Dynamic 3D Mapping using variable inputs

In previous projects, I have worked with opencv to try and build stereo and mono camerea vison systems with limited success. This is mainly due to hardware issues such as camera inaccuracies and incorrect settings. I want to build a system which is defensive against such issues. Applying filters, smoothing and other processes to bring out the true geomtery in the 3d scene while removing noise and other errors from things such perspective shift and distortion.