Dual-branch attention-based multi-joint data fusion for enhanced action recognition in IoT-enabled environments

To address the challenges of multi-joint data feature extraction and fusion, especially in the Internet of Things (IoT) environment, this paper proposes a multi-joint action recognition model AttnFusionGRU based on a dual-branch attention mechanism. Traditional action recognition methods have limitations in capturing dynamic changes and key joint information. AttnFusionGRU uses a dual-branch attention mechanism to focus on key features in spatial and temporal dimensions respectively, combined with a lightweight SqueezeNet module and bidirectional GRU (BiGRU), to achieve efficient and accurate action recognition. Combined with IoT devices, this model can process a large amount of data from multi-source sensors in real time and with limited resources. Experiments on UCF101 and Human3.6M datasets show that the proposed model outperforms existing baseline models in Top-1, Top-5 accuracy and mean per-frame accuracy (MPFA), demonstrating good generalization and robustness. Ablation experiments further verify the key role of each module in performance improvement. Despite challenges such as computational overhead and environmental adaptability, this model has broad application prospects in IoT-driven fields such as intelligent monitoring, motion analysis, and medical rehabilitation, and provides an effective solution for multi-joint motion recognition, especially suitable for deployment in edge computing environments.