Rectification of Uncalibrated Images for Stereo Vision

Abstract

This paper aims at rectification of uncalibrated stereo images. We use the direct method to compute the rectifying transformations, not to deduce from the fundamental matrix, so to reduce the uncertainty; Minimize the distance from the point to its epipolar line to ensue the uniqueness of the two rectifying transformations. We only extract a few correspondent points, make use of the condition of the correspondent points in the rectified images also meet the polar constraint, minimize distance of all of the correspondent points to the corresponding epipolar line, then obtain the system equations, through solving system of equations, thus get the rectifying transformation matrix and rectify the stereo image pairs. Experiments show that, the size of the rectified images change not so seriously, the vertical disparity is eliminated, so the subsequent stereo correspondence can be fulfilled on the horizontal scan line of the rectified images completely. DOI: http://dx.doi.org/10.11591/telkomnika.v11i1.1903