Corner feature based object tracking using Adaptive Kalman Filter

The efficiency and accuracy of moving object (MO) tracking in outdoor area is an important issue for the surveillance community. This paper modifies the prototype of Kalman Filter and proposed a corner feature based adaptive Kalman filter (AKF) for MO tracking. Unlike pixel-level feature, corner feature is insensitive to dynamic change of outdoor scenes such as water ripples, plants and illumination. Thus, the corner feature can robustly describe MO in outdoor area. The main contribution of this paper is to take advantage of corner points to describe MO and then use the variation in the number of occluded corner points across consecutive frames to design an AKF. The proposed tracking method is able to track MO in the presence of occlusion and velocity variation. Experiments on challenging video clips demonstrate our method can portray the trajectory of MO accurately and stably.