| Autor: |
Zang, Tuo, Tu, Jianfeng, Duan, Mengran, Chen, Zhipeng, Cheng, Hao, Jiang, Hanrui, Zhao, Jiahui, Liu, Lingfeng |
| Předmět: |
|
| Zdroj: |
Applied Intelligence; Sep2024, Vol. 54 Issue 17/18, p7621-7635, 15p |
| Abstrakt: |
Human action recognition from skeleton motion sequences is widely applied in various fields such as virtual reality, human-computer interaction and kinematic rehabilitation. With the wide use of graph neural networks for extracting spatial features from skeleton anatomy connectivity, spatial-temporal extension of single graph models on human skeleton may improve the network performance. In this paper, we propose a priori separation graph convolution (PS-GCN) network composition with a priori mixed GCN by introducing a hypergraph representation of the skeleton spatial features, and a dynamic adaptive GCN to describe the respective graph model for each sample at each layer of the network of spatial features. For temporal feature analysis, a global attention unit is added to describe the long-term relationship. Moreover, a feature fusion structure is applied for short-term temporal features in the input of the network. The proposed model is evaluated on the NTU-RGB+D, NTU-RGB+D 120 and NW-UCLA datasets via a comprehensive ablative study. The results show that our model is comparable in accuracy to the state-of-the-art methods. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink