Alex Goldmark is the senior producer of Note to Self, a storytelling show about how technology is changing society. Subscribe here to get Note to Self shows delivered right to your devices. Follow him on Twitter @alexgoldmark.
Algorithm Predicts Red Light Runners
Monday, December 05, 2011 - 07:00 PM
MIT has developed an algorithm to predict which vehicles will run a red light. That's useful for a future where cars talk to each other, or traffic signals respond dynamically to traffic.
There are more than 2.3 million accidents a year at U.S. intersections causing 7,000 deaths. About 700 of those fatalities are from red light running. MIT's algorithm aims to reduce that number by giving advanced warning to drivers -- or even just the car's computer braking system -- that another vehicle is approaching that is statistically likely to roll on through an intersection.
The MIT algorithm used sensors capturing vehicle speed, deceleration, and distance from intersection to predict with 85 percent accuracy which cars would run red lights and which wouldn't.
The U.S. Department of Transportation outfitted a Virginia intersection with censors that tracked the behavior of more than 15,000 vehicles, including when they ran a red light. This is part of a larger effort by the DOT to plan for a future where cars talk to each other in numerous ways that improve safety. As we've reported before, once cars can talk to each other, they will need to know what to say that actually improves safety. Early warning of a dangerous approaching vehicle would be near the top of the list.
The new MIT algorithm, with 85 percent accuracy, is an improvement of 15 - 20 percent over previous algorithms. As important as accuracy, the predictions proved early enough to be acted upon. Researchers found a “sweet spot” one to two seconds in advance of a potential collision where the algorithm was most accurate, and while a driver still had time to react if, say, a red danger light on the dashboard began flashing.
According to an MIT News Release, the researchers report their findings in a paper that will appear in the journal IEEE Transactions on Intelligent Transportation Systems.
Hat tip Gizmag.