关键词: 动作识别, 开放环境, 时空图卷积网络, 骨架序列, 深度学习

Abstract: With the rapid development of human-robot collaboration technology,human action recognition faces several challenges in complex environments.Existing action recognition methods are mainly optimized for closed set scenarios and lack effective recognition of unknown categories in open set environments.To address this challenge,an open set action recognition algorithm based on higher order spatial-temporal self-attention mechanism was proposed,which integrated Spatial Multi-Head Self Attention(S-MHSA)and Temporal Multi-Head Self Attention(T-MHSA)modules for enhancing the model's ability to capture long-term dependencies and global contextual information.On the other hand,the OpenMax layer was introduced to compute the distance distribution between the sample and the mean activation vector,which enabled the algorithm to distinguish between known and unknown categories,improving recognition accuracy in open environments.Extensive experiments were conducted on both public and custom datasets.The results showed that the proposed algorithm outperformed the original algorithm in terms of accuracy and open set recognition capability.Furthermore,its practical application value was validated in real-world scenarios.

Key words: action recognition, open environment, spatiotemporal graph convolutional network, skeleton sequence, deep learning