A Hierarchical Algorithm for Fast Projective Flow Estimation

Abstract

This paper proposes a two-stage featureless hierarchical algorithm to alleviate the problems encountered in the VideoOrbits (and other related) algorithms for estimation of the projective coordinate transformation between images. Translation, rotation and zoom parameters are estimated by phase correlation methods in the first stage, which are used as initial estimates of Orbits in the second stage. In particular, a hybrid algorithm devised to use the first stage only when the projective motion is large demonstrates the most significant improvements in efficiency and accuracy over using Orbits alone on a motion sequence with large interframe movements