Chessboard corner detection algorithm based on minimum cross entropy

The shortcoming of the present B/W chessboard corner detection algorithm is analyzed and a new method based on cross entropy is proposed. Firstly, the pixels around the corner are divided into 4 quadrants, and initial selection of corners is carried out based on the gray value difference between the adjacent quadrants; secondly, the cross entropy of the diagonal quadrant is defined, and the corner screening is done using the principle of minimum cross entropy; thirdly, the idea of non-maximum suppression of local gradient amplitude is introduced to solve the problem of local overlap of the candidates; at last, sub-pixel coordinates of corners are calculated using Frostner Operator. Experiments and their analysis prove preliminarily that: (1) the detection result of this algorithm is better than the classical Harris Operator and SV Operator; (2) the sub-pixel accuracy obtained is almost the same as that obtained with the Matlab Camera Calibration Toolbox, and it is suitable for online camera calibration.