Vision/Performance Driven

Facial analysis will undoubtedly play an important role in facial animation control strategies for two principal factors : (1) automatic control parameter extraction, and (2) validation of facial articulations. Deriving facial motion parameters from video images is an area of active research within the vision community (for more details see the NSF report: Facial Expression Understanding [38]). Therefore this section briefly identifies two key approaches used to date for facial animation:

1. Facial animation control has primarily focused on manually produced sequences by artists who carefully craft keyframes. This process is time-consuming and laborious. Therefore, automating facial parameter extraction is highly desirable and has resulted in a number of physical tracking approaches involving head-mounted devices attached to 6 DOF measuring devices [125]. The performance is then captured using multiple sensors and directly applied to facial parameters.

2. An alternative strategy is to derive parameters from real peoples' faces in motion using vision-based techniques. However, the current experience on facial motion tracking, and in particular expression tracking, is limited. Usually the work is restricted to frontal images of the face under suitable illumination. Furthermore, the ability to identify the position, orientation, and scale of the head and facial features in advance of tracking impedes progress.