Keypose and Style Analysis Based on Low-dimensional Representation

• Item type: SIG Technical Reports(1)
• 公開日: 2009-06-02
• タイトル: Keypose and Style Analysis Based on Low-dimensional Representation
• タイトル / 言語: en
• タイトル: Keypose and Style Analysis Based on Low-dimensional Representation
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: D論セッション1
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: The University of Tokyo
• 著者所属: The University of Tokyo
• 著者所属: The University of Tokyo
• 著者所属(英): en
• 著者所属(英): The University of Tokyo
• 著者所属(英): en
• 著者所属(英): The University of Tokyo
• 著者所属(英): en
• 著者所属(英): The University of Tokyo
• 著者名: Manoj, Perera; Shunsuke, Kudoh; Katsushi, Ikeuchi
• 著者名(英): Manoj, Perera; Shunsuke, Kudoh; Katsushi, Ikeuchi
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Human motion analysis is a complex but extremely interesting and important research area in computer graphics and computer vision. This paper addresses three vital topics in human motion analysis related to keyposes: how to extract keyposes from dance motions, how to utilize them, and how to recognize the person and task that constitute a keypose. As our first topic, we propose a new method to extract keyposes from a given dance using the energy flow of the motion. Our experimental results and comparison with a previous keypose extraction approach show the high accuracy of keypose extraction with our new method. As our second topic, we propose a new method to reconstruct low-dimensional motion based on keyposes, and we illustrate the effect of keyposes in a given motion space on human perception. We utilize the keyposes extracted with our new method, formulate a model, and derive a low-dimensional motion based on our model. We also construct low-dimensional motion using uniform sampling poses, and we compare the results with those obtained from our method. As our third topic, we propose a novel approach to decompose motion into common and individual factors using the Multi Factor Tensor (MFT) model. By this method, we recognize person and task from the motion sequence.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Human motion analysis is a complex but extremely interesting and important research area in computer graphics and computer vision. This paper addresses three vital topics in human motion analysis related to keyposes: how to extract keyposes from dance motions, how to utilize them, and how to recognize the person and task that constitute a keypose. As our first topic, we propose a new method to extract keyposes from a given dance using the energy flow of the motion. Our experimental results and comparison with a previous keypose extraction approach show the high accuracy of keypose extraction with our new method. As our second topic, we propose a new method to reconstruct low-dimensional motion based on keyposes, and we illustrate the effect of keyposes in a given motion space on human perception. We utilize the keyposes extracted with our new method, formulate a model, and derive a low-dimensional motion based on our model. We also construct low-dimensional motion using uniform sampling poses, and we compare the results with those obtained from our method. As our third topic, we propose a novel approach to decompose motion into common and individual factors using the Multi Factor Tensor (MFT) model. By this method, we recognize person and task from the motion sequence.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA11131797
• 書誌情報: 研究報告コンピュータビジョンとイメージメディア（CVIM）; 巻 2009-CVIM-167, 号 2, p. 1-16, 発行日 2009-06-02
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会