統計的相関を用いて運動と言語の構造を結びつけたデータベースの設計

概要

論文の詳細を見る
The motion capture systems have been improved, and widely used for motion analysis and synthesis in fields of robotics, animation, rehabilitation, and sports engineering. A lot of captured human data have been accumulated. These prerecorded motion data are expected to be reused. The retrieval of a specified motion data is a fundamental technique for the reuse. This paper describes a novel approach to retrieve motion data from word queries out of a large motion dataset. The motion data are trained by Hidden Markov Models, each of which symbolizes a motion pattern. The motion data are also manually given word labels. The mapping between motion symbols and word labels are optimized through canonical correlation analysis so that correlation between them can be maximized. This mapping makes it possible to project a word query to motion features, and to search for motions similar to the motion features. The validity of the proposed approach was demonstrated on captured motion data.