| Autor: |
Siqi Liu, Nan Wu, Haifeng Jin |
| Rok vydání: |
2022 |
| Zdroj: |
Proceeding of 2021 International Conference on Wireless Communications, Networking and Applications ISBN: 9789811924552 |
| Popis: |
A human action recognition network (AE-HRNet) based on high-resolution network (HRNet) and attention mechanism is proposed for the problem that the semantic and location information of human action features are not sufficiently extracted by convolutional networks. Firstly, the channel attention (ECA) module and spatial attention (ESA) module are introduced; on this basis, new base (EABasic) and bottleneck (EANeck) modules are constructed to reduce the computational complexity while obtaining more accurate semantic and location information on the feature map. Experimental results on the MPII and COCO validation sets in the same environment configuration show that AE-HRNet reduces the computational complexity and improves the action recognition accuracy compared to the high-resolution network. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
