Research Abstracts - 2006

Fast Accurate Skinning for 3-D Characters

Robert Wang, Kari Pulli & Jovan Popovic

Controlling realistic human avatars appears in many aspects of computer graphics, from movie production to computer gaming. Usually, this is done by controlling the joints of a skeleton. The skeleton in turn deforms a surface mesh, or skin. This mapping from skeletal poses to skins is referred to as the process of "skinning."

Previous work has attacked skinning from two fronts. On one hand, physical simulations have attempted to reproduce the exact properties of muscles, tendons and ligaments to relate them to animator-controlled bones. On the other hand, approximate techniques such as skeletal subspace deformation (SSD), also known as single-weight enveloping, have been used in video games to quickly apply skeletal controls to a surface mesh. The former technique, currently found in medical simulations and movie studios, is not yet computationally feasible for video games. The latter technique, while popular in video games, suffers from gross approximation errors at extreme poses (see figure). The limitations of SSD have prevented its use in high-quality animation work and are becoming increasingly obvious in today's highly detailed video games. We are developing a technique that gets the best of both worlds. Our technique hopes to combine the speed of SSD while approximating more sophisticated models generated by physical simulations or by hand.

References:

[1] A. Mohr and M. Gleicher. Buiding efficient, accurate skins from examples In ACM Transactions on Graphics, 22(3), pp. 562--568, 2003.