Investigating Driver-Specific Variability in Vision-Based Driver Fatigue Monitoring: A Subgroup Evaluation Design Tested in a Controlled Simulation

Abstract Driver fatigue is a significant concern for road safety, and vision-based driver monitoring systems offer a non-invasive approach to identifying fatigue-related behaviours. However, variations in individual eye characteristics, eyewear, head pose, and illumination conditions may influence the reliability of fatigue indicators across different drivers. This study investigates driver-specific variability in vision-based fatigue monitoring through a controlled simulation-based subgroup evaluation framework. The proposed evaluation examines the effects of four factors—eye geometry, eyewear, head pose, and illumination—on eye-based fatigue indicators and compares fixed-threshold monitoring with driver-specific calibration approaches. The framework incorporates subject-disjoint evaluation, subgroup-specific performance assessment, sensitivity analysis, interaction analysis, and statistical checks to examine how these factors may affect monitoring outcomes. Particular attention is given to the relationship between individual variability and the consistency of fatigue classification across simulated driver subgroups. The study aims to establish a reproducible evaluation methodology for identifying potential sources of performance variation in camera-based driver fatigue monitoring. The findings are intended to inform the design and evaluation of more adaptive and reliable driver monitoring systems. Since the investigation is based on controlled simulation, its outcomes should be interpreted within the assumptions of the simulation framework and require further validation using real-world driver recordings.