Abstract
Recently, motion capture has become one of the most straightforward and efficient solutions to realistic motion synthesis. In this talk, we will present an efficient approach to capturing dense facial motion parameters from multi-view video. For automatic tracking, spatial proximity of facial surfaces and temporal coherence are utilized to find the best trajectories and rectify missing and false tracking.
Furthermore, we will introduce an example-based motion synthesis technique to reuse motion capture data. Based on a pre-computed motion transition graph, the proposed method can rapidly and smoothly transit between captured motions or even parametrically synthesized ones.