Kalman Filtering Based Motion Estimation for Video Coding
Nai-Chung Yang, Chaur Heh, Chung Ming · InTech eBooks · 2009
In this chapter, we have introduced two types of motion estimation based on Kalman filter, without and with rate-constraint. The first type employs the predicted motion information and the measured motion information to obtain an optimal estimate of motion vector. The predicted motion is achieved through the use of AR models which characterize the motion correlation of neighboring blocks. The measurement motion is obtained by using any conventional block-matching fast search scheme. The results indicate that the method provides smoother motion vector fields than that of the full search scheme, and saves computational cost significantly.