How to find errors in stereo calibration and rectification?

edit

Look at the samples in cpp folder. You have visual clue with the epipolar lines (if they get thought the same corner/parts of the chessboard). You have also a reprojection error computation, using the epipolar geometry constraint: m2^tFm1=0 (as state in the sample stereo_calib.cpp). If you want to measure the stereo reconstruction, look at the sample stereo_match.cpp which could save a point cloud. Opening this point cloud, with meshlab for example, allow to measure the size of object in 3D (like the size of the square if you know it and have used the right size in the calibration process). You could also automatize this process by taking each 3D positions of the chessboard corners and compute the means distance between them.