High-Order Graph Convolutional Network for Skeleton-Based Human Action Recognition

Abstract

Skeleton-based action recognition plays an important role in the field of human action recognition. Recently, with the introduction of Graph Convolution Network (GCN), GCN has achieved superior performance in the field of skeleton-based human action recognition. In this work, we propose a high-order GCN model. In this model, we introduce the expression of high-order skeletons and establish a new high-order adjacency matrix. Through this matrix, the relationship between skeleton nodes and non-neighbor nodes has being established. In addition, based on the degree of node association of different hierarchical neighborhoods, the value of the matrix expresses the importance of different hierarchies. As a result, the proposed model extracts the co-occurrence feature of the skeleton which is superior to the local features and improves the recognition rate. We evaluate our model on two human skeleton action datasets, Kinetics-skeleton and NTU RGB+D, and then further explore the influence of skeleton nodes based on different hierarchies on the recognition results.