Radial Lens Distortion Calibration from Spheres: Theory and Method

Abstract

This paper addresses the problem of lens distortion calibration from a single image of a sphere. Almost all calibration methods based on spheres have focused on estimating all the camera parameters except the lens distortion ones which are left for another independent calibration process not using spheres. The paper presents for the first time a radial lens distortion calibration method based on the analysis of local shading measures in the sphere image. Several properties of these measures are pointed out and validated. While these measures are robust to noise, they are shown to vary nonlinearly as a function of lens distortion. As such, they can work as a basis for lens distortion calibration. The proposed method is easy to use and requires no specific feature extraction from the images. Experiments on simulated and real data are reported with comparisons to well-known techniques.