Minimum checkerboard images for stereo calibration and correction
What is the minimum number of pairs of checkerboard images for mathematical calibration and correction of two cameras? One pair is viewed as one kind of checkerboard by each camera, ending with a left and right image of the same scene. As far as I know, we only need one pair for the stereo, as the stereo is looking for the relationship between the tow cameras.
source to share
Stereo calibration provides not only rotation and translation between two cameras, but also internal and distorting parameters of each camera. You need at least two images to calibrate each camera separately to get the built-in functionality. If you have already calcined each camera separately, then yes, you can use one pair of checkerboard images to get R and t. However, you won't get very good accuracy.
Typically, you need 10-20 pairs of images. You need enough images to cover the field of view and have a good distribution of the board's 3D orientations.
To calibrate a stereo pair of cameras, you first calibrate the two cameras separately, and then perform one more joint optimization of the parameters of both cameras, as well as rotation and translation between them. So one pair of images just doesn't work.
Edit: The camera calibration algorithm used in OpenCV, Caltech Calibration Toolbox, and Computer System Tool for MATLAB is based on the work of Zhendguo Zhang . His article explains it better than I ever could.
The crux of the problem is that the points on the chessboard are coplanar, which is a degenerate configuration. You simply cannot solve this problem with just one kind of planar board. You need more than one kind, with the board in different 3D directions. Views in which the boards are in parallel planes do not add any information.
source to share
"One image with three corners gives us 6 pieces of information that can be used to solve both internal and distorting."
I think this is your main mistake. These angles are not independent. A 100x100 checkerboard pattern does not provide more information than a 10x10 pattern in your ideal world, since the points are on the same plane.
If you have one kind of checkerboard, the closer distance to the board can be compensated for by focus so that you are not (even in your ideal world) able to calibrate the internal and external parameters of your camera.
source to share