Binocular Camera Calibration Based on Three-dimensional Reconstruction Error

Abstract

This paper presents a new high precision binocular camera calibration method. In the classical calibration method, the camera parameters are usually obtained by the optimization calculation of the re-projection error. However, for the system used for three-dimensional optical measurement, this error does not represent the result of three-dimensional reconstruction. In this method, a specially designed calibration board is photographed at multiple angles. (there is no strict requirement in the calibration process.), the intrinsic parameters of the two cameras are obtained by the image. Then, the rotation matrix and translation matrix between the two cameras are computed by the centroid distance increment matrix method. Thus the degree of coupling between the parameters is reduced. At the same time, the space intersection method is used to reconstruct the 3D coordinates of calibration points. Finally, the reconstruction error is calculated and minimized to optimize the calibration parameters. This error directly reflects the efficiency of 3D reconstruction and is more suitable for the quality evaluation of binocular camera calibration. Experiments show that this method is simple, accurate and significantly improves the accuracy.