Enhacing workplace safety compilance using YOLOV8: real-time detection of safety gear and equipment

Abstract

In this thesis,we explored the design and implementation of an AI-based system for real-time detection of Personal Protective Equipment (PPE) compliance and worker iden- tification on construction sites. Leveraging the YOLOv8 object detection algorithm and deep learning-based face recognition using an Improved ResNet-50 backbone and ArcFace loss, the system effectively identifies safety violations and links them to individual work- ers. Trained on a custom dataset of over 2,500 annotated images, the system achieved 92% precision in PPE detection and 95% accuracy in facial recognition. Real-time alerts are delivered to site supervisors via a Telegram bot, enabling rapid intervention. The sys- tem’s deployment on the NVIDIA Jetson Nano platform ensures low-latency, on-device processing without reliance on cloud services. While environmental factors such as oc- clusion and low lighting posed challenges, post-processing filters and model quantization techniques improved robustness. This work highlights the potential of edge-based AI systems in enhancing safety practices in high-risk industrial environments, fostering a culture of accountability, and reducing human error in compliance monitoring.