Research On Video-based Face Detection And Tracking

Video-based face detection and tracking is one of the key tasks in computer vision. With the rapid development of computer technology, it has a wide range of applications in video conference, human-computer interaction, judicature identification, video surveillance, and entrance controlling, etc. Compared with still image, color image sequence provides more information, such as color, motion, and so on. However, it should be more robust to different imagery conditions, illumination conditions and complex background. It also can deal with the face pose, occulision, motion and so on effectively. Additionally, it demands lower computational cost. The study on video-based face detection and tracking is valuable in both theory and practice.This thesis mainly focuses only on video-based face detection and tracking. Do research in face detection and tracking in varity of ways, a multi-faces tracking system is designed in that targets tracked are initialized by face detection. The main contributions of this thesis are as follows:Regarding the face detection, this thesis adopts Adaboost integral method and the extended rectangular features put forward by Rainer Lienhart. Compared with those rectangular features that Viola used, this rectangular features library adds the 45°rotated rectangular features, and extends the training range, increases the hit rate, and decreases the false alarm rate. In this thesis, a cascade method is presented. Detect the image in frontage and profile separately, then merge all of the regions to gain a complete face. This method achieves the purpose of multi-pose face detection.Regarding the face tracking, the recent developments and main algorithms of face tracking are introduced in this thesis, and the Continuously Adaptive Mean Shift (Camshift) algorithm is discussed in detail. Since the algorithm can robustly track target of different shape and size with the immunity against illuminant fluctuation and noise inference, and has low CPU load, it can be served as an efficient human-computer interface. However, Camshift performs unsatisfactorily when flesh-like interference and occlusion occur. Some methods, such as human face feature, accessory information and so on, are proposed to enhance the robust of Camshift. Finally, many methods such as kalman filter, particle filter, finally adoption of a simple linear motion-prediction to predict face are researched. Empirical data has testified that the proposed algorithm can overcome the defects of the Camshift algorithm, enhance the tracking accuracy and reduced the computation, ensure the real-time face tracking.Based on the research of face detection and tracking algorithm, a complete face tracking intelligent video surveillance system framework is built, which is also a constructive attempt to drive the academic research into practical applications.