A real-time surveillance system with multi-object tracking

Multi-object detection and occlusion tracking in the computer vision field is an important research topic on the modern surveillance system. These studies largely rely on computer vision work. However, most algorithms are too complex and not practical to be used as real-time systems. This paper proposes a real-time surveillance system. The proposed method mainly improves the foreground detection to get low-complexity and high-quality effects of tracking. A novel occlusion-adaptive tracking method is also applied. It can immediately track multi-objects in successive positions without color cues and an appearance model. To track moving objects, the proposed method uses labeling information to eliminate noises and group moving objects. Additionally, we are also concerned about several cases of occlusions to increase the tracking efficiency. The comparison has been performed with other algorithms of background subtraction. Experimental results show that the proposed method has better performance than other foreground detection methods in terms of both computation speed and detection rate.