Mitigating Road Accidents through Driver Drowsiness Detection using Facial Landmarks

Drowsiness-related Road accidents are becoming more common, highlighting the need for effective driver monitoring systems. This paper introduces a real-time drowsiness detection system using facial landmark technology. It analyzes 478 facial points to calculate important metrics: Eye-Aspect-Ratio (EAR) for eyelid movement and Mouth-Aspect-Ratio (MAR) for tracking yawns. The Mouth-over-Eye (MoE) ratio is then computed to differentiate between normal and fatigued states. A key contribution of this work is the integration of MoE as a major composite feature, which significantly improves detection accuracy by reducing misclassification caused by isolated facial movements like talking or blinking. The study utilizes a dataset of 868 labeled images to train and assess various machine learning models, including Logistic Regression, Support Vector Machine, K-Nearest Neighbors, Decision Tree, Random Forest, and XGBoost. Results showed high accuracy (97.1% to 97.7%), with the Decision Tree being the best choice for practical use, achieving 97.7% accuracy and the fastest response time at just 0.77 milliseconds. The system is practical due to its use of affordable camera hardware and real-time processing. However, performance can vary based on lighting and camera quality. This research demonstrates the effectiveness of combining facial features with machine learning models to improve road safety by providing timely alerts for drowsiness, while addressing the balance between accuracy and computational efficiency. This research contributes to safer driving environments by enabling timely drowsiness alerts through lightweight and accessible technology. Future work will explore the capability of detecting various facial states, improved robustness under diverse real-world conditions and integration with multimodal driver monitoring systems.