Application of Depth Camera-based Action Recognition with Graph Convolutional Networks in Elderly Care and Smart Education

Application of Depth Camera-based Action Recognition with Graph Convolutional Networks in Elderly Care and Smart Education

This study explores the innovative applications of depth cameras combined with Graph Convolutional Networks (GCN) for action recognition in two critical domains: elderly care and smart education. We harness the capabilities of depth cameras to capture spatial and temporal features, alongside our robust GCN algorithm, to develop models capable of accurately recognizing and classifying human actions. In elderly care, our model is particularly focused on detecting and analyzing falls, which are crucial for enhancing care safety and supporting the independence of elderly individuals. Experimental results demonstrate that our depth camera-based action recognition model achieved an impressive average accuracy of 96.3% in fall detection within real-world scenarios, while also maintaining low rates of false positives and false negatives. In the realm of smart education, our depth camera-based model is specifically designed to recognize students’ hand-raising actions in real-time, which is crucial for comprehensively assessing student engagement in the class, and accordingly adjusting teaching strategies. Experimental results show that our model achieves an average accuracy of 89.7% in realworld scenarios, while maintaining low rates of false positives and false negatives. Overall, this study showcases the powerful potential of integrating depth cameras with GCNs for action recognition, significantly enhancing both the safety and efficiency of elderly care, as well as the interactivity and educational quality of smart education.

Creators Zhang Qingqi

Creators Wu Ren

Creators Ge Qi-Wei

Award Project on Training of Doctoral Students at Yamaguchi University through Interdisciplinary Research Practice [Crossref Funder] https://doi.org/10.13039/501100001691 JP-MJSP2111