A Motion Filtering Algorithm Based on Kalman Filter

The Electronic Image Stabilization (EIS) contains three parts: motion estimation, motion filtering and motion compensation. By motion estimation, motion vectors will be acquired, but the global motion vectors are comprised of random jitter and scanning movement of camera, so it is necessary to eliminate the vibrations and keep the active movements. In this paper, it is a self-adaption Kalman filtering algorithm which is applied to motion filtering. The Kalman filtering makes the assumption that the system noise and observation noise are sequences of discrete white noise which is subject to Gaussian distribution with the known variance. However, in most cases, the assumptions are not correct, so in order to separate the scanning movements from the motion parameters, a Sage-Husa self-adaption motion filtering based on Kalman filter will be applied to motion filtering to get the active movements. The experimental results have shown that the Sage-Husa self-adaption filtering algorithm can get the motion vectors more accurately than the classical Kalman filtering.