| This thesis presents new algorithms for face detection and head pose estimation in the domain of multi-media applications.; A new feature is developed based on motion contour analysis. Complemented by color and other facial features, it is employed to detect and locate faces of interest from video sequences. A hybrid real-time face tracker is presented in which initial talker locations are estimated acoustically from microphone array data. Multiple faces can be identified and distinguished from other objects with similar shapes and colors. The system is robust to nonlinear source motions, complex backgrounds, unsatisfactory lighting conditions, and a variety of source-camera depths.; By relying on a facial criterion that is easily extracted from video images acquired across a range of lighting and zooming conditions, an initial head pose estimation algorithm is proposed which is capable of accurately evaluating head orientations over a complete 360 degree interval.; A generic classification algorithm is then presented in which discriminative features are selected to boost Fisher classifiers. The new classifier is applied on pose estimation so that no heuristic features are required in determine the pose orientations. When training on face images under various lighting conditions, the illumination effect is minimized.; While the direct focus of this work is automated video conferencing, the proposed techniques have utility to many multimedia and virtual reality applications. |
