Early Drowsiness Detection via Second-Order Derivative Analysis of Heart Rate Variability: A Non-Contact ECG Approach with Machine Learning

Drowsy driving accounts for approximately 20% of traffic fatalities, yet current detection systems rely on behavioral indicators that manifest only after significant impairment has occurred. This study investigates whether first and second derivatives of heart rate variability (HRV) can provide earlier drowsiness detection, enabling proactive safety interventions before crash risk escalates. Data were collected from 25 participants across 49 driving simulator sessions using capacitive ECG electrodes embedded in the seat backrest—a non-contact, privacy-preserving approach suitable for vehicle integration. Ground truth labeling combined crash-proximity metrics (30%) and behavioral indicators (70%), explicitly excluding HRV derivatives to ensure unbiased evaluation. A Neuroplastic + NADN Vision Transformer achieved 87.5% accuracy (F1 = 0.85) for binary Alert/Light drowsiness classification. Critically, HRV derivatives alone achieved 78.1% accuracy without any visual features—demonstrating the feasibility of camera-free monitoring. Temporal analysis revealed that derivative-based detection preceded behavioral manifestations by 5–8 minutes and crash events by 6.8 ± 2.3 minutes, providing a substantial early warning window for graduated driver alerts or automated interventions. Analysis of 2056 crash events revealed that driving impairment manifests during the transition toward drowsiness: 56.2% of crashes occurred in Alert state and 43.8% in early Light drowsiness, demonstrating that HRV derivatives detect physiological precursors before traditional drowsiness thresholds are reached.