Application Of The Active Face Tracking Based On The Mean Shift Algorithm In Video Surveillance System

Human face with a lot of visual information, such as identity validation, expression information, face recognition, tongue identification, which are all built on the basis of human face. Consequently, face detection and tracking turn to be one of the important research contents in intelligent visual surveillance, but also is widely concerned as the forefront direction in the field of computer vision in recent years.Based on the research of face technology, the main content of this dissertation is to build a basic framework and system platform for the active face tracking in accordance with face detection, face tracking and Pan&Tilt feedback control, and do research on the key issues of each step. Main work and innovation are as follows:(1) Face Detection Algorithm Based on the Skin-colorFace detection is the foundation of face tracking. A fast accurate face detection algorithm must be designed firstly. Based on the real-time requirement of video surveillance system, the skin-color feature is selected as the main features of human face in video and the facial geometric feature is used for face validation. Experiment shows that the method can achieve good results in speed and accuracy.(2) Face Tracking Based on the Improved Mean Shift AlgorithmAt the stage of face tracking with background motion and drawn lessons from the method of target tracking, the paper gives some comprehensive research on the various faces tracking algorithm at present. Aimed at the real-time requirement of video surveillance system, a real-time accurate face tracking algorithm based on Mean Shift algorithm is designed. The object template and the radius h of tracking window are updated by estimating the value of comparability variationsΔρ, which solve the impact of face size change and face rotation and improve tracking adaptability.(3) Research on Face Occlusion and the Active Face TrackingFirstly, on the research of the key issues of faces tracking, this dissertation improves the problems of the face cross and occlusion by the method which combines Mean Shift algorithm with Kalman filter. The accuracy of face tracking is promised.Secondly, on the research of the active face tracking with background motion, established the corresponding relations for the angle of target movement to the angle of Pan&Tilt turn, Pan&Tilt turn can be controlled better through face tracking feedback, and face can be locked in the center of the camera.In addition, based on the image process platform DM642, this dissertation transplants the code in DSP and sets up an integrated active face tracking system. The experiments prove that the algorithm is effective.