I’m trying to calibrate a couple of ELP USB 3.0 cameras to be eventually used in a stereo configuration, but have had less than stellar results trying to undistort images on each camera so far. In particular, I do not see much radial distortion, if any, on the original camera images, but there is a lot of distortion, especially at some edges, of “undistorted” images after they have been run through cv2.undistort(). Specifically, I’m using 62 calibration images with a 9x14 interior point chessboard, with code identical to that in the opencv Camera Calibration tutorial:
Camera Calibration — OpenCV-Python Tutorials 1 documentation (opencv-python-tutroals.readthedocs.io)
I’ve only modified the number of interior points of the chessboard where required. The images look like so:
I realize the undistorted one is naturally warped at the edges, but even the box the chessboard is attached to appears warped to some extent (not a perfect rectangle), and certainly does not appear undistorted to me. Can anyone point me to what I am doing wrong? Perhaps the original image is close enough to being undistorted that I can skip right to rectification?