Radial distortion correction in a vision system

Abstract

Accurate correction of lens distortion is the core technology in visual based measurements. Radial distortion is the most significant lens distortion in a vision system. In this paper, a radial distortion correction method named the iteration geometrical similar method is proposed. The initial method is used to obtain a geometrical similar degree of image points. The distortion coefficients are then estimated by using the iteration approximation model. The radial distortion coefficients and other camera parameters can be successfully decoupled. Both simulative and practical experiments have been carried out to evaluate the performance of the proposed method. First, the influence of image noise and distortion degree on correction results is discussed in computer simulations. The results show that correction performance of the proposed method is manifested well. The proposed method was also compared with linear regression analysis (LRA) and the diversion model (DM) in the practical experiment. The results indicate that the average straightness calculated by the proposed method is only 0.195 pixels, compared with 0.394 pixels and 0.438 pixels by LRA and DM, respectively. In addition, the rectified images are used to calibrate the camera. The mean reprojection error using rectified images corrected by proposed method is only 0.137 pixels, which is significant less than 0.421 pixels and 0.462 pixels by the LRA and DM, respectively.