Correcting Radial Lens Distortion by Two-View Point Correspondences

This paper presents a two-view point correspondences based method to correct radial lens distortion, and it can deals with lens radial distortion of different levels with only two images captured at different orientations. The distance of adjacent feature points is adjusted to make them distribute as evenly as possible in overlapping areas. Then symmetric transfer error function is used to construct correction model of radial distortion, and a progressive complexity strategy is adopted to solve calibration parameters. In case of severe distortion, it corrects the coordinates of all feature points with last solved parameters and rematches the feature points. In this wary, it increases the number of inliers and makes them move from center to peripheral of overlapping area, thus, to improve the accuracy and robustness of lens correction.