0

I have been trying to calibrate stereovision camera setup for a while. After a lot of testing I came to the illogical conclusion that the more chessboard pictures I use for the calibration, the higher the reprojection error of both individual and stereo calibration (roughly speaking). How is that possible considering that the larger set you use, the smaller error should emerge? The pictures were taken following these remarks, so I am almost certain that the pictures are of good quality.

Here are some results:

82 pictures in total

40 pictures in total

10 pictures in total

Johnny Doerman
  • 11
  • 3
  • It would be difficult for anyone to answer with the few data you give. You can check if the chessboard corners are correctly detected (try `cv::drawChessboardCorners()`), look at the reprojection error per image, etc. It can also be a experimentation problem, for example the relative position between your cameras changed between images. – vvanpelt Mar 08 '21 at 16:09
  • Just to clarify a few points, camera is fixed rigidly while I setup the board to vary on each axis. I also thought that maybe the chessboard coroners were not found, however I only use the image points of the pictures where all the points were found. I also display each image after the cv::drawChessboardCorners() to visually check if the corners are where they are supposed to be. I will try to do the reprojection error per image and post the results. – Johnny Doerman Mar 09 '21 at 10:26

0 Answers0