An intelligent edge protection system for construction sites based on improved YOLOv11 and multi-sensor fusion

Abstract

The frequent occurrence of safety accidents on construction sites highlights the urgent need for intelligent Personal Protective Equipment (PPE) monitoring. However, existing deep learning solutions often rely on cloud computing, suffering from high latency, and are susceptible to false alarms in complex environments due to their reliance on purely visual data. To address these challenges, this paper proposes a real-time, highly robust edge computing protection system that fuses the latest YOLOv11 algorithm with multi-source environmental sensors. Specifically, a lightweight YOLOv11s model is trained on a custom PPE dataset and deployed on an NVIDIA Jetson Orin Nano edge node. By leveraging NVIDIA TensorRT for engine-level acceleration, the model achieves an optimal balance between accuracy and computational efficiency. Furthermore, a multimodal fusion warning mechanism is introduced to cross-validate visual detection results with real-time temperature and TVOC gas sensor data, thereby triggering hardware countermeasures (e.g., relays and exhaust fans) strictly under verified high-risk conditions. Experimental results demonstrate that the proposed YOLOv11s model achieves a mean Average Precision (mAP@0.5) of 92.3% and a real-time inference speed of 35 Frames Per Second (FPS) on the edge device. The multi-sensor fusion significantly reduces visual false positives, providing a highly reliable, low-latency, and deployable closed-loop safety paradigm for smart construction environments.