Drowsiness Detection Using Convolutional Neural Network with Eye Feature Analysis

Driver fatigue represents a critical yet often overlooked factor in road safety. This research presents a novel three-stage machine learning approach to detect and prevent driver drowsiness through advanced computer vision techniques. The proposed system integrates three sequential models: (1) an object detection model to identify the driver, (2) a facial landmark model to precisely locate eye regions, and (3) a Convolutional Neural Network (CNN) to analyze eye states and detect potential drowsiness. The CNN model, trained on 48,000 grayscale images of open and closed eyes, achieved an impressive 99.03% accuracy after fine-tuning with images containing glasses. The system operates through a dashboard-mounted camera, continuously tracking facial expressions and triggering warning mechanisms such as hazard lights, gentle braking, horn activation, and steering adjustments when prolonged eye closure is detected. By offering a proactive and scalable solution, this research contributes to the broader goals of reducing road fatalities and promoting safer driving practices. The system is publicly accessible via a Streamlit web application, allowing users to test individual models and the complete drowsiness detection pipeline, demonstrating both the practical implementation and potential real-world application of the proposed technology.