Motion prior based on general motion statistics

Motion estimation plays an important role in scene matching, SLAM (Simultaneously Localization And Mapping) in computer vision field. In BMA (Block Matching Algorithm) for the video compression, motion estimation is also considerable. Motion prior which could be an initial guess for the next motion has an effect on how well motion is estimated. In many applications, motion model is assumed to be Gaussian model with constant velocity. In this paper, we propose a new motion model based on general motion statistics and demonstrate proposed motion model helps motion estimation become more accurate.