Intrinsic camera calibration refinement from independent SFM reconstructions

[kossi]

I am often puzzled by the large variations in intrinsic camera parameters from settle changes is scene conditions. For repeat reconstructions using the same camera (physical lens physically unchanged) it would be nice to have a way to enforce consistency in intrinsic camera parameters across independent 3D reconstructions. In other words, if I use the same camera to reconstruct the Eiffel tower and another object of the same scale, I should expect to have the same intrinsic parameters but currently these parameters tend to vary widely (especially focal length and principal point). The requested optimization approach would find the best set of intrinsic parameters that simultaneously minimizes errors across multiple independent sfm reconstructions.

Thanks