Known Bugs

Known Bugs Pending Resolution

Recalibration & Recapture - BUG FIXED

Fix: Use Rectangular Chessboard

Issue Summary:

• Recapturing chessboard images and recalibrating causes program to crash after recalibration is complete even when chessboard corners are found.

Manifestation:

• OpenCV assertion failure error and abort (width > 0 && height > 0).

Explanation:

• During recalibration, detected chessboard corners are highlighted over each chessboard image.
• It is likely that the corners for the left and right chessboard images will be interpreted in orthogonal orientations (i.e. the left image will have corners detected horizontally and the right vertically).
• This will cause the recalibration error to be significantly large (>1) and as a result, the valid region of interest (ROI) for stereo matching between the left and right images will have no overlap and will be zero (a rectangle with no width or height).
• The ROI determines how the edges of the input images are cropped so that only rectified pixels are operated on, hence a zero ROI results in images being cropped away, leading to the OpenCV error manifested when trying to read the input images into the algorithm.

Temporary Fix:

• Provide precomputed intrinsics.yml and extrinsics.yml files and do not use the RECALIBRATE AND RECAPTURE flags.