Automatic Computation of the Fundamental Matrix from Matched Lines

Abstract

This paper addresses the recovery of epipolar geometry using homographies computed automatically from matched lines between two views. We use lines because they have some advantages with respect to points, particularly in man made environments. Although the fundamental matrix cannot be directly computed from lines, it can be deduced from homographies obtained from them. We match lines lying on a plane, estimating simultaneously a planar projective transformation with a robust method. A homography allows us to select and to grow previous matches that have been obtained combining geometric and brightness image parameters. Successive homographies can be computed depending on the number of planes in the scene. From two or more, the fundamental matrix can be obtained.