Face feature tracking with automatic initialization and failure recovery

Face feature tracking is a well known and quite challenging area in computer vision. This paper mainly focuses on two important aspects of feature tracking, viz., automatic initialization and automatic detection of tracking failure followed by system update. We present a dynamic framework to automatically initialize and update the face feature tracking process. In addition, a novel approach to self-occlusion handling is also presented. The system consists of - initialization, feature tracking and system update modules. A reliable and efficient technique, that can quickly initialize a face feature tracking system in subject independent manner, has been presented. The initialization module relies on a scale independent accurate feature positioning algorithm based on binarized motion differencing approach. Face feature tracking module is based on the multi-resolution motion tracking algorithm. The system also enables automatic tracking failure detection and re-initialization, with practically minimal subject intervention. In the end, a new technique, to handle the problem of features occlusion, has been proposed. The combined model not only makes the tracking system more efficient and quicker but also helps it to act in a self supervised manner.