Multiple human detection and tracking based on head detection for real-time video surveillance

Multiple human detection and tracking is a very important and active research topic in computer vision. At present, the recognition performance is not satisfactory, which is mainly due to the fact that the full-body of human cannot be captured efficiently by cameras. In this paper, an improved method is developed to detect and track multiple heads by considering them as rigid body parts. The appearance model of human heads is updated according to fusion of color histogram and oriented gradients. An associative mechanism of detection and tracking has been developed to recover transient missed detections and suppress transient false detections. The object identity can be kept invariant during tracking even if unavoidable occlusion occurs. Besides, the proposed method is fast to detect and track multiple human in a dynamic scene without any hypothesis for the scenario contents in advance. Comparisons with state-of-the-arts have indicated the superiority and good performance of the proposed method.