem happen, and I think you are right about the trees playing a role. I guess probably the best way to fix this is to add GCPs and georeference the model with just the D10 imagery, then to add the EOS M images and re-align everything. I wonder if I will need to somehow make the camera position and orientation for the D10 imagery stay fixed, or if the GCPs will be enough?

I agree that trees are sometimes a problem. At this location I often lose images from the outer flight lines, but the inner ones hold together pretty well.

My images are evenly separated. The reason they didn't look that way in the other image was because of the angle break, and because one camera is firing slower than the other and they are on different wings of the plane. If I only showed one camera at a time, the images would look more evenly spaced (like in the attached image).