Object Tracking Using an Improved Kernel Method

Abstract

An improved object tracking scheme is presented based on the Kalman filter and mean-shift approach. And this scheme is robust to disturbance and occlusion of both the object and the scene. The object is selected by using FG/BG detection and represented by its center point and probability distribution. The mean-shift approach estimates the object position based on the result of the Kalman filter. A threshold of the Bhattacharyya coefficient is set to judge occlusion and when object being occluded the Kalman filter estimates the object position. Since the proposed scheme combines the space information with probability distribution, it is robust to disturbance and occlusion.