| dc.description.abstract | Motion capture technologies, especially those combined with multiple kinds of sensory technologies to capture both kinematic and dynamic information, are widely used in a variety of fields such as biomechanics, robotics, and health. However, many existing systems suffer from limitations of being intrusive, restrictive, and expensive.
This dissertation explores two aspects of motion capture systems that are low-cost, non-intrusive, high-accuracy, and easy to use for common users, including both full-body kinematics and dynamics capture, and user-specific hand modeling.
More specifically, we present a new method for full-body motion capture that uses input data captured by three depth cameras and a pair of pressure-sensing shoes. Our system is appealing because it is fully automatic and can accurately reconstruct both full-body kinematic and dynamic data. We introduce a highly accurate tracking process that automatically reconstructs 3D skeletal poses using depth data, foot pressure data, and detailed full-body geometry. We also develop an efficient physics-based motion reconstruction algorithm for solving internal joint torques and contact forces based on contact pressure information and 3D poses from the kinematic tracking process.
In addition, we present a novel low-dimensional parametric model for 3D hand modeling and synthesis. We construct a low-dimensional parametric model to compactly represent hand shape variations across individuals and enhance it by adding Linear Blend Skinning (LBS) for pose deformation. We also introduce an efficient iterative approach to learn the parametric model from a large unaligned scan database. Our model is compact, expressive, and produces a natural-looking LBS model for pose deformation, which allows for a variety of applications ranging from user-specific hand modeling to skinning weights transfer and model-based hand tracking. | en |
