Motion Estimation using Kinetic Energy with Controlled Friction

This paper presents a new energy minimization approach to the motion estimation. The conventional energy minimization approaches yield local minimum solutions corresponding to initial states. In the case of video processing which includes many objects moving randomly, it is impossible to give a proper initial state to each object. We propose to introduce "kinetic energy with controlled friction" in order to get better solution automatically. The experimental results showed the proposed method found the lower energy than conventional methods.