Maximum-likelihood motion estimation of a human face

An algorithm for estimating the three-dimensional motion of a human face from a monocular image sequence is investigated. For motion estimation, the shape of a human face is described by a three-dimensional rigid triangular mesh and its motion by six parameters: one three-dimensional translation vector and three rotation angles. The motion parameters are estimated by maximizing the conditional probability of the frame to frame intensity differences at observation points. The conditional probability is a function of the motion parameters, the frame to frame intensity differences and the covariance matrix of the intensity error at the observation points. The intensity error is supposed to be the result of the camera noise and the position error attributed to the shape estimation errors and the motion estimation errors occurred by the motion analysis of previous frames. The algorithm was applied to different real image sequences depicting a moving human face with very encouraging results.