1. The pattern of the calibration plate
The calibration plate can be three-dimensional, two-dimensional, circular, grid, random pattern, and generally use multiple angle views of the flat chart instead of constructing a three-dimensional object.
2. The number of checkerboard corner points and the number of parameters
1. Number of parameters: internal parameters: 4 camera internal parameters, 5 distortion parameters;
External parameters: 3 rotation parameters, 3 translation parameters
2. Assuming that K pictures of the calibration board from different angles are taken, and each picture has N corner points, then K pictures can provide 2NK constraints (each corner point can provide 2 constraints), and temporarily ignore the distortion parameters, then There are 4 internal parameters and 6K external parameters. The conditions that can be solved are: 2NK≥6K+4, that is, (N-3)K≥2. The homography matrix was mentioned earlier, and a homography relationship can be represented by 4 points , so no matter how many corner points there are on the checkerboard, only 4 points are actually useful (refer to three, lens distortion___windlike's blog-CSDN blog )
So N≥4, so K≥2. That is to say, at least 2 checkerboard pictures need to be taken.
3. Considering noise and numerical stability, generally more checkerboard pictures will be taken, and more image information will be collected. Generally, more than 10 pictures, 7*8 or larger checkerboard grids will be used.