One of the most reliable face detectors in digital images has been deveopled by Rowley, Baluja and Kanade. It is described in the technical report CMU-CS-95-158R (see Face Detector Demo at the Carnegie Mellon University). Their system recognizes about 90% of all frontal faces while detecting only sporadic non-faces as faces. We have recreated their proposed neural network-based frontal face classification system and use it as a basis for our frontal face detector in video sequences. See how well it works:

Short Demo Video (MP4)