Towards A Robust and Real-time Face Detection and Tracking Framework

Traditional research and development in face detection and tracking are mostly focused on either face detection in still images, or face tracking in image streams, without remarkable combination of both. As a result, few impressive face detection and tracking systems can be seen nowadays. The paper contributes to effectively incorporation of two modules, detection module and tracking module, into one integrated and coupled framework. In the detection module we construct AdaBoost detectors for face detection in video stream, which are able to detect multi-view faces and are robust to illumination changes in video stream. The tracking module is based on the improved mean shift tracking, which has advantage of target model adaption. The detection module, triggered every ten frames, is not only responsible for initiating a tracking thread, but also, together with tracking module, contributes to target model update. The system is able to benefit from the strengths of different techniques and overcome their respective limitations. Experiments on real image sequences show that the system is robust and real-time in realistic environment.