Multiframe optical flow estimation with oriented smoothness in the direction of motion

A new computational method to estimate the optical flow field is considered. A multiframe approach is taken which takes into account the high temporal correlation among successive images. Given an image sequence, a global energy functional over the spatiotemporal volume of the sequence is defined. The regularization term in the energy functional is based on the assumption that the optical flow field is smooth in the direction of motion. With a motion-based smoothness criterion, the proposed method implicitly preserves spatial discontinuities of the optical flow field. A stochastic relaxation method is used as a test-bed optimization tool. Experimental results on synthetic and real image sequences show that this approach yields an optical flow field estimate that is accurate and boundary preserving.