Robust face detection and tracking under natural conditions

This paper proposes a new method to find face in real-time videos by combining detection and tracking. Basically, the method contains two complementary modules: detection by Viola Jones method and tracking by correlation filters. Detection in current frame is independent to previous frames, but the performance may downgrade under harsh conditions in terms of lighting, rotation, occlusion and others. Contrarily, the filter tracker is capable of handling such harsh factors at a high speed, but its drawback is that the tracker heavily relies on previous frames and often drifts to the background. The proposed method combines merits of the aforementioned two techniques and could achieve promising face finding performance. In addition, this paper introduces a generalized correlation filter named Generalized Unconstrained Filters (GUF), which works well for online tracking.