I am trying to calibrate the camera with a checkerboard using the well-known Zhang method, followed by the beam adjustment, available in both Matlab and OpenCV. There are many empirical recommendations, but from my personal experience, the accuracy is pretty random. Sometimes it can be really good, and sometimes very bad. The result can actually vary quite simply by simply placing the chessboard in different places. Suppose the target camera is straightforward with a horizontal FOV of 110 degrees.
Does the number of squares in a chessboard affect accuracy? Zhang uses 8x8 in his original article without explaining why.
Does the length of the square affect accuracy? Zhang uses 17 cm x 17 cm without explaining why.
What is the optimal number of pictures with reference to different positions / orientation of a chessboard? Zhang uses only 5 images. I saw people offering 20-30 images with chessboards at different angles, filling the entire field of view, tilting left, right, top and bottom, and suggested that there should not be a chessboard placed in a similar position / orientation, otherwise the result will be shifted to this position / orientation. Is it correct?
The goal is to define a workflow to get a consistent calibration result.
, , " ", , , : RMS- . , , , ( ).
, . , ( , , ). , , , , , (1) , , , ( 2) , FOV. !
[ ]
, , , RMS , RMS 1 , 2, . , , , , ( , , , )? : auto-anything - , ). , , , , (, ). , , - , , . , , , .
, , . , . , . , , , . , 20-30 .
, . . , .
, Calibrator MATLAB. , , .