Low-light driver drowsiness detection for real-time safety assistance using dual attention mechanisms in deep learning model

This research presents a robust real-time driver drowsiness detection system employing deep learning, attention mechanisms, and explainable AI (XAI) techniques to address this critical safety concern. The system integrates a fine-tuned InceptionV3 baseline with dual attention mechanisms, i.e., spatial and channel attention mechanisms, alongside a Low-Light Fine-Tuned LLFormer, to enhance detection performance in complex scenarios such as low-light conditions and occluded facial features. Additionally, the ResNet-50 model is utilized for feature extraction, while XAI techniques like Grad-CAM, LRP, etc., are incorporated to provide interpretability and transparency to model predictions. Multiple drowsiness indicators, including head tilting, blinking, and yawning, are analyzed using temporal factors, supported by facial landmark key point detection and a multi-browser distraction detection module for comprehensive monitoring. Experimental results reveal significant improvements, achieving up to 98.4% accuracy even under challenging conditions such as drivers wearing glasses, low light, and varied levels of facial occlusion. The model is optimized for real-time deployment on mobile and embedded platforms with minimal computational overhead. By incorporating these innovations, the proposed solution demonstrates the potential to significantly reduce drowsy driving-related risks, providing a practical, scalable, and interpretable tool for advanced driver assistance systems aimed at enhancing road safety.