Automatic Labanotation Generation from Motion-Captured Data Based on Hidden Markov Models

Abstract

Labanotation is a powerful tool for the recording and archiving of traditional dances. In this paper, we propose a Hidden Markov Model based method to automatically generate Labanotation from motion-captured data by recognizing each category of body movements that corresponds to a Labanotation symbol. The body movements across frames are modeled with Hidden Markov state and each state is modeled with a mixture of Gaussian models. Furthermore, we extract better features from motion-captured data that are more conducive to modeling movement segments with Hidden Markov Models. Therefore, our model is able to generate much more reliable Labanotation records than previous works. In our experiments, We achieve an accuracy of about 90\% for the generated notations in the support column of Labanotation.