Vision-Based Driver Fatigue Detection Using Convolutional Neural Networks and Behavioral Metrics

Driver drowsiness is a major cause of road accidents, especially during long-distance or late-night driving.
This paper presents a real-time Driver Drowsiness Detection and Alert System that uses a webcam to monitor the
driver’s facial behavior. YOLO (You Only Look Once) is employed for accurate face detection, and Convolutional
Neural Networks (CNNs) analyze facial features. Two key metrics, Eye Aspect Ratio (EAR) and Mouth Aspect Ratio
(MAR), are calculated using Dlib and OpenCV to detect prolonged eye closure and yawning. When thresholds are
crossed, an alarm is triggered to alert the driver. The system also logs EAR and MAR values with timestamps in a
CSV file and generates visual plots for analysis. A user interface developed with Flask and Tkinter provides control
and ease of use. Experimental results show more than 90% detection accuracy under normal conditions with an alert
delay of ~1–2 seconds. This work demonstrates an effective blend of computer vision and deep learning to enhance
driver safety and reduce fatigue-related accidents.