Smarter Vehicles, Safer Roads

October 13, 2010

North Carolina has the tenth highest rate of traffic fatalities in the United States, according to the National Highway Traffic Safety Administration. More than 1,400 people died on state roads in 2008, the most recent year for which figures are available. That's roughly the same number who died in wrecks 15 years earlier, even though traffic fatalities had dropped 8 percent nationwide. NC State researchers are trying to improve those grim statistics, studying ways to help drivers stay in their lanes and avoid crashes.

Dr. Wesley Snyder has developed a 'seeing' car that can keep the vehicle in its lane. Full-scale demonstrations will need more computing power to move at higher speeds.  (Photo: Roger Winstead) 
Dr. Wesley Snyder has developed a 'seeing' car that can keep the vehicle in its lane. Full-scale demonstrations will need more computing power to move at higher speeds. (Photo: Roger Winstead)

The DARPA Grand Challenge, a Department of Defense competition to build driverless vehicles, got Dr. Wesley Snyder thinking a few years back about ways to build cars that use human-like vision. The professor in the Department of Electrical and Computer Engineering (ECE) didn't participate in the contest, but he used his expertise in robotics to take a step toward the goal of a "seeing" car by designing a system to keep a vehicle in its lane of traffic. Such a system could be crucial if a driver has a heart attack, seizure, or other medical condition that would cause a loss of control.

Snyder developed algorithms to mimic some of the biology believed to be behind the way people see. His research team then placed a video camera on the hood of a toy jeep and hooked it up to a laptop computer on the back of the jeep. The camera sends images of the road ahead to the computer, where the algorithms recognize lane markers, stop signs, tail lights of other vehicles, and traffic lights. The curvature of the road helps the computer infer the jeep's speed and distance from other objects as the vehicle moves. Snyder is now working on boosting computing power because the laptop can process only two images each second, limiting the jeep to a snail's pace. "We've got part of it solved, but we still have more work to do," he says. "Driving in tight traffic at high speeds is a very tough problem."

Dr. Mo-Yuen Chow uses tiny robotic cars and a street grid mapped out on the floor to test his accident-avoidance system, which uses a central traffic-control system to communicate with vehicles. Improved in-car GPS systems, above, would warn drivers of the need to take quick action.  (Photo: Roger Winstead) 
Dr. Mo-Yuen Chow uses tiny robotic cars and a street grid mapped out on the floor to test his accident-avoidance system, which uses a central traffic-control system to communicate with vehicles. Improved in-car GPS systems, above, would warn drivers of the need to take quick action. (Photo: Roger Winstead)

Dr. Mo-Yuen Chow, Snyder's ECE colleague, is taking a different tack to try to solve that problem. Using the concept of "intelligent space," or the control of distributed networks, he is working on a system to turn future vehicles into the equivalent of planes being tracked by air-traffic control. State transportation departments already have control centers to monitor traffic in metro areas, and Chow says global-positioning system (GPS) devices could one day beam information collected from various vehicle sensors to the control centers and receive real-time warnings not only about traffic congestion ahead, but also when the driver needs to take quick action to avoid a collision.

Using biological names to describe his work, Chow compares the overall project to an immune system for vehicles and says the first step is to create a "gene library" of situations a driver might face. Factors like vehicle speed, weather conditions, the curve of the road, and traffic combine in different ways to produce a range of outcomes, he says, much like genes do in humans. Eventually, human factors like driver age and fatigue will be in the gene library, too. As vehicle information is fed into traffic control centers from numerous GPS devices, a program will run through the database to assess the risk that two or more vehicles are headed for a collision and will alert them as needed to take corrective action. "This isn't like air traffic, where you have miles between planes," Chow says. "We need to process data in real time for split-second decisions."


View the full article - Results - Research and Graduate Studies at North Carolina State University