Want to Stop Car Crashes? Study the Near-Misses

For once, something intuitive turns out to be true: near-miss car crashes are a very good predictor of actual crashes on that same roadway — and cities can save lives by changing road designs in high-conflict areas before more people die, a groundbreaking new traffic study shows.

In a first for any U.S. city, Bellevue, Wash. collected over 5,000 hours of data from a network of high-definition video cameras throughout its street network, cameras that analyzed persistent patterns of dangerous driver behavior to observe road conditions that most frequently lead to traffic “conflicts” (a.k.a.: near-miss collisions) on their streets.

The results show that crashes happen where the near-misses happen…

The Bellevue project was unique because it collected traffic conflict data across 40 cameras spanning an entire city— and because is used state-of-the-art artificial intelligence to isolate patterns in the data.

The city worked with Transoft, which automatically identifies road user types, and speeds, trajectories, and vehicle sizes — without collecting anyone’s personal information.