One of the most dangerous drivers is the “red-light runner” — one who, whether through inattention or impatience, doesn’t stop at a red light. Researchers are now working to predict which drivers are going to do this and potentially program your car to warn you.
Red-light runners can be identified because they don’t have the right “braking profile,” says Jonathan How, R.C. Maclaurin professor of aerospace at the Massachusetts Institute of Technology (MIT).
“They’re not slowing down, which could be a case of just having missed the light,” he says. “Or someone braking but not sufficiently, so the system would detect that they wouldn’t likely stop in time. We’re looking at acceleration, velocity and position.”
By using these parameters, the MIT researchers came up with an algorithm that predicts when an oncoming car is likely to run a red light.
Testing with data collected from an actual intersection, they correctly identified red-light runners 85 per cent of the time, predicting within a couple of seconds whether a driver would run the light.
That could give other drivers enough time to react, providing they were given the information.
The ideal warning, How says, is through an existing technology called vehicle-to-vehicle (V2V) communication, in which cars wirelessly “talk” to each other to exchange information.
Using the information in the algorithm, a car that suspects its driver isn’t going to stop would broadcast that to other cars approaching the intersection. Those cars would flash warnings to their drivers, who could then brake or take other evasive action.
While these oncoming cars could also be physically detected through sensors, How believes the V2V system is superior.
“It’s important for visibility issues,” he says. “Sensors can’t see around corners or if someone’s blocking you, like a big truck beside you at a light. V2V circumvents much of that because vehicles have the ability to say, ‘Here I am, here’s what I’m doing.’ In many instances your visibility isn’t what it should be, and this provides more information.”
The researchers also have to ensure that the system isn’t too sensitive.
“We had to think about low false-alarm rates and seeing how you can predict things well, but not be skittish,” How says.
If a system overreacts too often, the driver will probably just shut it off.
How says the system, built into new cars and retrofitted to existing ones, could potentially hit the market in five to 10 years.
