Skeleton-based online action recognition using inertial sensors in industrial scenarios

Online action recognition is a crucial in industrial settings to boost productivity, enhance safety, and streamline automation. This research focuses on leveraging inertial sensors attached to the body to generate precise skeleton representations, facilitating the real-time recognition of complex actions performed in industrial environments. The primary goal is to accurately distinguish between similar movements associated with different tasks. To address this challenge, we employ the InfoGCN++ neural network, known for its effectiveness in online skeleton-based action recognition. Our experiments in simulated industrial scenario demonstrate that this approach not only provides high accuracy in distinguishing between similar actions but also maintains robust real-time performance. The results highlight the potential of integrating inertial sensors with advanced neural networks to improve movement recognition and classification. This study contributes to human action recognition by offering a practical solution for real-time action monitoring in industrial settings.