A continuous tracking algorithm for long-term memory motion estimation [video coding]

It is known that long-term memory motion estimation can bring a significant coding gain to a video coding system. In this paper, a fast algorithm for long-term memory motion estimation is presented. The proposed algorithm attempts to track the direction of the best long-term motion vector in a frame-by-frame manner from the most recent reference frame to the oldest reference frame. Nine direction patterns for the direction tracking are used and the search locations of the long-term memory motion estimation are adaptively selected according to the chosen direction pattern. Simulation results demonstrate that the proposed algorithms can speed up long-term motion estimation by more than 40 times.