Edge and Node Graph Convolutional Neural Network for Human Action Recognition
Gang Li, Shengjie Yang, Jianxun Li · 2020
The skeleton-based method in the task of human action recognition has been a research hotspot in recent years. Graph Convolutional Neural Networks (GCNs) are often utilized in extracting the spatial features of skeleton joints data. However, Skeleton joints, the product of pose estimation algorithms, cover only part of spatial information. Skeleton edges, which are in the representation of orientations and center coordinates of bones, also contain critical information. Our work explores how skeleton edges and skeleton nodes work with each other. Furthermore, we design a two-stream network with a specially designed fusion mechanism, named Edge and Node Graph Convolutional Neural Network (EN-GCN). Experimental results on the NTU-RGB+D large-scale dataset confirm the superiority of our model.