Attention module-based spatial–temporal graph convolutional networks for skeleton-based action recognition

Abstract

Skeleton-based action recognition is a significant direction of human action recognition, because the 9 skeleton contains important information for recognizing action. The spatial temporal graph convolutional networks 10 (ST-GCN) automatically learn both the temporal and spatial features from the skeleton data, and achieve remarkable 11 performance for skeleton-based action recognition. However, ST-GCN just learn local information on a certain 12 neighborhood, but does not capture the correlation information between all joints (i.e., global information). 13Therefore, we need to introduce global information into the spatial temporal graph convolutional networks. In this 14 work, we propose a model of dynamic skeletons called attention module-based Spatial Temporal Graph 15 Convolutional Networks (AM-STGCN), which solves these problems by adding attention module. The attention 16 module can capture some global information, which brings stronger expressive power and generalization capability. 17Experimental results on two large-scale datasets, Kinetics and NTU-RGB+D, demonstrate that our model achieves 18 significant improvements over previous representative methods. 19 20

Geometric Deep Neural Network using Rigid and Non-Rigid Transformations for Human Action Recognition

“…Presently, deep learning methods for human action recognition are preferred over traditional skeleton-based ones, which tend to focus on extracting hand crafted features [15,39]. The former methods can be categorized into three major sets: methods based on Recurrent Neural Network (RNN) [19], methods based on Convolutional Neural Network (CNN) [7], and methods based on Graph Convolutional Network (GCN) [17].…”

Risk Prediction of Theft Crimes in Urban Communities: An Integrated Model of LSTM and ST-GCN

“…ST-GCN was quickly applied to motion recognition and traffic. For example, Kong Y et al [36] built a dynamic skeleton model based on ST-GCN, combined with the attention module; Geng X et al [37] proposed a spatial-temporal multigraph convolutional network based on ST-GCN (ST-MGCN) to forecast the demand for rides. ST-GCN has a good prediction effect on the transition of crimes in a topological space, but it attaches less importance to timing changes on a single node.…”

KShapeNet: Riemannian network on Kendall shape space for Skeleton based Action Recognition

“…Presently, deep learning methods for human action recognition are preferred over traditional skeleton-based ones, which tend to focus on extracting hand crafted features [17,39] The former methods can be categorized into three major sets: methods based on Recurrent Neural Network (RNN) [21], methods based on Convolutional Neural Network (CNN) [8], and methods based on Graph Convolutional Network (GCN) [20].…”