Computation of Optical Flow Using Dynamic Programming

This paper presents an original algorithm for the computation of optical flow called Orthogonal Dynamic Programming (ODP) as well as several enhancements to it. The principle is to minimize a sum of square differences (SSD) between a pair of images. The originality of the approach is that an optimal matching is searched for entire image strips rather than for pixel neighborhoods. Dynamic programming is used to provide very robust strip alignments and a multiresolution iterative process is used to compute the velocity field. Extensions to the comany intermediate image within the sequence (with non integer image index). Compared to classical methods, this one has the advantage to be able to operate on multi-band (color) images and to provide a dense flow field for the whole image (neither holes nor border shrinks). Continuity and regularity constraints enforced by dynamic programming leads to a very good flow field estimation even in homogeneous or aliased areas. 2 The Orthogonal Dynamic Programming algorithm putat& of the velocity field for non integer image Dynamic Programming is a very robust technique indexes, to the use of more than two images, and to the search for subpixel velocities, are presented. Refor searching optimal alignments between various types of patterns because it is able to include order sults obtained for the Barron, Fleet and Beauchemin and continuity constraints during the search. Howperformance tests appear to be at least as good as or better than those obtained using classical optical ever, it is applicable only for the search of monodimensional alignments (the reason is that no natural flow detection methods. order can be found for a multidimensional set) and