Predict and prevent: Curbing traffic accidents with machine learning

The crowd-sourcing tool could use your help.

In 2016, road crashes resulted in 40,000 deaths and 4.6 million injuries in the United States. For young people under age 19, these collisions were the leading cause of death. Until now, cities and transportation agencies have had to review many years’ worth of crash data to detect deadly patterns of car accidents before acting.

The goal of preventing death and serious injury is now within reach thanks to a partnership among the UW, the city of Bellevue and Microsoft. “The role for the UW team is to provide the definition of near misses and analyze near-miss data for safety and operational improvements,” said Yinhai Wang, professor of civil engineering. “The goal is to eliminate traffic deaths and serious injuries.”

The project aims to use footage from traffic cameras across North America to “teach” computers how to recognize near-miss collisions. Data from these machine-learning systems will enable transportation engineers to predict where crashes will occur and take proactive measures to prevent them.

The project is looking for people to use a crowd-sourcing tool to analyze video and teach computers to identify a person in a wheelchair, on a bike or in a car, as well as patterns of movement in intersections. The more people who take part, the better computers will learn to recognize near-miss collisions. Learn more.