Fast approach to checkerboard corner detection for calibration

Abstract. To reduce accuracy lost in the calibration process for high-precision optical systems using interferometry, an approach is proposed to detect checkerboard corners based on the level set evolution principle. Compared with existing corner detection methods, no image gradients are required for segmentation of checkerboard patterns. It has the capability of doing corner detection for the images acquired under more complex imaging environments, like underwater, low-contrast, blurred, and heavily distorted images. In addition, no iteration is required in the level set evolution procedure, and a fast speed is achieved. In this implementation, the grids that consist of a checkerboard pattern are first found as level set curves by segmenting the checkerboard pattern image. Then, noting that checkers might be recognized as quadrangles, the four corners of a quadrangle can be located by checking the varying of points of its boundary in slope. Alternatively, they also could be located according to the maximal distance at specific orientations between a point and the center of the closed curve. Finally, several experiment results are presented to validate the proposed approach and to demonstrate its robustness and correctness.