以轮廓为对象的体态特征情绪分类与预测

采用摄像机对人体行为进行采集，基于OpenCV库对人体行为进行特征提取，在已有学者研究的基础上，拓展所采集低层运动特征，引入PAD情绪量表对个体进行评估.采用中国民族音乐作为情绪诱发素材，对10名受试者进行正、中、负性情绪诱发，受试者通过外显行为将自身情绪自然表达出来；通过PAD情绪量表评估每段诱发素材的情绪状态，视频采集该情绪状态下受试者的行为；采用轮廓法提取视频中个体行为的体态特征，主要包括个体轮廓相对面积及变化特征；结合情绪量表的评估结果，采用支持向量机对个体体态特征所蕴含的情绪进行分类学习，实现体态特征的情绪预测.结果实现了体态特征的情绪分类，且以正、负性情绪为类别进行体态特征分类的情绪预测准确率达到76.92%.
Abstract: The features of the human behaviors were collected by camera and extracted with OpenCV in order to predict the emotions of individuals with the help of PAD scales based on previous studies. Ten subjects were requested to express their emotions through body movements, which were induced by Chinese folk music involving positive, middle and negative emotions. The states of subjects' emotions were evaluated by PAD scales while the behaviors of subjects were collected by camera. The features of individual behaviors were extracted with the characteristic of silhouette, including the relative areas of the individual silhouettes and their rate of change. The emotions of behaviors were classified by support vector machine with the evaluation of emotional scales in order to realize the emotional prediction of the behavioral features. The results realized the emotional classification of behaviors, and the prediction accuracy rate came to 76.92% with the categories of positive and negative emotions.