Hello,
I have a project with less-than-ideal photo overlap, as it's using historic aerial photography. We successfully created reality meshes and point clouds from these photos in an earlier phase of this product, but now we are trying to create georeferenced models using surveyed ground control points and failing repeatedly. The original format of these points is Nevada State Planar NAD 27, in US Survey feet, with elevation and height above ellipsoid in NVGD 29 . I tried the GCPs with the coordinates in US Survey feet and the ellipsoidal height in meters, then with everything in meters. I have converted the points to NAD 83 geographic with elevation in NGVD 88 metyers and choosing "NAD 88 + NGVD 88" wherever spatial reference system is asked for. According to our surveyor, we can't get height above ellipsoid in NGVD without going back to the field. We have had many tries failing on Aerotriangulation. We've varied the settings in many ways including "one-pass" vs "multi pass" and "use control points for rigid registration" vs "use control points for adjustment". We mostly use "keypoint density normal" and "pair selection exhaustive" but have tried varying these as well. We have tried Selection of Matching pairs "for structured aerial data set." We leave most other things as the default that auto-fills. We have a sensor size of 230 mm and focal length of 153.36 mm because of the aerial photography camera. We have usually been entering this info, but have tried leaving it out too. We had failed ATs and ATs that looked good from the accuracy report, except that all the GCPs turned red because the reprojection error was too high. We finally got an AT that looks great. With good accuracy, no photos dropped, and no reprojection error problems. Looks normal on the 3D view. We ran a production from this AT, and it showed it as successful on all tiles, but the 3mx file just looks like a line. and in 3D view the points all look stacked. What are we doing wrong? Please help. I have spent way too much time on these repeated failures.
Thanks,
Heather Gang
Which version are you using? I would recommmend use CC Update 13 as it has some improvements regarding aerotriangulation. Also do not add GCPs until successful aerotriangulation as bad GCPs may drop photos which otherwise would align.
I have tried also some historic photos which are scanned negatives with quite good results - [Contextcapture] Reality model from German Flown Aerial Photography, 1939 - 1945
What it seems is that you should get first estimations as close as possible so if camera parameters are known then it is already good start. Ideally if the camera is still available you could do camera calibration. Also reusing camera calibration from other softwares may help for example try doing alignment in Agisoft or Pix4d (demo is free and you don't need save option) and then import in Contextcapture.
Hello Oto, Thanks for your advice. We have CONNECT Edition, update 11. I don't know if we will be able to get the newer version. I did not know you could add the GCPs after Aerotriangulation. I will try that. The camera is not still available to us to do calibrations. We do feel fairly confident in the sensor size and focal length, because this info was on the film canister. I hadd run across references to your work with the German photos, which is part of what inspired us to try this. Ours are also scanned negatives.
Also Oto, what spatial reference system did you use? I have a feeling that our problem may be somehow related to that. We are not dropping photos during AT, although some don't have many tie points, creating large holes on reconstruction. Do you know if you have to use the same SRS for input and output? There are several places during the process where SRS is entered. I've mostly tried to keep the same SRS throughout. I think your advice on the Bentley forum about your aerial photos was where we got the settings we've mostly been using (like multi-pass and keypoint selection exhaustive, etc.) but we've tried a lot of different settings. Did you add user tie points? Thanks again!
Internally Contextcapture works with ECEF coordinates so there are no multiple SRS systems. Coordinate systems can only be defined only when importing image positions, adding GCPs, Point clouds and when generating final production models. In viewer you can choose any coordinate system it only changes readout.
I have tried adding user tie points but it mostly only works if the photos do not align correctly if they do not align at all then it won't help much.
>but the 3mx file just looks like a line. and in 3D view the points all look stacked
This happens if Contextcapture starts with real coordinates but drop them because can't align photos correctly using real coordinates so it turns back to relative coordinates and leaves ECEF coordinates by mistake and that causes floating point rounding issues.
What you could do is using Google Earth add these photos as overlays scale them relative to ground and find center coordinates for each image. Then these positions could be imported and at least first aerotriangulation will be in approximate coordinates and that could help you to figure out why survey points are not working maybe there are issues with these points.