Citation

Li YE, Hao H, Gibbons RB, Medina A. Transp. Res. Rec. 2020; 2674(12): 291-302.

Copyright

(Copyright © 2020, Transportation Research Board, National Research Council, National Academy of Sciences USA, Publisher SAGE Publishing)

DOI

10.1177/0361198120953432

PMID

unavailable

Abstract

Crashes involving roadway objects can cause severe injuries and property damage. Utilizing data from the Second Strategic Highway Research Program (SHRP 2) naturalistic driving study (NDS), this study investigated crashes involving roadway objects and their implications for the potential of machine vision-based driving systems in preventing such crashes. A comprehensive statistical and machine learning analysis was first conducted to identify major factors affecting the occurrence and severity of such events. Machine vision performance metrics (based on the SHRP 2 NDS cameras) and human driving decisions were then analyzed to identify opportunities where machine vision systems could particularly mitigate risk factors. The results suggest that driver behaviors/errors, speed, reaction time, and object characteristics played the most significant role in the occurrence and severity outcome of the SHRP 2 events. The average object detection distance based on the SHRP 2 cameras was approximately 20 m for all objects. The average reaction time provided by the SHRP 2 cameras was 1.5 s for all events but 1.1 s for events involving animals and roadway debris. In general, the machine vision reaction time was longer than the driver reaction time for approximately 95% of all analyzed events and 75% of the events in which drivers reacted before collisions. Drivers were able to make safe evasive maneuvers in 56% of all analyzed events and 72% events involving roadway debris and animals. Based on these results, the paper discusses in detail when and how machine vision systems could assist in preventing crashes involving roadway objects.

Language: en