A Review of Graph Convolutional Networks for Skeleton-based Human Action Recognition

Nan Ma^a, Mengye Zhang^b, Mohan Wang^c, Genbao Xu^d, Chuansheng Xiao^e and Shangyuan Li^f

Faculty of Information and Technology, Beijing University of Technology, China.

^amanan123@bjut.edu.cn

^bzmengye@emails.bjut.edu.cn

^cwangmohan@emails.bjut.edu.cn

^dxugenbao@emails.bjut.edu.cn

^excs_self@emails.bjut.edu.cn

^fs202373050@emails.bjut.edu.cn

ABSTRACT

Human action recognition is one of the most popular research areas. With the advent of deep learning techniques, human action recognition has made significant progress. In the current mainstream data types, using skeleton data for human action recognition has many advantages. Skeleton-based human action recognition algorithms have attracted many researchers' attention. Over the past few years, the graph convolutional networks(GCN) have been demonstrated the powerful functionality in human action recognition. In this review, we give a detailed introduction to the human action recognition data modes, and make a comprehensive overview of GCN-based skeleton action recognition algorithms. Then, we propose a classification of GCN-based improved skeleton action recognition networks from different perspectives, and detailedly study some mainstream algorithms in recent years and make a comparative analysis. Finally, we state the future research focus and discuss the emerging development trend of this field. We put forward the future research direction from the perspective of combining graph convolutional networks with transformer and hypergraph.

Keywords: Human action recognition, Graph convolutional networks, Human skeleton data, Deep learning.

---

Download PDF