Estimation of the Lens Distortion Model by Minimizing a Line Reprojection Error

Abstract

Most techniques for camera calibration that use planar calibration patterns require the computation of a lens distortion model and a homography. Both are simultaneously refined using a bundle adjustment that minimizes the reprojection error of a collection of points when projected from the scene onto the camera sensor. These points are usually the corners of the rectangles of a calibration pattern. However, if the lens shows a significant distortion, the location and matching of the corners can be difficult and inaccurate. To cope with this problem, instead of point correspondences, we propose to use line correspondences to compute the reprojection error. We have designed a fully automatic algorithm to estimate the lens distortion model and the homography by computing line correspondences and minimizing the line reprojection error. In the experimental setup, we focus on the analysis of the quality of the obtained lens distortion model. We present some experiments that show that the proposed method outperforms the results obtained by standard methods to compute lens distortion models based on line rectification.