Recursively estimating optical flow from a noisy image sequence

A token-based method for recursively estimating optical flow from a noisy image sequence is presented. Motion characteristics of the perspective projection of a 3D moving object are analyzed and a dynamic model is built. The recursive solution is achieved through an iterated extended Kalman filter and two schemes for the velocity propagation of match points are discussed. Experimental results indicate that this method, which requires a long image sequence but rather few match points, is efficient and robust even in very poor conditions