Detection-Aware Generation of Fatigue-Conditioned Ocular and Head Behaviors for Virtual Drivers

Open Access
Article
Conference Proceedings
Authors: Woojin Kim, Seung-Jun Lee, Yonghyun Kim, Myeongkyu Lee, Brandon J. Pitts
Abstract

Driver fatigue induces systematic changes in observable behaviors, which should be represented in virtual drivers beyond predefined animations. This study presents a data-driven framework for generating stochastic ocular and head behaviors conditioned on subjective sleepiness. Empirical behaviors were obtained from a controlled driving experiment and characterized across Alert, Intermediate, and Drowsy states based on the Karolinska Sleepiness Scale (KSS). Significant fatigue-dependent changes were observed in eyelid-closure characteristics, PERCLOS, and head-pitch behavior, and were incorporated into probabilistic behavioral models. A detection-aware generation method was developed to synthesize eyelid trajectories while accounting for discrepancies between intended behavior and behavior re-measured after detection. Monte Carlo validation showed that generated PERCLOS distributions reproduced empirical distributions across all KSS states, with normalized Wasserstein distances of 0.14–0.24 and no significant differences detected by Kolmogorov–Smirnov tests. These results provide quantitative evidence that the proposed generator reproduces fatigue-dependent characteristics and population-level variability observed in real drivers.

Keywords: Virtual Driver, Driver Fatigue, Behavioral Generationn, PERCLOS, KSS

DOI: 10.54941/ahfe1008383

Cite this paper
Downloads
0
Visits
1
Download PDF

More from this volume

← Development, Comprehension Testing, and Preliminary Evaluation of Child-Adapted Measures of Visually Induced Motion Sickness for Virtual RealityDesign and Preliminary Evaluation of a Generative AI Role-Play System for Practicing Cognitive Empathy →
View all articles in Human Factors in Design, Engineering, and Computing →