A multimodal non-contact safety monitoring system for industrial personnel status detection

To address the intelligent and real-time monitoring demands of personnel status in complex industrial environments, this paper proposes a novel multimodal safety monitoring system that innovatively integrates head pose recognition and mmwave radar based vital sign detection within an edge-deployable framework. The system introduces a rule-driven fusion strategy at the decision level, enabling interpretable and efficient state inference from heterogeneous modalities. The proposed approach supports synchronous processing of visual and radar data on the Jetson AGX Orin edge computing platform, ensuring low-latency and high-reliability operation in dynamic industrial scenarios. Additionally, a 3D-printed mechanical enclosure is designed for robust deployment in real-world environments. Experimental results demonstrate an average behaviour classification accuracy of 91.3% and a stable runtime frame rate of 25 FPS, validating the system’s practical applicability, scalability, and real-time performance.