Performance Driven Facial Animation using Blendshape Interpolation

This paper describes a method of creating facial animation using a combination of motion capture data and blendshape interpolation. An animator can design a character as usual, but use motion capture data to drive facial animation, rather than animate by hand. The method is effective even when the motion capture actor and the target model have quite different shapes. The process consists of several stages. First, computer vision techniques are used to track the facial features of a talking actress in a video recording. Given the tracking data, our system automatically discovers a compact set of key-shapes that model the characteristic motion variations. Next, the facial tracking data is decomposed into a weighted combination of the key shape set. Finally, the user creates corresponding target key shapes for an animated face model. A new facial animation is produced by using the same weights as recovered during facial decomposition, and interpolated with the new key shapes created by the user. The resulting facial animation resembles the facial motion in the video recording, while the user has complete control over the appearance of the new face.