Authors :

Jonathan Jun Hao and Melissa Chua Hui Ling

Address :

School of Computing, National University of Singapore (NUS), 13 Computing Drive, Singapore 117417.

School of Computer Science, Singapore Management University (SMU), 80 Stamford Road, Singapore 178902.

Abstract :

Real-time human gesture recognition plays a vital role in enhancing human-computer interaction, surveillance, and assistive technologies. This study proposes a robust framework that leverages multiscale feature learning and Jellyfish Optimization to improve gesture recognition accuracy and responsiveness. Existing methods often struggle with low recognition accuracy in dynamic or occluded environments due to ineffective modeling of spatial-temporal relationships and fixed feature extraction strategies. To address these limitations, we introduce a novel Spatio-Temporal Graph Attention Network (ST-GAT) optimized using the Jellyfish Optimization Algorithm. ST-GAT models human skeleton joints as graph nodes, capturing spatial and temporal dependencies through attention mechanisms, while Jellyfish Optimization adaptively tunes the attention weights and temporal windows to refine feature learning. The proposed method is applied to real-time surveillance and gesture-controlled systems where high accuracy and fast response are critical. Experimental results demonstrate that our framework significantly outperforms traditional CNN and RNN-based models in both precision and inference speed, even in complex environments involving occlusions and variable gesture speeds. This approach enhances the system's robustness, making it suitable for real-world deployment in safety monitoring and interactive applications. The proposed method gradually improves the recognition accuracy by 98.3%, Inference Speed by 97.4%, precision by 90%, recall by 95%, F1 score by 94.8%, and Optimization Convergence Rate by 96.7%.

Keywords :

Human Gesture Recognition, Spatio-Temporal Graph Attention Network, Jellyfish Optimization, Real-Time Processing, Multiscale Feature Learning, Skeleton-Based Recognition.

DOI :