Human Motion Detection And Tracking

The detection and tracking of human motions is key to the visual analysis of human movement, and has also been attached more and more importance to computer visions recent few years. Research in this domain involves wide application (including security surveillance, human-computer interface, the details analysis of human movement, etc.). Especially after the event 9.11, security issues have been paid unprecedented attention around the whole world. The Intelligent Security Monitoring System based on computer vision not only completes the safeguard tasks efficiently, but also saves a great deal of human labor and efforts. Generally, research in the visual analysis of human movement provides a promising application future, as well as great economic benefits to the society.This paper presents an automatic human detection and tracking system, which could analyze and deal with the image sequence gathered from a fixed CCD camera, recognize and keep tracking of human motion. The system consists of motion object detection, human recognition and human motion tracking.The objects detection section describes background subtraction technique adopted by the system and discusses the features of background model and foreground detection in RGB space, HSI space and YUV space respectively. We also present a background model initialization algorithm with iteration form, which doesn't have the requirement that the motion foreground cannot exist in the training image sequence in background initialization stage as the traditional algorithm does.The human recognition section provides an efficient algorithm, which performs well in the system by employing the unique features of human figure and the priori knowledge of the scene.The human motion tracking section discusses the use of combined algorithm to prevent Kalman filter from diverge, while to estimate the motion state of human in image sequence accurately. When people mutually occlude forming group and then split, the algorithm in this paper matches human based on their color feature, resolves the problem to match human before forming group and after group splitting, and achieves satisfied results. The algorithm is fast and robust.The system has been demonstrated through a series of tests with image sequences gathered from both indoor and outdoor, showing great practicability.