Why is the Charuco Board Calibration so much better against a flipping Z-Axis then the normal OpenCV chessboard?

edit

Can you post the images with the chessboard when there is the issue?

Other people have experienced the Z-axis issue but as the number of points were low (4 points), it could be possible there is something like a singularity when solving for the pose. With much greater number of points, it makes sense if the issue disappears. In your case, it is strange that there is an issue with the chessboard as the number of points is high.

Thank you for your response. I am not able to post a picture but I am uploaded it here: http://www.bilder-upload.eu/show.php?...

Thanks for the image, I have understood the issue now.

Somehow, it looks like the intrinsic parameters estimated using the CharUco Board were "more accurate" than using the classical chessboard and do not impact negatively the pose estimation (solvePnP) from Aruco marker. Calibration can be a tricky step. Looks like you had the same issue than the post I have linked. I don't think you can do much about the Z-axis flipping. If you have the issue again, maybe:

• try to use the refine option when detecting the marker to have more precise corner coordinates
• try to detect if the rotation matrix is right-handed and try to "fix it" somehow, maybe with polar decomposition of the rotation matrix?

A big thank to you! I hope it will help other people to calibrate with CharUco instead of the "normal" chessboard :)

The camera calibration step can be tricky as it requires:

• a chessboard object the most possible flat
• multiple acquired images with good views (multiple orientations, positions in space), see for instance

To be honest, I cannot guarantee that calibrating with CharUco is better than the regular one as the main advantage of CharUco should be that it can handle occlusion of the CharUco board.