Retrieval and Generation of Human Motions Based on Associative Model between Motion Symbols and Motion Labels

Abstract

A large motion data has been stored by using a motion capture system such that humanoid robots or CG characters can perform human-like behaviors. However prerecorded data is not reused efficiently since it is difficult to retrieve a specified motion data from a large dataset, and to modify the motion data to fit desired motion patterns. We have studied an imitative learning model based on symbolization of motion patterns using Hidden Markov Models (HMMs), where each HMM("motion symbol") abstracts dynamics of some motion patterns and can be used for motion recognition and generation. So in this paper, we propose an novel framework to retrieve and generate of human motion data, which consists of original motion data, motion symbols and motion words. Each motion dataset is labeled with motion symbols. Moreover an association between motion symbols and motion words is stochastically formed. The association makes it possible to derive motion symbols from motion words and to search for motion datasets using the motion symbols. The motion symbols can also generate motion data. Therefore the framework can provide the desired motion data when only the motion words are input.