Automatic Facial Expression Tracking for 4D Range Scans

This paper presents a fully automatic approach of spatio-temporal facial expression tracking for 4D range scans without any manual interventions (such as specifying landmarks). The approach consists of three steps: rigid registration, facial model reconstruction, and facial expression tracking. A Scaling Iterative Closest Points (SICP) algorithm is introduced to compute the optimal rigid registration between a template facial model and a range scan with consideration of the scale problem. A deformable model, physically based on thin shells, is proposed to faithfully reconstruct the facial surface and texture from that range data. And then the reconstructed facial model is used to track facial expressions presented in a sequence of range scans by the deformable model.