Research On Real-Time Face Detection And Pose Estimation In Video Sequence

Science and Technology change quickly, various technical in the fields of computer pattern recognition and computer vision have also been researched and developed continuously and deeply, and also makes lots of practical application. Face detection and track and pose estimation is one of important topics which caused a long-term care and more and more research. During various kinds algorithm of face detection and pose estimation have been proposed and implemented continuously, face detection and pose estimation makes more and more research and development in Identification Recognition and Validation, Virtual Reality, Intelligent Video Surveillance, Human-Computer Interaction and so on. It has a wide development prospect and great commercial applications value, and has been the research hotspot in the fields of computer pattern recognition and computer vision.At present, there are lots of research and algorithm at home and abroad, and various classification of algorithm for face detection. In this paper, image which will be detected has been divided into two types by image source, static image and dynamic image. And also designates the problem of face detection in dynamic image can transformed into the problem of face detection in static image. After generalizing and making a brief introduction of face detection algorithm in static image, then choose AdaBoost learning algorithm for detecting face. According to the principle of the classical AdaBoost learning algorithm, has completed the work from classifier training to face detection. Select a specific size for the normalization of the training sample to ensure the training speed of classifier. And improve the scale of classifier zoom and search process to ensure the speed and the accuracy of face detection. Finally, do experiments and analyze the results, which verified the feasibility and effectiveness, and able to meet the requirements of the research.We use the method which is combining LK optical flow tracking algorithm and CamShift tracking algorithm to track face and estimate pose in order to ensure the speed and accuracy. Choose LK optical flow tracking as the predominent, and supplemented by CamShift tracking. To the result of comparing the information of the region location by tracking with the information of the based region location by setting do pose estimation. First locate facial feature points by ASM, and determine the positive face by the position relationship of facial feature points.While the location of the positive face has been matched, setting the information of the based region location and extract the feature points to start LK optical flow tracking, compare position of facial feature points to estimate pose.The result of CamShift tracking has been used to estimate pose when the optical flow tracking is failed. The experiment results show that the method is feasible.