Real-Time Drowsiness Detection and Classification with Deep Learning Model

Abstract

Drowsy driving is a major concern for road safety, leading to accidents and fatalities. This paper presents a novel approach called Optimized Dual-Tree Deep Learning (ODT-DL) for real-time drowsiness detection in drivers. The model uses advanced techniques like image preprocessing, feature extraction, and feature selection. It uses Hidden Markov Models for sequence modelling and classification, enabling accurate drowsiness detection. The experimental evaluation of ODT-DL on two benchmark datasets, YAWDD and NTHU-DDD, shows outstanding performance, with accuracy, precision, recall, and F1-Score consistently exceeding 99%. The model's high discrimination capabilities and low false alarm rates ensure reliable detection. Comparative analysis against other machine learning models, such as AlexNet, ResNet, Support Vector Machine, and ensemble methods, highlights the superiority of ODT-DL. The findings suggest the model's practical implications for enhancing road safety by preventing accidents caused by driver drowsiness, with potential applications in vehicle safety systems. The proposed ODT-DL model holds promise for real-world implementation and opens avenues for future developments in road safety technology.