Surface and Motion Capture for High Fidelity Synthesis of Digital Humans

University Of Washington, Seattle WA

Investigators

Abstract

Proposal #0098005 Popovic, Zoran U of Washington We propose a multi-layered approach to capturing and synthesizing realistic human shapes and motions. To capture the static shape of real humans, we will employ 3D scanning techniques including hierarchical light striping, simultaneous multi-striping, and photometric stereo. A feature-tracked motion capture system as well as 3D scanning techniques will generate motion data. The investigators will acquire this motion data at varying resolutions in order to drive the analysis of skeletal motion, body part deformation such as bulging due to flexing a muscle, and secondary motion such as leg vibrations that occur when stomping on the ground. This wealth of human data will then drive an analysis, modeling, and synthesis stage. The static scan data will be analyzed to construct the space of possible human shapes. This human shape model together with the body part motion capture and full-body motion capture will be used to construct a detailed kinematic model of the human body. Modeling human shape movement at such different levels of detail will allow control of the human motion on the coarse skeletal level while preserving the fine details such as muscle bulging. Furthermore, this multi-layered approach will enable selective replacement of different layers in the human model structure. For example, it will be possible to map the animated movement onto a different body scan and observe a different surface shape movement and creasing. The detailed kinematic human model will be further extended with a model of human dynamics by taking into account a number of physical properties of the human body such as muscle usage and mass distribution. This additional dynamic information provides a way to preserve the realism of motion even when the structure of motion is significantly modified. In addition, the investigators will extend the skeletal dynamic model with secondary motion simulations constructed to replicate the loose skin and tissue vibrations that occur in high-energy movements. The investigators will incorporate their work into new curriculum both at their university and in courses being offered to the professional community. This work will be folded into CDROM's that reach a wide audience, including the general public and a broad spectrum of high school students who may be considering careers in information technology. The results of the research will include complex databases of human shape and motion to be distributed to the general research community in order to encourage further research in this area.