Integrated Detect-Track Framework for Multi-view Face Detection in Video

An Experiential sampling and Meanshift tracker based Multi-view face detection in video is proposed in this paper. In this framework, instead of performing face detection at every position in a frame, we determine certain key positions to run the multi-view face detectors. These key positions are statistical samples drawn from a density function that is estimated based on color cues, past detection results, Meanshift tracker results and a temporal continuity model. These samples are then propogated using a Particle filter framework. We use a Meanshift tracker to track faces that are missed by the multiview face detectors. Our framework results in a significant reduction in computation time and accounts for the detection of complete 180 degree pose of the face. We also come up with a novel likelihood measure for track termination, which becomes important when used for detection purposes.