跳到主要导航 跳到搜索 跳到主要内容

Time-Dependent Body Gesture Representation for Video Emotion Recognition

  • Xi'an Jiaotong University

科研成果: 书/报告/会议事项章节会议稿件同行评审

6 引用 (Scopus)

摘要

Video emotion recognition has recently become a research hotspot in the field of affective computing. Although large parts of studies focus on facial cues, body gestures are the only available cues in some scenes such as video monitoring systems. In this paper, we propose a body gesture representation method based on body joint movements. To reduce the model complexity and promote the understanding of video emotion, this method uses body joint information to represent body gestures and captures time-dependent relationship of body joints. Furthermore, we propose an attention-based channelwise convolutional neural network (ACCNN) to retain the independent characteristics of each body joint and learn key body gesture features. Experimental results on the multimodal database of Emotional Speech, Video and Gestures (ESVG) demonstrate the effectiveness of the proposed method, and the accuracy of body gesture features is comparable with that of facial features.

源语言英语
主期刊名MultiMedia Modeling - 27th International Conference, MMM 2021, Proceedings
编辑Jakub Lokoc, Tomáš Skopal, Klaus Schoeffmann, Vasileios Mezaris, Xirong Li, Stefanos Vrochidis, Ioannis Patras
出版商Springer Science and Business Media Deutschland GmbH
403-416
页数14
ISBN(印刷版)9783030678319
DOI
出版状态已出版 - 2021
活动27th International Conference on MultiMedia Modeling, MMM 2021 - Prague, 捷克共和国
期限: 22 6月 202124 6月 2021

出版系列

姓名Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
12572 LNCS
ISSN(印刷版)0302-9743
ISSN(电子版)1611-3349

会议

会议27th International Conference on MultiMedia Modeling, MMM 2021
国家/地区捷克共和国
Prague
时期22/06/2124/06/21

学术指纹

探究 'Time-Dependent Body Gesture Representation for Video Emotion Recognition' 的科研主题。它们共同构成独一无二的指纹。

引用此