Smart Construction Safety and Alert System with Enhanced AI-Based Optimization

Construction sites are inherently dangerous environments, often resulting in injuries and losses due to lapses in Personal Protective Equipment (PPE) compliance. This paper presents an AI-driven smart safety system designed to automate real-time monitoring of construction workers using computer vision. A comparative evaluation of multiple object detection models—including YOLOv7, YOLOv8, YOLOv9, YOLOv11 variants, and Faster R-CNN—was conducted to identify the best solution. YOLOv11s was selected for its lightweight architecture, excellent precision $(0.908)$, high $\text{mAP}(0.847)$, and efficient inference speed, making it highly suitable for real-time applications. The model was trained on a merged dataset of over 4,000 images across ten PPE categories, using mosaic augmentation and hyperparameter tuning to improve performance and reduce overfitting. The system incorporates a smart alert mechanism that automatically sends email notifications when PPE violations continue for more than 10 seconds, enabling timely intervention. A WebSocket-enabled backend ensures low-latency video streaming and seamless edge device deployment.