| Title |
Contrastive Graph Convolutional Network for Skeleton-based Human Action Recognition |
| Authors |
이경현(GyungHyun Lee) ; 고병철(Byoung Chul Ko) ; 남재열(JaeYeal Nam) |
| DOI |
https://doi.org/10.5573/ieie.2022.59.11.90 |
| Keywords |
Skeleton-based action recognition; Skeleton graph augmentation; Contrastive learning; Semi-supervised learning; Graph convolutional network |
| Abstract |
Human action recognition studies require numerous data to be learned to achieve generalized performance, as they can have various structures depending on the size variation or occlusion of the object. Therefore, in this paper, we propose a contrastive graph convolutional network using graph data augmentation and contrast learning. We train various postures to model by applying data augmentation and contrastive learning. Then we increase the accuracy of recognition with learning relations of pose that change over time. The proposed model can achieve high generalized performance even with a lower amount of data and parameters, as well as provide high computational efficiency. We demonstrate the superior performance of our model via comparative experiments with state-of-the-art methods. |