A unified probabilistic approach to face detection and tracking

In this paper, we present a novel probabilistic approach to detecting and tracking human faces in video sequences. Specifically, with the use of multimodal observations a graphical chain model integrating smooth regularization is proposed. Using this model, the recovery of faces that are not detected or detected with only low confidence by a frame-based detector can be formulated as an overall optimization problem constrained by a set of reference faces, which are initially detected with high certainties. Experimental results demonstrate the effectiveness of our approach.