Kalman filter based multiple objects detection-tracking algorithm robust to occlusion

Abstract

Visual target tracking is one of the major fields in computer vision system. Object tracking has many practical applications such as automated surveillance system, military guidance, traffic management system, fault detection system, artificial intelligence and robot vision system. But it is difficult to track objects with image sensor. Especially, multiple objects tracking is harder than single object tracking. This paper proposes multiple objects tracking algorithm based on the Kalman filter. Our algorithm uses the Kalman filter as many as the number of moving objects in the image frame. If many moving objects exist in the image, however, we obtain multiple measurements. Therefore, precise data association is necessary in order to track multiple objects correctly. Another problem of multiple objects tracking is occlusion that causes merge and split. For solving these problems, this paper defines the cost function using some factors. Experiments using Matlab show that the performance of the proposed algorithm is appropriate for multiple objects tracking in real-time.