Re-examining the optical flow constraint. A new optical flow algorithm with outlier rejection

Most optical flow algorithms are derived from the commonly used optical flow constraint, which assumes the first derivative of the 3D intensity function with respect to time is zero. In this paper, we argue that since this constraint is true only in an infinitesimal neighborhood around the observation point, in practice, in order to improve the estimates and to cope with larger motion, the observation point should migrate gradually according to the current motion estimates. Based on this argument, we propose a new optical flow algorithm. Simulation findings show that the new algorithm leads to (1) faster and better convergence; (2) capability to handle larger motion. Moreover, a simple and effective outlier rejection mechanism is also proposed to handle motion discontinuities in the flow field. It is shown to effectively reduce the motion compensated intensity difference and motion error.