Q&A: How Researchers are Using Surveillance Cameras to Understand Phantom Traffic Jams
Earlier this year, I traveled to Nashville, Tennessee, for the Axis Communications Connect and Converge Conference (ACCC) to learn more about their technology and use case scenarios in the field.
One use case shared during the event was taking place right down the road: the I-24 Mobility Technology Interstate Observation Network (MOTION). The system uses 294 Axis PTZ cameras across a four-mile stretch of Interstate 24 to conduct research on vehicle behavior and traffic flow to improve congestion management across the state of Tennessee.
The Tennessee Department of Transportation (TDOT) owns and is responsible for the maintenance of the system, which was installed in 2022. Vanderbilt University, based in Nashville, is responsible for processing the data the I-24 MOTION project produces and coordinating experiments performed on the roadway.
William Barbour, Ph.D., senior research scientist, adjunct professor, Institute for Software Integrated Systems at Vanderbilt, gave a presentation at ACCC about how the university uses the system. Afterwards, he spoke with me about why traffic studies matter, how the I-24 MOTION system works, and some of the latest research it’s been used for.
Our conversation below has been lightly edited for clarity.
Security Technology (ST). What sparked your interest in studying transportation?
William Barbour (WB). I’ve always been fascinated by construction and infrastructure. I had some interest in sustainability, as well. I was interested in how we could make transportation work more efficiently and provide better experiences for everybody who’s out there on the roadway in different modes.
ST. Why does understanding how traffic moves matter?
WB. We, from a detailed point of view, don’t have a full picture yet of how traffic jams form, propagate, or evolve, and how they disappear. We have models for these sorts of things, but our understanding at the individual driver level—what causes that first bit of congestion and then how does that grow? We haven’t gotten to that level of understanding yet, at least from a data-driven perspective. That was some of the motivation for I-24 MOTION, to gain that understanding.
ST. As a researcher, is there a technical definition of what a traffic jam versus congestion is?
WB. I wouldn’t be completely willing to split hairs on the terminology…but I will point out a few different types of traffic. One would be a discreet event traffic jam—caused by either law enforcement activity on the side of the road or a tow truck or somebody pulled over needing some help changing a tire or something. Some discreet event that’s not recurring, but it’s still out there. That certainly causes a traffic jam—we’re all very well aware of how that happens.
The second category I’d throw out there is infrastructure-based. There’s a chokepoint on the roadway. We would often call that a bottleneck. As a bottleneck implies, trying to force a flow of traffic through that chokepoint—if there’s not enough infrastructure capacity at that point for the amount of demand—then we start getting a queue forming because of that bottleneck.
Then the final type, which is, I think, at the core of some of our gaps of understanding, are the recurring traffic congestion that’s not necessarily due to a bottleneck or other infrastructure cause and is not caused by some discreet event, but it happens every single day.
Some of those things we would call phantom traffic jams. That’s a colloquial term but could be used to capture traffic jams that don’t have an obvious cause that’s from an infrastructure or an event perspective. Those traffic jams are all over our roadways and we don’t fully understand why they’re there.
ST. I’m thinking about in the Washington, D.C., area when you’re driving on the Beltway and you get to a point where everyone has slowed down. You drive like that for 15 minutes and then, everyone speeds up again.
WB. The fundamental reason that happens…because we’re trying to push so many vehicles through that area, traffic density increases and the space between vehicles decreases. There’s a certain point at which everybody can no longer drive that close together at that speed, and the speed starts to drop. It doesn’t just drop a little bit, from free flow of 70 or 65 down to a nice even 50 or 45. It drops off really dramatically. This is what we call the flow breakdown, where you experience a big drop off in speed instead of something modest where if only everybody could drive at 40, we could all just have a nice smooth commute.
Instead, we get these really oscillatory patterns where you’re going 0 to 50, 0 to 50, that sort of thing. That’s really frustrating and not a very pleasant way to drive.
ST. Could you give a broad overview of what I-24 MOTION is?
WB. I-24 MOTION was envisioned as an instrument to measure the finer points of traffic congestion—the stop-and-go traffic, why we see capacity breakdown, and traffic congestion form to begin with. That’s the motion behind it, as well as to be an open road test bed for traffic experiments.
I-24 MOTION was envisioned to address that gap in what we could do on traffic engineering. It evolved into what amounts to a 300-camera system deployed over 4 miles of Interstate 24 southeast of Nashville, Tennessee. Those 4 miles, there's not a square inch of the roadway that's not covered by a camera. All of that video data is processed into anonymized vehicle trajectories for every single vehicle out there on the roadway. Those anonymized vehicle trajectories are the most powerful form of traffic data, because from that ultrafine-grained point of view, we can study, at the level of individual vehicles in an anonymized way, or we can roll that data up into more big-picture traffic metrics and perform data analysis that way.
That's the basic system. It's traffic observation as well as experiments and testing. Then the data product from all that video is anonymized vehicle trajectories. I place an emphasis on anonymized because we don't care what the color of anyone's car is. We don't care what's on their license plate. It's just, there is a vehicle, it is an SUV, or it is a truck, and it is driving in this way, measured by a bunch of ones and zeros.
ST. When you say vehicle trajectory, what do you mean?
WB. That would be the location at every point in time measured at 25 measurements per second of each individual vehicle as it drives through the corridor. That trip might take, let’s say, 5 minutes to drive those 4 miles and that would amount to…thousands of observations of that individual vehicle.
We can map out exactly where it traveled, in which lane, how fast, and other vehicles it interacted with in anonymized way.
ST. TDOT owns and is responsible for the system. How does the work Vanderbilt is doing play into how I-24 MOTION is used?
WB. We at Vanderbilt are responsible for the data processing, as well as coordinating experiments that are performed on I-24 MOTION. We ingest all the raw video from the test bed on a daily basis and process that into vehicle trajectories…then we discard the raw video. That’s performed everyday by servers at our data center. We wrote all the software to do that processing.
We have a great team here that does all the software. Then we publish that data for free to anybody who would like to take a look at it.
At TDOT, they are responsible for the maintenance [of the system] as well as its governance. If we want to run an experiment on I-24, let’s say we’d like to try and do some smoothing out of traffic, they’ll sign off on the safety plan, the management plan for the experiment, and make sure that it’s conducted in a safe and responsible manner since it’s an open roadway.
Gresham Smith is a consulting firm here in Nashville. They’ve been involved from the beginning. They’re the designer of record for all the infrastructure, and they are continually involved on providing management services, as well.
ST. At ACCC you mentioned an experiment done using I-24 MOTION in 2022. What was that experiment?
WB. This was an experiment by the CIRCLES Consortium, which Vanderbilt is a member of. That consortium is a number of universities—University of California Berkely, Rutgers University, and Temple University—as well as some industry partners from the automotive OEM space and other sectors.
The experiment was trying to deploy consumer-grade vehicles with a small amount of custom automation that we added with the express purpose of putting them out in live traffic and having them receive data about how traffic was operating up ahead, then performing some wave smoothing on the traffic stream.
What that looks like in practice is something that a lot of people already do as they’re driving around. If you’re coming up to a stoplight and you see that it’s red up ahead, and you’re expecting that it’s going to turn green, you let off the gas and start coasting in to that signal in hopes that it’s going to turn green and you have minimal disruption to your speed as you pass through.
That was the idea with the traffic wave smoothing—that there are these traffic waves that cause us to oscillate our speed 0 to 40, 0 to 40. Instead of riding each one of those waves, we could potentially smooth it out by trying to drive at more of an average speed.
Those vehicles would receive information about how traffic was looking up ahead and then a small computer inside each vehicle would make a decision about how fast that vehicle should drive for the next segment of time. There were 100 of those vehicles that were deployed as part of the CIRCLES experiment. It was a massive undertaking. Putting 100 vehicles out there requires more than 100 drivers just to support. Custom hardware in each vehicle required a staff of around 10 just for the hardware. On the software side, there were algorithms teams, data teams, as well as the I-24 MOTION team.
ST. What kinds of cars were you using?
WB. These were Nissan Rogues that were generously lent to the project by Nissan. We modified those Nissan Rogues with just a little bit of additional equipment that basically modified their adaptive cruise control.
We were basically overriding the cruise control set point, so changing what the maximum speed was. It was still running the safety logic of the regular cruise control, so never getting too close to the vehicle ahead. We were also changing the following distance—how you can set adaptive cruise control to one, two, or three bars.
ST. How long were they out there driving for?
WB. It was, overall, a five-day experiment and we had all 100 vehicles out there at the exact same time on Friday. They were driving laps around I-24 from around 6:00 a.m. to 9:00 a.m. or 10:00 a.m. It required drivers to take breaks because this is a stressful traffic environment and you don’t want drivers to become fatigued and make a mistake. They would drive for a couple hours before coming back for a break. We’d swap drivers with a reserve driver and get the car back out there.
ST. Are there any results you can share from the experiment?
WB. Some of those results are just now about to hit publication. Overall, we were looking for if we can decrease the overall energy consumption of the traffic stream. With those 100 vehicles, you hope that they’re going to have effects on other vehicles around them.
That was the metric that we were shooting for. I can say that the experiment did have a positive impact on traffic. The result for exactly how much, that’s awaiting publication in a peer-reviewed journal.
ST. Did you do any major experiments using the system in 2023?
WB. We had a bit of symbiosis between the I-24 MOTION project and the I-24 SMART Corridor. The I-24 SMART Corridor is another TDOT project that’s using active traffic management and integrated corridor management to try to address the traffic needs of I-24.
I-24 MOTION is an observational and testing area, but the SMART Corridor is really focused on the commuting public. That system has variable speed limit capabilities, lane control, and diversion routing onto parallel streets—that system was coming online in 2023.
I-24 MOTION provided some of the insights for best deploying the algorithms that determined what to put on the variable speed limit signs. We were using motion to identify how those algorithms were performing, as well as observing some of the congestion-related events from the system.
Overall, it helped us in our understanding of the variable speed limit system and helped us tune that system. It also gave us a way to quantify in high detail the safety benefits of the I-24 SMART Corridor.
ST. What would one of those safety benefits be? Or a recommendation based off of the system?
WB. The variable speed limit system is a high-fidelity warning of the fact that there’s going to be slow traffic ahead. As you’re driving into town or in the morning, you’ll see that variable speed limit system at the regular 70-mile-per-hour default roadway speed. At a certain point, you’re going to see that speed drop down to 60, 50, 40, and then eventually it’ll hit 30 miles per hour—the minimum allowed on that roadway. At that point, you can definitely expect there’s heavy congestion.
There are people, unfortunately, who are not aware that congestion is coming up and they get into a crash because they’re not expecting what could end up being a severe slowdown. The variable speed limit system is effective at back-of-queue warning so that people are noticing “there’s something happening with the speed, there’s some disturbance up ahead. I need to be on the lookout, I need to be more vigilant.”
We have seen a reduction in the crashes secondary to congestion like that. That’s one function of the system, and I think that’s where the primary safety benefit has manifested.
ST. Is there any other research using the I-24 MOTION system that you can share?
WB. We are very much still interested in additional safety studies and things around first responder safety on the side of the road. Tennessee is a move-over state, and we want to make sure that our first responders on the side of the road are as safe as possible. There’s an opportunity to study compliance with those sorts of regulations, and that’s definitely on our radar.
ST. The number of people who have been killed in car crashes rose for a few years. There’s lots of interest in understanding why that’s happening and what we can do to stop it.
WB. Because I-24 MOTION is permanent infrastructure, we get the benefit of seeing random or rare events out there on the roadway. We’re not in the business of crash analytics, but we are definitely interested in how traffic evolves in the wake of a crash.
We’re about to release over 100 days of traffic data from I-24 MOTION in January. That is a treasure trove to traffic researchers around the world. I believe we have users from 30 different countries and over 30 different U.S. states, if not maybe 40 by now. Tons of other researchers from academic institutions, as well as in industry and government, are all using the data from I-24 MOTION.
We hope that continues because there’s more information contained in all of that data than our team could explore in an entire lifetime. We hope that others continue to be involved with the project because that’s really going to unlock its full potential—having others involved and coming up with their own ideas and pouring their own research hours into it.
Megan Gates is editor-in-chief of Security Technology. Connect with her at [email protected] or on LinkedIn. Follow her on Bluesky.
