Machine vision’s eye on driving’s future
Wouldn’t it be great if traffic signals were half as smart as traffic cops? Some new technology is promising to improve the intelligence of traffic signals at major intersections. It is just one of several major advances in which cameras and computers are transforming law enforcement, highway safety and eventually driving itself.
Aldis Corp., an Oak Ridge, Tenn., technology company, is developing a camera and software system that promises by next year a significant improvement in automated traffic signals.
For decades, traffic engineers have gone out every few years and attempted to time signals to maximize the flow of vehicles on the majority of major streets. It is only somewhat effective.
Unless a street grid is perfectly symmetrical, engineers can time signals for only one direction of traffic. And if traffic signals are spaced more than half a mile apart, the platoon of vehicles from one signal to another breaks up, said Tom Hicks, vice chairman of the traffic engineering subcommittee of the American Assn. of State Highway and Transportation Officials.
To supplement timing, engineers have put detectors in the road to determine when vehicles arrive at an intersection. These detectors -- loops of wire embedded in the road -- are expensive and prone to constant failure. And more recently, engineers have used up to eight cameras to detect vehicles, an even more expensive solution.
Aldis plans to start testing next month a single camera that would be hung below a traffic signal, looking out about 600 feet in every direction with a fisheye lens. A computer with some fairly advanced software would calculate the speed of cars, trucks and buses, timing the signals to maximize the flow of vehicles, said Bill Malkes, who helped develop the system.
“If it sees you coming and there is nothing in the other direction, it changes the signal so you don’t have to stop,” Malkes said.
The custom camera and associated software and hardware will cost an estimated $15,000 per intersection, somewhat more expensive than buried loops but requiring less costly maintenance over time, Malkes said.
The setup not only would increase the capacity of the road, but it would also improve the fuel economy of vehicles and reduce emissions, which are greatest when vehicles accelerate from a stop. The safety potential also seems significant, Malkes said, allowing the system to hold cross traffic if it senses that a car is not going to stop for a red light. More than 800 Americans are killed each year when somebody runs a red light.
Another type of camera, mounted on a police car, promises more changes for American roadways. The camera can scan up to 30 license plates per second, convert the images into data and match license plate numbers against a list of wanted vehicles. The patrol officer is alerted on his car terminal to a hot car.
The system, developed by ELSAG North America Law Enforcement Systems, a unit of Finmeccanica, is in use by police departments around the country, including in Rialto, Upland and Baldwin Park.
Recently, it was adapted for use on school buses. The idea in that implementation is to capture license plate images of vehicles that don’t obey school bus laws and might endanger students during loading or unloading, a spokeswoman said. The data are transferred to a laptop in the bus and then reviewed for violators at the end of the day.
Sebastian Thrun, director of artificial intelligence for Stanford University, said machine vision will eventually transform our transportation system in myriad ways.
“We can make driving more efficient, safer and more fun,” Thrun said.
The potential may be greatest with automated vehicles. In November, a Stanford team came in second in a Defense Department “urban challenge” contest, in which fully automated vehicles navigated through mock intersections, traffic circles and streets at a 60-mile course in Victorville.
Machine vision still faces many challenges, Thrun said. Cameras and their image processing have trouble understanding something new and are not nearly as good as humans at judging distance.
The automated cars in Victorville had to use laser range-finders, which send out light and measure the amount of time it takes for the light to bounce back.
Nonetheless, Thrun predicts that within eight years cars will have a number of automation features, allowing drivers to set vehicles on automatic pilot and take a nap or direct an auto to park itself inside a garage.
“In the grand scheme of things, we are making a lot of progress,” he said.
