with a IP-camera with significant distortion and unprecise focal distance estimates, pose estimation from GCPs may become inprecise
This may cause distortion or elongated parts in areas that are enlarged by the camera lens for better FOV
Causing too large distances in areas uncovered by control points
Hence overestimation of velocities.
This needs to be tested with a set of known points and known reference such as the camera position itself with a camera that has significant distortion issues or a manipulated camera image (this would make the experiment more controlled).
Possible solution:
automated tuning of parameters with RMSE and/or known camera position
estimation of straight lines perspective and some kind of automated "straightening" through updating the distortion coefficients
Description:
Possible solution: