Quality assessment of self-calibration with distortion estimation for grid point images

Abstract

Recently, a camera self-calibration algorithm was reported which solves for pose, focal length and radial distortion using a minimal set of four 2D-to-3D point correspondences. In this paper, we present an empirical analysis of the algorithm's accuracy using high-fidelity point correspondences. In particular, we use images of circular markers arranged in a regular planar grid, obtain the centroids of the marker images, and pass those as input point correspondences to the algorithm. We compare the resulting reprojection errors against those obtained from a benchmark calibration based on the same data. Our experiments show that for low-noise point images the self-calibration technique performs at least as good as the benchmark with a simplified distortion model.