A Personalized Edge AI Framework for Adaptive Driver Fatigue Monitoring and Alert Escalation

This research presents a Personalized Edge Artificial Intelligence (AI) framework for real-time driver fatigue monitoring and adaptive alert escalation. The study addresses two key limitations of existing vision-based driver monitoring systems: the use of fixed detection thresholds that may not account for individual differences in facial characteristics, and the reliance on single-level alerts that do not reflect the severity of driver drowsiness. The proposed framework integrates facial landmark detection, Eye Aspect Ratio (EAR), Mouth Aspect Ratio (MAR), lightweight Convolutional Neural Networks (CNNs), and per-driver calibration to improve fatigue detection reliability. Designed for edge deployment, the system processes camera input locally, reducing dependence on cloud connectivity while supporting real-time performance and enhanced privacy. An adaptive, multi-level alert mechanism maps estimated drowsiness severity to graduated warnings, ranging from low-risk notifications to critical intervention alerts. The research adopts a mixed-methods approach combining a structured literature review with a proposed controlled comparative experiment. It evaluates the framework against conventional fixed-threshold, single-level alert systems using metrics such as accuracy, precision, recall, false-alarm rate, inference latency, power consumption, and alert-timing appropriateness. A working prototype, the Personalized Edge AI Driver Monitor, demonstrates real-time facial monitoring, personalized threshold calibration, severity-based alert escalation, and simulated INT8-quantized MobileNetV3 inference, with observed latency of approximately 40–45 milliseconds. These prototype observations provide an initial demonstration of feasibility; comprehensive statistical validation across a diverse driver population remains future work. The research aims to contribute a reproducible framework for more personalized, computationally efficient, and context-aware driver fatigue monitoring, with potential applications in intelligent transportation systems and embedded automotive safety technologies.