Body Movement Based Emotion Recognition

In the era of artificial intelligence,human-computer interaction is so frequent that the need for recognizing human emotions by computers and machine is becoming more and more urgent.At this stage,emotion recognition mainly based on facial expressions and speech signals.However,as an important part of body language,body movements also contain a wealth of emotional information.We take body movement based emotion recognition as our topic.Two related datasets are collected to design our own model to achieve effective recognition.The main contributions of this article are as follows.Firstly,the two large-scale,multi-view datasets we collected is large enough to use data-hungry methods like deep learning.At this stage,the number of public accessible datasets in this field is under ten,and each contains less 10,000 samples so deep learning is rarely used.Secondly,we pioneered use human pose estimation model to get skeleton joint points,and then fuse both RGB feature and skeleton feature to recognize emotion from body movement.It is completely different from other studies in this field to manually extract features and use traditional classifiers to recognize.Our model extracts feature and recognizes emotion with deep learning networks and performs better in emotion recognition.Thirdly,we analyze and compare the results of previous research in this paper and realize that skeleton feature is more effective in emotion recognition.Therefore,we use an attention model to get new skeleton feature and get satisfied improvement of recognition accuracy.Our model achieves effective recognition using only smallscale skeleton data,and it has obvious advantages in the aspects of recognition accuracy and computational efficiency compared with other models.