Face Tracking Algorithm And Its Implementation On The Dam6416 Platform

Face detection and tracking is an important research sub-field in artificial intelligence and computer vision. As a key technology of facial information processing, it has a broad application in many fields such as video surveillance, content-based image and video retrieval, videoconference, human-computer interface, etc.In this thesis, a study on face tracking in video sequence has been presented. After a literature review of existing algorithms on this particular topic, study was performed along two categories, namely tracking face in terms of object detecing and tracking face in terms of a general object tracking problem.From the view point of object detecting, three face tracking schemes using different features of images were studied. The first scheme employed a skin color model to extract a face region with the assistant of shape information, and accomplished the real-time face tracking using motion prediction model for the following frames. Thersecond scheme used a two-dimensional ellipse model to approximate facial contour. A local search was employed to determine the position and size of the face by maximizing the normalized sum of the image gradient magnitude around the perimeter of the ellipse.The third scheme was based on Adaboost learning algorithm. A rapid face detection system was built using rectangular features and by improved training algorithm. Experimental results have demonstrated the availability of the above schemes, and the cons and pros of them were also analyzed.From view point of object tracking, two commonly used algorithms, namely mean shift tracking and tracking based on particle filter were studied. Furthermore, a combination of the two algorithms was implemented, in which the respective merits were integrated. Exprimental results have shown the effectiveness and the robustness of real time tracking of faces in videos.In this thesis, a software implementation of the face tracking system on PC which has integrated all algorithms studied above was also reported.. And a single face tracking system which can capture, process and play video automatically is also implemented on a DAM6416P platform.