<![CDATA[Driver fatigue is a major crash factor, yet camera/physiology methods face cost, privacy, and calibration barriers. This paper presents an embedded, noninvasive system using four OBD-II PIDs—speed (0x0D), RPM (0x0C), throttle (0x11), and engine load (0x04)—polled in real time via ELM327 on Jetson Orin. Engineering values (Phys_) and raw ECU payloads (RAW_) are logged for DoCAN/ISO-TP auditing. The real-world dataset contains 51.80 h (5 drivers, 10 sessions, 2 contexts) at 2 Hz, labeled by KSS and mapped to a continuous fatigue score. Decoding fidelity is high (r≥0.9996, MAE<0.25). From 60 s/10 s windows we extract instability features (SD, 0.05–0.3 Hz corrective-oscillation bandpower). Wilcoxon exact tests show significant shifts at high fatigue (p≤0.001953). Cluster-robust regression improves R² 0.266→0.716; LODO reduces RMSE 0.198→0.122 and raises prediction correlation 0.525→0.823. Findings match reduced vigilance and sensorimotor control stability under fatigue. This supports 4-PID OBD-II as a low-cost edge fatigue-warning modality.]]>
Copyrights © 2025
