2017 Research Project

Development and Evaluation of Vehicle to Pedestrian (V2P) Safety Interventions

Principal Investigator
Missy Cummings
Duke University
View Bio

Full Report

Research Brief

Project Slide Deck


Globally, pedestrian deaths account for almost a quarter of all traffic related deaths and are also increasing (World Health Organization, 2018). In the US, pedestrian fatalities now account for approximately 16% of all motor vehicle crash-related deaths (Retting, 2018), with an 81% increase in injuries to distracted pedestrians since 2005 (Nasar & Troye, 2013). These increasing injury and fatality rates are concerning given that cars, in theory, have more safety devices on them today than ever before. Moreover, with increasing worldwide focus on autonomous self-driving vehicles, it is not clear that such advanced technology can account for vulnerable users such as pedestrians. It is also not clear how much pedestrian risk will be increased with the arrival of more automated vehicles and what could be done to mitigate such risks when these cars are more commonplace.

This research effort, the first to conduct a controlled experiment of crossing pedestrians in a field setting with smartphone-based alerts, demonstrated that in a group of 30 participants given smartphone aural and visual alerts of varying reliability while engaging in distracted walking, only 2% exhibited a tendency towards unsafe crossings, while 18% tended towards risky crossings. These results parallel similar observational studies. Non-US-born participants, representing half the test population, were statistically more likely to engage in risky crossing behavior despite developing accurate trust models of the alert reliability. This was particularly true for non-US-born participants with higher than average neuroticism personality scores.

These results suggest national origin plays an important role in the use of technological interventions meant to promote positive behaviors and solutions effective in one setting may not generalize to other nations. Moreover, technology-focused interventions are currently not producing effective solutions, especially across different nationalities. While the subject pool was small in this study and more research is needed in a larger population, this research suggests design criteria might be elucidated from such use of machine learning classification methods in concert with controlled experiments. In this experiment, whether people stopped at or before approximately two feet from the road’s edge predicted safer crossings. Such a threshold could be critical for the designers of autonomous cars who need to prioritize the tracking of multiple entities in congested environments. Those pedestrians that move, for example, inside two feet with constant or increasing velocity or acceleration can become high priority entities to track.

More research is needed to determine such thresholds, including variations due to nationality, road and sidewalk design, and proximity to particularly vulnerable populations, i.e., high school and college campuses with higher numbers of people like to engage in distracted walking. However, given that cars like those from Tesla and Waymo already collect this information at levels researchers never could, allowing non-partisan researchers to access this data and develop safety-based models to be shared across all manufacturers would help prevent future fatalities.

Project Details

Project Type: Research
Project Status: Complete
Start Date: 3-1-2017
End Date: 5-1-2019
Contract Year: Year 1
Total Funding from CSCRS: $200,000
Collaborating Organizations:  Duke University; University of North Carolina, Chapel Hill