Binocular stereo vision calibration based on alternate adjustment algorithm

This paper presented a high precision camera calibration algorithm of binocular stereo vision system based on alternative adjustment. This method could be achieved by taking multiple pictures on a professionally designed calibration board in different directions. The watershed algorithm is used to extract feature points, and time domain coding is used to match image pairs. The camera initialization parameters including radial and tangential distortion coefficient are obtained by the classical monocular camera calibration algorithm. We built a binocular vision calibration system which is based on the left and right camera coordinates as the reference coordinate system respectively, and the internal parameters of the two cameras are optimized by alternately adjusting experiments to obtain the optimal value. Then, all the internal parameters are used to optimize the external parameters, so as to obtain the optimum distortion parameters and internal and external parameters. Compared with Zhang’s calibration method, the results show that the mean absolute relative error of the two camera parameters is less than 3.45e−6, and the mean re-projection error of the 10 sets of data is about 0.0725 pixels. The mean re-projection error obtained from the distorted model of Zhang is about 0.0852 pixels. Compared with other calibration methods, the proposed algorithm has higher accuracy and convenient, cost-effective and easy to implement.