A tracking system with multiple region background subtraction algorithm for night scene

The tracking system is designed to track humans by subtracting objects of interest i.e. humans, from the background. For the day scene, performance is quite good; however, in the night scene, tracking becomes quite tough. In this paper, a tracking system that includes a new foreground algorithm and tracking algorithm is proposed to thoroughly consider the night scene. We first propose a new foreground detection algorithm to process the complicated scene including shadow and illumination. It applies pixel illumination and background classification in multiple regions. By use of the proposed multiple-region background subtraction algorithm, the interferences of illumination or shadow would be removed. With the help of this technique, the segmentation result would be more accurate even at night scenes. Then we propose a modified silhouette tracking algorithm by using boundary and velocity information of the object is proposed. It can deal with the occlusion caused by other foreground objects. Experimental results prove that the proposed system outperforms the accuracy of foreground detection for the night scene.