Implementation of a Multiple Switch Time Approach to Style-Based Motion Segmentation
Sheng, Yu, Department of Systems Engineering, University of Virginia
LaViers Minnick, Amy, Department of Systems and Information Engineering, University of Virginia
This thesis presents progress on segmenting human movement based on a notion of movement quality. This research is an extension of a style-based motion classification where, here, this classification is used to segment long motion phrases into smaller, discrete motion snippets. In particular, this thesis presents a given trajectory that is segmented into three shorter trajectories that each has their own length and quality. The objective of the thesis is to refine this segmentation, extend it to an arbitrary number of segmentation points, apply it to motion capture data and explore other extensions. A key novel contribution of this thesis is the analytical derivation of first order necessary conditions for optimality. The research may be used to build a library of motion primitives and aid the study of motion recognition in automation.
MS (Master of Science)
inverse optimal control, human motion
All rights reserved (no additional license for public reuse)