講演抄録/キーワード |
講演名 |
2020-10-10 10:30
TED Talksにおけるジェスチャーの分析と分類方法の検討 ○手嶋仁志(九大)・和家尚希(マイクロソフト)・ディエゴ トマ(九大)・中島悠太(阪大)・川崎 洋(九大)・池内克史(マイクロソフト) PRMU2020-35 |
抄録 |
(和) |
(まだ登録されていません) |
(英) |
Most of automatic gesture generation methods have focused on generating beat gestures only. Among these methods, many generate gestures from audio data, but only a few consider text data as additional input to generate meaningful gestures. Our intuition is that text data contains many semantic information that is strongly correlated to the type of gesture that people use in their speech. Using text data is therefore a powerful way to generate many kind of gestures and mimic human behavior. To enable such gesture generation, a large amount of data with annotated gesture type labels on the video is essential to investigate the correlation between text and gestures. In this paper, we propose a new method to automatically detect and classify the gestures of a speaker and their types from a video with text data. We study speech data from TED Talks and introduce a new dataset with annotated gestures. We classify the gestures into three categories: Imagistic Gesture that complements the content of the text, Beat Gesture that has nothing to do with the content of the text, and Deictic Gesture that points to something. |
キーワード |
(和) |
/ / / / / / / |
(英) |
Gesture Classification / Gesture Generation / Gesture Data / / / / / |
文献情報 |
信学技報, vol. 120, no. 187, PRMU2020-35, pp. 104-109, 2020年10月. |
資料番号 |
PRMU2020-35 |
発行日 |
2020-10-02 (PRMU) |
ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
PRMU2020-35 |