Real-Time Deep Learning-Based Drowsiness Detection: Leveraging Computer-Vision and Eye-Blink Analyses for Enhanced Road Safety

Abstract

Driver drowsiness is one of the foremost causes of road traffic accidents because it reduces attentiveness, delays reaction time, and impairs decision-making. This paper presents a real-time, deep learning-based driver drowsiness detection system that combines convolutional neural network (CNN) image classification with facial-landmark-based geometric analysis to improve road safety. A custom image dataset comprising four driver states — Open, Closed, Yawn, and No_Yawn — was collected and used to train a MobileNetV2-based CNN through transfer learning. In parallel, the MediaPipe Face Mesh framework extracts facial landmarks to compute the Eye Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR), enabling continuous monitoring of eye closure and yawning behaviour. The fusion of imagebased classification with geometric feature analysis improves detection robustness across varying illumination, head pose, and driving conditions. The proposed MobileNetV2-CNN model is benchmarked against an existing Random Forest (RF) classifier that relies solely on handcrafted EAR/MAR features. Experimental results show that the proposed model achieves an accuracy of 99.19%, precision of 99.22%, recall of 99.19%, and an F1-score of 99.19%, substantially outperforming the existing RF baseline, which attained an accuracy of only 83.01%. These results confirm that combining deep-learningbased visual recognition with geometric eyeblink analysis provides a reliable, noninvasive, and computationally efficient solution for real-time driver monitoring that can be integrated into Advanced Driver Assistance Systems (ADAS) to reduce fatiguerelated accidents.