I used a DJI Inspire 2 w/X4S camera to obtain nadir and 45 deg oblique imagery in both directions for the use of reality modeling in ContextCapture. Extensive ground control was placed and recorded to constrain the model. After running a quality check against the DTM, the points were all within 0.1 ft with two exceptions at approx. 0.35 ft. The quality report from ContextCapture also exhibited good results from AT (see below).
When checking elevations from the 3SM to the same DTM used to check GCP quality, the center of the mesh matches in elevation but the extremities of the mesh are upwards of 15 feet apart. The mesh is distorted into a concave shape even though there is control at the extremities of the alignment/mesh.
Could you provide some guidance on how to address this issue? Thank you.
At first don't put all GCPs as control points, leave some as checkpoints to control the results.
Secondly it is best to do a good first aerotriangulation and only then add GCPs as GCPs doesn't seem to correct camera calibration only tries to warp the model on top of GCPs. Reuse the calibration with same focal length from other project where there is fly around some structure so the calibration is good. For good calibration the focal length after aerotriangulation shouldn't much differ from input values. Also which CC version do you use?
Thanks for the response! We have a run a calibration on all our cameras to get the calibration values independently (. Would it be better to manually input those values instead of having CC determine them and can you do that in CC? I am using CC Update 12.
<width>5472</width> <height>3648</height> <f>3673.98076996467</f> <cx>14.7645309021603</cx> <cy>-6.39059760546769</cy> <k1>0.00579541035354978</k1> <k2>-0.00485341520383045</k2> <k3>0.00476312838462881</k3> <p1>0.00197524278308092</p1> <p2>-0.00126717847972054</p2>
Yes you can input them before aerotriangulation.
I'm currently running the reconstruction with only 5 GCPs arranged in a "+" pattern to focus on the subject highway, and input the independent camera calibration values. I will let you know the results when it is finished and checked against the DTM in InRoads.
After the most recent reconstruction there is still some concave shape to the mesh but it brought the vertical error down to a max of 0.8' at the edge. This is still surprising since there is a nearby GCP. Also there are horizontal offsets up to 1' in places. My other workflow involves using Agisoft Metashape were I can consistently reconstruct with a vertical accuracy of 0.1' to 0.2'. Is it reasonable to expect similar results from CC?