Modern cities produce an absolute avalanche of data every second of every day, from cameras and sensors belonging to dozens of separate systems. In such a smart world, it's amazing that so many traffic lights are still so dumb. A new project beginning in Melbourne, Australia aims to harness this avalanche of data using deep learning AI and predictive modeling, and use it to reduce travel times, reduce emissions, and influence behaviors in all sorts of other ways as part of "the world's smartest traffic management system."
The "Intelligent Corridor" is a three-year trial to be run on a 2.5-km (1.5-mile) stretch of Nicholson Street, Carlton – one of Melbourne's busiest roads. The University of Melbourne has partnered with Austria's Kapsch TrafficCom and the Victorian Department of Transport, among others, on the project.
The system, launched this week, draws in live and historical data feeds from an enormous and varied network of sensors, including CCTV camera feeds, Bluetooth sensors, air quality monitors, live public transit information, TomTom live traffic data, weather data, live traffic light signal and phase data, intersection logic data, and more. Some of these data feeds are already available city-wide, others were pre-installed in the wider Carlton area as part of a separate AIMES (Australian Integrated Multimodal EcoSystem) project, which itself is described as "the world’s first and largest ecosystem for testing emerging connected transport technologies at large scale in complex urban environments."
This is the first time, claims the project team, that such a broad range of live and historic data has been brought together into a traffic management project, to be crunched in real-time by AI and deep learning algorithms. And the insights this "EcoTrafiX" system brings forth will be used for a variety of purposes.
Firstly, and perhaps most directly relevant to you and I, it'll be able to control the traffic lights at every intersection in the corridor for optimal flow. At the project launch, Kapsch's David Bolt gave an example: "We're taking a video stream from one of the hundreds and thousands of cameras across the network, we're using our deep learning versatile platform to analyze and annotate it, and we're then forming insights. We're looking at queue length detection, for example, at a lane level. That influences the signal phase and timing. I can start to dynamically adjust and optimize this intersection, and further up, other intersections down the corridor."
Getting the dynamic traffic light logic right won't just reduce frustration for car commuters – every stop and start takes its own toll on the city in the form of noise and emissions, particularly where heavy cargo trucks are involved.
But things go far beyond that. The system has a number of ways to communicate back to road and public transit users, to influence traffic flow either in response to an incident, or simply as a means of load-balancing and optimizing things. If an accident prevents trams from getting through a certain intersection, the system can match each upcoming tram it's likely to affect with recommendations that'll get its passengers where they're headed, either by switching trams or by taking a short walk to another way in, and then get that message through to the tram driver.
There's also a safety angle – the system can monitor pedestrian crossing zones and give feedback to drivers through infrastructure-to-vehicle communications to connected cars. One example the team chose to highlight was a particular intersection at which drivers turning a tight corner can't see people crossing the road until they're almost upon them – real-time warnings will now be sent to certain connected Lexus cars if the system finds drivers are about to run into this kind of situation.
Incident management will be a big part of the system, which will pop up operator warnings when it detects something strange is going on, or predicts a problem emerging. Operators will be able to choose from a list of auto-generated actions in response to a situation, or else dig straight into the data streams, right down to the level of viewing live camera footage, to figure out what's happening. They'll also be able to call up similar incidents from the history of the area, including what actions were taken in the past and what the flow-on effects were, and the system is designed to allow as much, or as little, human oversight and interference as a particular city wants.
These are the kinds of things the system can do today, on launch. But over the course of the next three years, the team expects to trial all sorts of ideas, from things like making sure connected emergency vehicles see nothing but green lights, to intelligently routing traffic around school zones at pickup/drop-off times, to re-routing traffic in response to air quality mapping, to auto-texting owners of cars that are parked in clearways in the hope that they'll shift their own vehicles and clear the road faster than tow trucks can.
"We're able to communicate to drivers through the APIs and plugins that we've bolted onto the platform," said Bolt. "It's a social engineering challenge to change the habits of how you drive ... This is all about preparing the infrastructure for what's needed going forward. So how do we send out information to connected vehicles? How's that information being sent to non-connected vehicles? How do we prepare for autonomous vehicles?"
The project will naturally capture before-and-after data to measure and track the system's effectiveness. Kapsch says the system is designed to scale from small single intersection and short corridor deployments like this one, all the way up to massive city-wide implementations, since it's location-independent and can work with whatever data is available.
This certainly seems like an excellent and sorely needed use of AI and advanced analytics. City-wide traffic optimization is a data-heavy problem with an extraordinary number of inputs, dependencies and outcomes to track. There's a huge opportunity to do it better, plenty to be gained, and if AI and computer science are up to the task, we're fascinated to see what will come of this.
You can watch the Intelligent Corridor launch event in its entirety in the video below.
Source: Melbourne University / Kapsch
