Double articulation analyzer for unsegmented human motion using Pitman-Yor language model and infinite hidden Markov model
Tadahiro Taniguchi, Shogo Nagasaka · 2011
We propose an unsupervised double articulation analyzer for human motion data. Double articulation is a two-layered hierarchical structure underlying in natural language, human motion and other natural data produced by human. A double articulation analyzer estimates the hidden structure of observed data by segmenting and chunking target data. We develop a double articulation analyzer by using a sticky hierarchical Dirichlet process HMM (sticky HDP-HMM), a nonparametric Bayesian model, and an unsupervised morphological analysis based on nested Pitman-Yor language model which can chunk given documents without any dictionaries. We conducted an experiment to evaluate this method. The proposed method could extract unit motions from unsegmented human motion data by analyzing hidden double articulation structure.