Driver drowsiness is a major contributor to road crashes worldwide, motivating the development of driver monitoring systems based on behavioral, vehicular, and physiological signals. Among physiological modalities, electrocardiography (ECG) has attracted increasing attention due to its ability to capture autonomic nervous system activity associated with fatigue and reduced alertness, while remaining relatively easy to acquire in real-world driving settings through unobtrusive sensors embedded in the steering wheel or vehicle seat. Most existing ECGbased drowsiness detection approaches rely on handcrafted heart rate variability (HRV) features derived from domain knowledge. While these features are well established, it remains unclear whether additional information relevant to drowsiness can be learned directly from the raw ECG waveform. In this work, we explore whether an end-to-end deep learning framework can learn complementary representations directly from raw ECG signals and achieve competitive performance without explicit feature extraction. The proposed architecture combines convolutional neural networks (CNNs) to extract relevant features from the physiological data with long short-term memory (LSTM) networks to capture their temporal dependencies and dynamics. We evaluate the proposed framework on a driving dataset comprising approximately 120 hours of data acquired during simulated and naturalistic driving conditions and compare its performance against a conventional approach based on handcrafted HRV features. Experimental results demonstrate that the CNN–LSTM architecture achieves an accuracy of 60.55% under a leave-onesubject- out (LOSO) evaluation strategy, demonstrating that automatically learned ECG representations can provide competitive performance while avoiding explicit feature extraction. These findings suggest that deep learning offers a promising alternative for ECG-based driver drowsiness detection and motivate further investigation on larger and more diverse datasets
| ID | pc700 |
| Full Text | |
| Tags |
