Real-Time Driver Drowsiness Detection Using Facial Analysis and Machine Learning Techniques

Area of Science:

• Computer Science
• Artificial Intelligence
• Road Safety Engineering

Background:

• Drowsy driving is a major global road safety concern, causing numerous accidents and fatalities.
• Existing driver drowsiness detection (DDD) systems often suffer from intrusiveness and slow response times.

Purpose of the Study:

• To develop and evaluate a real-time, non-intrusive driver drowsiness detection system.
• To systematically assess various machine and deep learning algorithms for DDD performance.

Main Methods:

• Utilized facial analysis and machine learning techniques for drowsiness detection.
• Evaluated K-Nearest Neighbors (KNNs), Support Vector Machines (SVMs), Convolutional Neural Networks (CNNs), and advanced Computer Vision (CV) models (YOLOv5, YOLOv8, Faster R-CNN).
• Tested algorithms on NTHUDDD, YawDD, and UTA-RLDD public datasets.

Main Results:

• KNNs achieved 98.89% accuracy, 99.27% precision, and 98.86% F1 score on UTA-RLDD.
• YOLOv5 and YOLOv8 demonstrated 100% precision and recall with 99.5% mAP@0.5 on UTA-RLDD.
• Faster R-CNN showed lower performance with 81.0% accuracy and 63.4% precision on UTA-RLDD.

Conclusions:

• Advanced CV models, particularly YOLOv5 and YOLOv8, show significant promise for highly accurate, real-time driver drowsiness detection.
• The developed system has the potential to substantially improve road safety through proactive, real-time alerts.