Research On Face Detection And Tracking In Video Surveillance System

Face detection ,which is the pre-work of face recognition, has the application in a wide range of fields such as personal identification, visual surveillance and content-based retrieval. Human-face concerned applications is getting nearer and nearer to practical use, thus resulting in more and more emphasis over research for human-face detection. In this dissertation, we have made a deep study on automatic face detection and tracking in video surveillance system.The research of face detection advanced from simple image process to complex real-time video process. However, face detection is a most challenging task because of the complicated pattern and the frangibility of human face, most methods have the weakness of large computation, low efficiency and many false reports among the detection result; but, Viola present a fast face detection method based on AdaBoost learning algorithm in 2001, which made PC can see human face via camera. We can use integral image to quickly calculate the feature, and construct weak classifier by the feature; then weak classifiers are combined to a strong classifier in a linear way. The final classifier is built in a cascade structure, which could reject most non-face samples in the early layer. But as a new method, AdaBoost also need deeply development. Our experiments show that AdaBoost learning algorithm can trulydetect face rapidly, but face detection rate and false rate are not satisfying, especially videos in normal environment.Human skin color has been proven to be an effective feature and widely used in face detection in resent years. The results of experiment show that this detection approach features a rapid inspecting speed and is insensitive to the face posture. But how to improve the performance of the skin detector is a challenging problem.In this dissertation, firstly, an overview and analysis is given of existing typical algorithms for human-face detection; secondly, a multi-stage algorithm is proposed to detect human faces whose numbers sizes and positions are all unknown. The algorithm uses AdaBoost algorithm to preliminarily extract human-face regions, and skin color models are used to verify the selected result; Then we use region shape to filter them, which made the rate detection failure is reduced. Finally, a fast face detection system for color image sequences with complex backgrounds is proposed and implemented. Our test result shows that the detection is improved and false detection rate is much reduced, and our algorithm are effective and feasible.