“Decomposing Animation into BlendShapes” by Lewis
Conference:
Type(s):
Interest Area:
- Production & Animation
Title:
- Decomposing Animation into BlendShapes
Session/Category Title: Animation Techniques
Presenter(s)/Author(s):
Abstract:
This sketch describes some preliminary work on decomposing animated motion into a set of localized “blendshapes” (also called morph targets and shape interpolation) that can be combined to produce that motion. In computer animation, information traditionally flows in the opposite direction. Predefined blendshapes or other primitives are combined to produce the animated movement. To motivate blendshape decomposition (BSD), we note that several computer-vision techniques allow model streams to be acquired from cameras, and these techniques may become routine in the future. The form of the captured data (a stream of unlabeled meshes, often without frame-to-frame registration) does not lend itself to post-capture editing, however. This leads to the frequent comment that motion capture does not reduce the animator’s workload. Some post-capture editing of the data is presumed nec- essary, but this requires constructing an animator-friendly model (often using blendshapes) and then animating this model to mimic the captured performance. Motion capture can obviously drive an animator-friendly model through some automated process, but producing this model remains an issue. Our work is motivated by this issue.
References:
1. Chen, S. & Donoho, D. L. (1995). Atomic decomposition by basis pursuit. Stanford Dept. Statistics TR 479, May 1995.
2. Paatero, P. (1997). Least squares formulation of robust non-negative factor analysis. Chemometrics and Intelligent Laboratory Systems 37, 23-35.
