Automatic 3D Motion Synthesis with Time-Striding Hidden Markov Model

In this paper we present a new method, time-striding hidden Markov model (TSHMM), to learn from long-term motion for atomic behaviors and the statistical dependencies among them. TSHMM is a 2layer hidden Markov model, which approximates a variable-length hidden Markov model by first-order statistical dependencies. An EM algorithm is proposed to learn the TSHMM.

• Type: Conference
• Year: 2005
• Where: ICMLC
• Authors: Yi Wang, Zhi-Qiang Liu, Li-Zhu Zhou
• Keywords: Hidden Markov Model, ICMLC 2005, Machine Learning, Time-striding Hidden Markov, Variable-length Hidden Markov