Non-Rigid Modeling of Body Segments for Improved Skeletal Motion Estimation

Abstract

A necessary requirement for many musculoskeletal modeling tasks is an estimation of skeletal motion from observations of the surface of a body segment. The skeletal motion may be used directly for inverse kinematic calculations or as an observation sequence for forward dynamic simulations. This paper describes a fundamentally new approach to human motion capture for biomechanical analysis. Techniques for generating three-dimensional models of human skeletal elements from magnetic resonance imaging data are described, along with a methodology for corresponding these highresolution internal models to externally observable features. A system for generating dynamic visualizations of these skeletal models from retro-reflective, skin-mounted marker motion capture data is also developed. Next, a set of techniques for estimating body segment shape and pose without the need for retro-reflective markers, from single and multiple, calibrated and un-calibrated cameras is developed. Example results from both synthetic and actual data sequences are presented.