Abstract While road safety continues to improve with the increasing prevalence of advanced driver assistance systems (ADASs) and automated driving systems (ADSs), it remains a challenge to achieve widespread acceptance. Understanding driver behavior in context is essential to design a safe and effective system that drivers trust without finding annoying. This study analyzes a large set of realistic (semi-naturalistic) on-road driving data to shed light on how drivers’ glance behavior changes in response to changes in crash risk, measured objectively through looming (TTC; optically defined time to collision). Statistical analysis revealed that drivers adapt their off-road glance behavior as a function of looming as they approach a lead vehicle. The drivers in this study avoided initiating an off-road glance at looming levels above 0.2–0.3 s-1 and the tails of the off-road glance duration distributions became substantially thinner at high looming levels while the median duration remained relatively stable. These findings indicate that drivers adapt to the greater crash risk associated with high looming by reducing the frequency of long off-road glances, and highlight the importance of including contextual information in driver inattention detection systems. The results also support the advancement of driver-adaptive ADASs and behavior-based driver models for ADS evaluation.
Svärd et al. (Fri,) studied this question.