Embedded Real-Time Multi-Risk Detection: An EdgeML-Powered System for Driver Monitoring

Driver drowsiness and distracting behaviors are leading causes of road accidents. This paper presents a cost-effective, real-time driver monitoring system (DMS) that leverages Edge Machine Learning (EdgeML) to detect multiple risk factors: drowsiness, phone usage, smoking, and seatbelt non-compliance. The proposed solution integrates an ensemble of models, object detection (OD), facial landmark analysis, and posture estimation, deployed on a Raspberry Pi 5 (RPi 5) with a Coral USB Edge TPU (ETPU) accelerator. A dual-validation logic cross-confirms biometric anomalies with OD within a temporal window to minimize false positives. The system achieves an average F1-score of 98.66% and 90.46% for two- and five-class classification, respectively, and supports real-time processing at 40 FPS with low resource utilization (21% CPU, 500 MB RAM). This work offers a retrofittable EdgeML solution for real-time driver monitoring, democratizing access to advanced safety features for the billions of cars already on the road. The source code and model weights are publicly available1.