Fast Face Orientation Estimation from an Uncalibrated Monocular Camera

In this paper we present a model based method for fast face orientation estimation and pose recovery from a monocular image sequence captured by an uncalibrated camera. To accomplish this method, CANDIDE-3 model is used to construct 3D individual face model manually. Four facial feature points (outer corners of eyes and mouth) are tracked by SIFT algorithm. Taking advantage of these feature points and 3D individual face model, facial pose recovery can be quickly completed with a Fast Face Orientation Estimation Algorithm. An experimental system has been built to carry out the estimation method. The estimation results are robust and flexible.