Real-Time Safety Helmet Detection Using Enhanced YOLOv5 Object Detection

Personal Protective Equipment (PPE) regulations require construction workers to wear safety helmets to ensure site safety. However, monitoring PPE compliance consistently in fast-paced, dynamic construction environments poses a significant challenge. In response, we developed a sophisticated object detection system that automates the real-time verification of helmet use, thereby improving safety standards and reducing the likelihood of accidents. Extensive research was conducted to analyze all feasible algorithms that can be implemented in the safety helmet detection system and compare the proposed model with an existing one to ensure the proposed system can give high accuracy and high inference speed. Therefore, YOLOv5 was identified as the ideal choice in terms of accuracy and speed, and it was then enhanced with optimized transfer learning. We began our methodology by pre-training a comprehensive Kaggle dataset before refining the model using Roboflow on a specialized dataset. Using PyTorch and YOLOv5, we conducted exhaustive model training, testing, and evaluation. Our system achieved a lightning-fast inference speed of 39.8 milliseconds and a remarkable 91.4 percent accuracy in identifying helmet compliance. The implementation of such object detection technologies has the potential to significantly increase safety helmet compliance, thereby creating a safer environment for construction workers.