Driver Stress Detection Based on Electrocardiogram Signals using Deep Learning Techniques

This study presents a deep learning-based method to detect driver stress from a single electrocardiogram (ECG) signal. Driver stress is a significant contributor to traffic violations and accidents, impairing driver concentration, judgment, and reaction time, thereby leading to aggressive driving and poor decision-making. Real-time detection and management of driver stress are crucial for enhancing road safety and improving drivers’ overall health. Existing approaches for detecting driver stress using multimodal biosignals require various types of sensors, which presents practical challenges in terms of cost and complexity when applied in real vehicles. To address the limitations of existing methods, the proposed approach integrates a DenseNet-based 1D convolutional neural network (CNN) model with state-of-the-art attention mechanisms, including dual-channel attention mechanism (DCAM) and deep reinforcement fine-tuning (DeepRFT). The preprocessing phase includes baseline removal, normalization and data augmentation, with model evaluation conducted using the leave-one-out technique. The results indicate that models using attention modules achieved the highest accuracy and F1-scores, surpassing traditional methods in identifying stress states such as rest, low stress and high stress. Specifically, the DCAM-based model achieved an average accuracy of 86.2 % and an F1-score of 83.5 %, showing a significant improvement over baseline models. These findings suggest that ECG-based stress detection can enhance driver assistance systems, contributing to proactive stress management and safer driving environments.