高级检索
付心仪, 薛程, 李希, 张玥泽, 蔡天阳. 基于姿态的情感计算综述[J]. 计算机辅助设计与图形学学报, 2020, 32(7): 1052-1061. DOI: 10.3724/SP.J.1089.2020.18350.z43
引用本文: 付心仪, 薛程, 李希, 张玥泽, 蔡天阳. 基于姿态的情感计算综述[J]. 计算机辅助设计与图形学学报, 2020, 32(7): 1052-1061. DOI: 10.3724/SP.J.1089.2020.18350.z43
Fu Xinyi, Xue Cheng, Li Xi, Zhang Yueze, Cai Tianyang. A Review of Body Gesture Based Affective Computing[J]. Journal of Computer-Aided Design & Computer Graphics, 2020, 32(7): 1052-1061. DOI: 10.3724/SP.J.1089.2020.18350.z43
Citation: Fu Xinyi, Xue Cheng, Li Xi, Zhang Yueze, Cai Tianyang. A Review of Body Gesture Based Affective Computing[J]. Journal of Computer-Aided Design & Computer Graphics, 2020, 32(7): 1052-1061. DOI: 10.3724/SP.J.1089.2020.18350.z43

基于姿态的情感计算综述

A Review of Body Gesture Based Affective Computing

  • 摘要: 情感计算的理论与算法研究是近年来人机交互领域的热点话题.当前,常见的情感计算集中在基于面部表情、语音、文本、人体姿态等方向,既有单一模态的算法,又有多模态的综合算法.基于面部表情和语音模态的算法占据多数,国内外基于人体姿态的算法相对较少.文中针对基于姿态的情感计算所面临的几个关键科学问题展开了综述,包括情感的心理学模型、人体姿态估计算法、姿态的情感特征提取算法、情感分类与标注算法、姿态情感数据集、基于姿态的情感识别算法等.具体来说,首先介绍了几种常用的情感计算心理学模型,评述了各类模型的适用场景;随后从人体检测和姿态估计2个角度对人体姿态估计的常用算法进行了总结,并讨论了2D和3D姿态估计的应用前景.针对特征提取算法,分析了基于全身和上半身身体动作的姿态特征提取算法.在情感标注方面,介绍了表演数据和非表演数据的情感标注算法,并指出了半自动或自动的标注非表演数据将是未来的重要发展趋势之一.针对姿态情感数据集,列举了近年来常见的14个数据集,并主要从是否是表演数据、数据维度、静态或动态姿势、全身或非全身数据等几个方面进行了总结.在基于姿态的情感识别算法方面,主要介绍了基于人工神经网络的情感识别算法,指出了不同算法的优劣之处和适用的数据集类型.文中的综述研究,总结提炼了国内外该领域经典且前沿的工作,希望为相关的研究者提供研究帮助.

     

    Abstract: Research on the theory and method of affective computing has been a hot topic in the field of human-computer interaction in recent years.At present,the common research on affective computing in related fields focuses on facial expression,speech,text,human gesture,and other directions.There are both single-modality research and multi-modality comprehensive research.Among them,the researches based on facial expressions and speech modalities are the majority,and the research based on human gesture is relatively few.In this paper,we conduct a research survey on several key problems faced by gesture-based affective computing,including the emotional psychological model,human pose estimation,body emotional feature extraction method,emotion classification and labeling method,gesture-emotion dataset,and gesture-based emotion recognition algorithm.Specifically,we first introduce several commonly used emotional computing psychology models,review the applications of various models.Then we summarize the common methods of human pose estimation from two perspectives of human detection and pose estimation,and discuss the application prospects of 2 D and 3 D pose estimation.For feature extraction methods,we analyze feature extraction methods based on body movements of the whole body and the upper body.In the aspect of emotion annotation,we introduce the emotion annotation methods of performance data and non-performance data.We also point out that semi-automatic or automatic labeling of non-performance data would be one of the important development trends in the future.For the posture and emotion datasets,we list 14 most commonly used datasets in recent years classified by performance or non-performance data,data dimensions,static or dynamic poses,full-body or non-full-body data.In terms of gesture-based emotion recognition algorithms,we mainly introduce emotion recognition algorithms based on artificial neural networks,pointing out the advantages and disadvantages of different methods and their applicable datasets.This article reviews and summarizes the classic and cutting-edge research work in related fields,hoping to provide a good research basis for researchers in similar directions.

     

/

返回文章
返回