I have a stereo camera that I am trying to calibrate for depth perception. The individual cameras calibrate fine with low error. Sometimes, and this seems to depend on resolution, the remapped images are bad and either zoomed way out or zoomed way in, and don't have horizontal correspondences.
II have a jupyter notebook with all of the code and calibration images:
https://github.com/berickson/race-vision/blob/d0882be8d4ddb17391b0ca207afa60835dbbeeaf/stereo-calibration.ipynb
The 640_480 images work fine while the 320_240_d images don't work correctly despite what appears to be good individual camera calibrations. Notice that the "remapped L" and "remapped R" images are bad. I need the low resolutions to work as I am planning on using this on a resource limited computer on a robotic car. Am I doing something wrong that is causing these bad projections?
