Distortion correction using a single image based on projective invariability and separate model

A distortion correction method that is independent of camera parameters is proposed based on projective invariability and separate model. Cross ratio and collinearity of feature points in a single target image are utilized as geometrical invariability, and the separate model constraint for distorted points is also considered. Distortion coefficients and corrected points are acquired by non-linear optimization algorithm simultaneously under the mentioned constraint conditions. Only one view of the chessboard target image is needed to determine the complete distortion coefficients. Subsequently, the distorted points in multiple images are all corrected by means of the calibrated distortion model and the camera intrinsic parameters are calculated linearly according to the mapping relationship between two sets of points. Performance of the proposed method is comprehensively compared with the existing methods under the circumstances of different lens distortion degree. Simulative and real experiments show that the correction performance of the method presented in this article is advantageous in terms of the calibration accuracy. The proposed method is simple and feasible for machine vision system, since the correction results can be applied directly to the subsequent camera calibration process resulting in a satisfactory precision.