Past Iteration Algorithm For Optical Flow Computation

Abstract

In this paper, we propose the successive over-relaxation (SOR) method to speed up the optical flow computation. We derive a linear algebraic system having a large sparse matrix from the elliptic partial differential equations by finite different method. Using the successive over-relaxation (SOR) method, the system can be solved by gaining speed factor of at least two over other algorithms. Comparing to the conjugated gradient (CG) method, this method requires local neighbour connections only for each iteration without involving the whole large matrix interaction and it takes much less computation time for each iteration. Therefore the SOR method is more efficient.