Crash risk associated with eyes-off-road duration by road control type and intersection type

Abstract

Introduction: Scanning the driving environment is crucial for safe driving. However, the complexity of the driving environment affects a driver’s ability to maintain sufficient situational awareness. This study aimed to quantify the crash risk associated with eye-off-road (EOR) behavior across different road control and intersection types, providing insights for Driver Monitoring Systems (DMSs). Method: Using data from the Second Strategic Highway Research Program Naturalistic Driving Study (SHRP 2 NDS), we analyzed crash risk associated with EOR glance durations within a 6-second window. We compared odds ratios across three EOR duration groups: ≤ 2 s, 2 to 3 s, and 3 to 4 s. Additionally, we used a General Additive Model (GAM) to explore the non-linear relationship between EOR duration and crash risk. Results: Our findings reveal that crash risk is higher on uncontrolled access roads compared to controlled access roads. Among intersection types, turning at intersections presents the highest risk, followed by going straight at intersections and driving on straight non-intersection segments. Notably, the crash risk increases linearly only for short EOR durations but rises much faster than linearly when EOR durations exceed specific thresholds: 1.3 s for controlled access roads, 1.2 s for uncontrolled access roads, and as low as 0.9 s for turning at intersections. Conclusions and Practical Applications: These results challenge the commonly used 2-second threshold in driver distraction research, offering more refined insights that could enhance DMS alert settings based on driving environment. This study underscores the importance of considering road and intersection types when assessing the risks of EOR behavior, contributing to improved road safety strategies. Note that the study is not intended to specify exact values for DMS timers, which must be determined in practice by considering multiple factors. Additionally, the false alert rate requires further evaluation.