Three types of reprojection error on spherical epipolar geometry
Abstract
To compute an epipolar geometry for spherical images, error evaluation function, which is define as the sum of squares of reprojection errors, is generally minimized. For pin-hole camera, it is natural to measure the reprojection error as Euclidean distance, which is the quantities defined on image plane. The same as this, for spherical camera, it is natural to measure the reprojection error as the quantities defined on image sphere, not the quantities defined on plane. In this paper, three types of distance are defined as the reprojection error for spherical image. The distances are distance along geodesic, difference of longitude and arc length along colatitude. For three types of distances, the essential matrices for epipolar geometry are computed and the performances evaluated by comparing to the eight-point algorithm under three-dimensional reconstruction error of synthetic data.
Origin : Files produced by the author(s)