3D Morphable model parameter estimation

Abstract

Estimating the structure of the human face is a long studied and difficult task. In this paper we present a new method for estimating facial structure from only a minimal number of salient feature points across a video sequence. The presented method uses both an Extended Kalman Filter (EKF) and a Kalman Filter (KF) to regress 3D Morphable Model (3DMM) shape parameters and solve 3D pose using a simplified camera model. A linear method for initializing the recursive pose filter is provided. The convergence properties of the method are then evaluated using synthetic data. Finally, using the same synthetic data the method is demonstrated for both single image shape recovery and shape recovery across a sequence. © Springer-Verlag Berlin Heidelberg 2006.