Real-Time Driver Vigilance Assessment Using Yolo and Deep Learning

Driver fatigue remains one of the leading contributors to road accidents, yet most commercially available monitoring systems fail to provide reliable real-time assessment of a driver’s alertness. This work proposes a fully integrated vigilance detection framework that combines a custom-trained Ultralytics YOLO model with a temporal behavioural analysis layer to identify early signs of drowsiness, yawning, and reduced attention. The system processes live video streams using an optimized detection pipeline that includes brightness normalization, real-time state sampling, and adaptive thresholding to ensure robustness under varying illumination and head-pose variations. A multi-modal alert mechanism— featuring visual flash cues, audio warnings, and automated emergency notifications via WhatsApp—enables timely intervention during high-risk situations. The solution is deployed as a multi-threaded Python application with built-in analytics, trip log generation, and session reporting for post- drive evaluation. Experimental results demonstrate reliable real-time performance, achieving high detection accuracy and low latency, making the proposed system a practical, low-cost safety enhancement for intelligent transportation and driver assistance applications.