Validation of Commercial Wearables for Driver Monitoring: Assessing Heart Rate Accuracy and Reliability for Early Detection of Sudden Illnesses

Sudden medical illnesses (SMI), including myocardial infarctions and occult arrhythmias, represent a primary etiology of catastrophic motor vehicle collisions. Although consumer wearable technologies provide a non-invasive modality for continuous driver monitoring, their measurement accuracy within dynamic automotive environments requires systematic evaluation. This investigation establishes an empirical validation paradigm for assessing the pulse-rate (PR) accuracy of consumer-grade wrist and finger photoplethysmography (PPG) devices against a reference heart rate (HR) value derived from an electrocardiographic (ECG) signal. To secure an uncompromised physiological reference, an analytical signal-processing pipeline was implemented, incorporating an Adaptive Layer-Dependent Thresholding Function (ALDTF) coupled with Maximal Overlap Discrete Wavelet Transform (MODWT) decomposition to suppress high-frequency motion artifacts and baseline wander (BW) in the ECG signal without inducing morphological distortion. Nineteen real-world driving sessions, totaling over 18 h of continuous physiological telemetry, were evaluated via Lin’s Concordance Correlation Coefficient and Bland–Altman limit-of-agreement statistics. The findings demonstrate a distinct performance bifurcation governed by device form factor and internal data filtering. The evaluated smartwatch exhibited reliable tracking (MAE = 2.73 BPM) and preserved underlying autonomic variability, attaining distributional alignment with the ECG reference in 47% of trials. Conversely, ring-based form factors demonstrated substantial perfusion decoupling induced by isometric manual gripping forces, alongside aggressive proprietary smoothing filters that suppressed micro-fluctuations in heart rate variability (0% distributional alignment). These outcomes indicate that while wrist-worn form factors possess sufficient fidelity for detecting subclinical autonomic shifts, their implementation in automotive safety systems requires access to raw data to circumvent proprietary algorithmic filtering.