I'm trying to create a 3d model of a site in Australia. After running the AT i get a bunch of photos that just go off on a different plane and a corresponding section of points that run off with them. Has anyone seen this before? Does anyone know that the fix is?
Seen it plenty. Basically the AT is stuffed .. or needs help.
Check your control points first - make sure it isn't user error - picking the wrong point, using the wrong coordinates .. wrong system .. wrong height offsets .. whatever.
I mentioned control as it is on the edge of your model .. probably where your control is.
But - see all the orange photos? It couldn't auto tie them .. so supply some user tie points in the error to help it work out how the photos join together. There could be some repetitive patterns in there or something .. have a look at the photo list - see which one it starts to break on .. go back a couple - start creating manual tie points .. then re-run.
This is typical broken model with multiple components. Do a camera precalibration so camera parameters are locked and it will at least drop photos which are not possible to tie instead of creating multiple models then it will be easier to add manual tie points. Either overlap was too small or the scene lacks details or there is repeating patterns in it.
Also could try tie point density: high and enable multi component run in aerotriangulation.
Thanks James this makes sense. It is a desert scene so perhaps it couldn't find enough detail. So just to clarify we didn't use any GCP's on this job. Also you suggest using tie points? So add tie points in the photos prior to the break and then run the AT on top of the existing AT?
Hi Oto thanks for your response. To clarify what do you mean by camera calibration? On the Drone? Overlap was 80 /80 so would of through overlap is high enough? Its a sort of desert with red sand. Some buildings on it. Your thoughts?
Hard to comment on all of this without seeing the data set in detail. 80% overlap should be enough .. that said - if you have tall trees and you are flying over the top of them at the same height - the algorithm will struggle. Firstly the tie points on the trees are all moving (wind) and the distance from the UAV to the top of the tree is substantially less than the distance from the UAV to the ground - ie changing your overlap.
The camera calibration is a PDF somewhere. Basically do an orbit around a complex shape and run multipass on the optical calcs. Google contextcapture camera calibration and it will doubt turn up.
So with the tie points .. create some near where your are having issues .. nominate in the photos .. re-run the AT.
If you are going across vast expanses of crops, sand .. this without much unique detail - you run into issues .. or tops of trees.