Why am I stuck in traffic? A Tennessee project could provide answers

If you drive, you know the situation: You’re on your way to work, to meet friends or to start a week of vacation at the beach or a cabin up in the mountains. You’re cruising along the interstate when all of a sudden you see lots of red tail lights ahead. You hit your brakes and spend the next five or so minutes sitting in traffic, slowly inching forward. And just as suddenly as you stopped, the traffic breaks up and you are on your way again.

According to researchers in Tennessee, these “phantom traffic jams”—slowdowns caused by driver behavior rather than a crash or other obstruction—are far more common than many people realize. It’s one lesson traffic engineers have learned from a new “test bed” of high-definition cameras placed along a highly congested stretch of freeway in the Nashville area. 

By studying the data, researchers at Vanderbilt University are unlocking crucial information about how drivers affect the flow of traffic in their interaction with each other, with traffic laws and with the environment. 

“It was completely unknown how common phantom traffic jams are, whether they’re once in a blue moon, or every day,” said Daniel Work, a Vanderbilt University civil engineering and computer science professor who helps lead the project. “We’re seeing 20 waves in a three-hour window every morning on Interstate 24.”

Vanderbilt researchers chose the stretch of interstate, in part, because it has a high crash rate. State officials blame the crashes on congestion, including those phantom traffic jams, and on aggressive driver behavior.

But there’s another advantage to the test bed location: The study area is on a part of the interstate where the Tennessee Department of Transportation is rolling out new tech features designed to alleviate the chronic congestion there.

Last month, for example, the state started imposing dynamic speed limits. The agency changes the top speed depending on real-time data it collects from radar sensors along the corridor. Other improvements include overhead signals that permit or prohibit the use of specific lanes or that divert traffic onto parallel roads. Eventually, the state also plans to add ramp metering, cameras and message boards along a parallel highway.

TDOT hopes that the information collected by the network of cameras will help it determine how effective those other improvements are, said Lee Smith, the interim director of the department’s traffic operations division, in an email.

“In addition, TDOT will be able to evaluate other traffic operations strategies such as the effectiveness of the Move Over Law, presence of law enforcement in work zones and many others,” he wrote, noting that the state has already used information from the camera project to improve its variable speed limit algorithm. “This is a tremendous benefit compared to other deployments of [variable speed limits] in other parts of the country,” Smith wrote.

“This is the first time (as we are aware), that a state DOT has developed a test bed with these capabilities, particularly placed in open traffic,” he added. “TDOT has plans to use the test bed, and we encourage other state DOTs, researchers, auto manufacturers, etc., to come test their traffic management strategies and products on this test bed.”

For years, engineers have relied on basic information about the flow of cars and trucks on highways, such as how many vehicles used a stretch of road and how fast they were going. But that data doesn’t capture the behavior of individual drivers or vehicles, Work explained.

Existing datasets also don’t help researchers understand how new technology like adaptive cruise control or autonomous vehicles might change the way traffic moves on busy corridors.

“It’s like we’ve been sitting around with X-rays trying to understand why my muscle hurts and not seeing anything and drawing no conclusions,” Work explained. “At some point, we got tired of hearing, ‘Wouldn’t it be better if?’ and we started talking to our friends at the Tennessee Department of Transportation.”

Those conversations led to the installation of 294 high-resolution cameras along a four-mile stretch of Interstate 24. The pan-tilt-zoom cameras, which are made by Axis, are placed on rings of six cameras a piece, and mounted more than 110 feet off the ground. The high perch allows them to capture vehicles’ movements continuously as they move along the stretch. Artificial intelligence helps trace the path of vehicles from one camera to another, with the data stored anonymously.

“It’s an area that gets roughly 150,000 vehicles a day, so you’re looking at hundreds of millions of vehicle miles traveled every year,” Work said. “Every question you could possibly want to know about vehicle interaction, it’s in that data.”

The one-of-a-kind installation will allow researchers to get more comprehensive data than prior approaches, he explained.

In the 1970s, for instance, researchers tried to gather similar information by flying helicopters over a freeway.

In 2005, the Federal Highway Administration mounted seven cameras on a 30-story building along Interstate 80 in Emeryville in the San Francisco area. The study only tracked movements for 45 minutes along less than a third of a mile, but it is still frequently cited in industry research.

More recently, Work said, researchers have started using drones to capture vehicle movements. But drones only capture a short section of the highway, and they’re limited by the battery life.

It’s only recent advancements in technology that have made the Tennessee test bed possible. The project uses 4K resolution cameras, which can be mounted high enough to track vehicles and still capture enough information. Advancements in computer vision algorithms also unlocked the ability of researchers to translate the video into data that they could study.

Still, the amount of data collected by the cameras requires a “small server farm” to store and process at Vanderbilt University some 10 miles away, said William Barbour, a Vanderbilt research scientist who helps run the project. Putting that much information on the cloud would be impractical and expensive.

The test bed’s first experiment came last November, when carmakers deployed 100 automated vehicles with modified adaptive cruise control to try to smooth phantom traffic jams. They found that just 5% of cars driving the “right way” is enough to substantially reduce stop-and-go driving and improve fuel efficiency for everyone, Work said.

One of the next tests will be to see whether changing speed limits can smooth traffic flow during rush hour. “If we can get just enough influence on, call it 5% of the drivers out there who are willing to not just blast 30 mph over the limit through a slow speed zone,” Work said, “they can have a significant positive impact on all the vehicles that travel behind them in reducing speed variability, improving fuel efficiency and not having any negative impact on travel time.”