Robust detection of calibration markers using one circle

It is necessary to calibrate a camera for the metric reconstruction of a world using cameras. Usually stereo system is adopted and each camera should be calibrated in advance. While recent self-calibration algorithm only requires correspondences of feature points on image for the calibration of a camera, traditional calibration algorithm usually uses a calibration target. The extrinsic and intrinsic parameters of a camera are computed using one-to-one correspondence between 3-d points on calibration target and their 2-d points on image. Most calibration algorithms concentrate on camera model, computation method, and error analysis. Little attention is given into the automatic detection of control points on image. Detection of control points on image is usually done in semi-automatic or automatic way using their own tuned code. This paper presents a robust algorithm that can detect control points on chessboard image under the variation of scale and rotation. Symmetric properties on chessboard pattern are used for the automatic detection of control points. Cross-ratio is used in grouping control points. First, we generate candidates using Lucas-Kanade feature detector. Then, we verify each candidate by investigating symmetric properties using a circle as a probe. Finally, we can obtain one-to-one correspondence between 3-d points on calibration target and their 2-d points on image. Experimental results show the feasibility of proposed algorithm.