Why sensing traffic is no longer enough

Mandar Chincholkar, General Manager, Smart Transportation Edge AI at Intel, explains why cities need to move beyond sensing and reporting towards transport infrastructure that can anticipate problems and support better decisions.

For years, cities have invested heavily in connected transport infrastructure. Cameras, sensors and traffic signals now generate enormous quantities of information about how networks operate.

But collecting more data does not necessarily make infrastructure more intelligent.

The challenge is what happens next. Looking back at data to understand why congestion occurred yesterday has value, but it does not prevent the same problem happening tomorrow. The next step is cognitive infrastructure that can perceive what is happening, anticipate what could happen next and help people make decisions before a problem develops.

Consider congestion at one intersection that begins affecting five or six intersections around it. Existing systems can tell a traffic manager that vehicles are backing up. A cognitive system could recognise the conditions leading to congestion, understand the likely knock-on effects and help determine how traffic could be managed differently before the network becomes gridlocked.

The same principle applies to flooding. If a city knows a road is vulnerable when rainfall reaches a certain level, why wait until it floods? Combining historical information, forecasts and conditions on the ground could allow traffic to be redirected before disruption occurs.

That is the fundamental shift: from reacting to anticipating.

Bringing intelligence closer to the problem

Achieving this means thinking carefully about where computing happens.

Historically, much of the data generated by connected infrastructure has been sent to the cloud. The cloud remains essential, but sending everything to a centralised environment is not always efficient when decisions may need to be made immediately.

Increasingly capable AI-enabled edge computing allows information to be processed closer to the cameras, sensors and intersections generating it.

At Intel, we are exploring this through work involving cities and transport authorities in the US and Europe. In Austin, Texas, work with the University of Texas at Austin is examining connected corridors, linking multiple intersections and looking at traffic flows across a wider area rather than treating each junction in isolation.

We are also working with Santa Clara County in California, including on emergency vehicle pre-emption. If a fire engine approaches a congested intersection, giving it a green light is only part of the challenge. The system also needs to consider what that intervention means for surrounding traffic and how disruption elsewhere can be minimised.

In Sweden, work with AstaZero is exploring similar questions around connected transport infrastructure. We are looking at how conventional video analytics can be combined with vision language models and large language models to bring together different sources of information and provide greater context.

Get more from what cities already have

Moving towards cognitive infrastructure should not require cities to rip out everything they already have.

Cities have invested heavily in cameras, sensors and roadside equipment. Some will inevitably be too old or limited, but much of that infrastructure can potentially be made more capable.

For example, we are exploring techniques such as super resolution, where AI can improve the fidelity of images from lower-resolution cameras before systems perform tasks such as object detection and classification. That could provide cities with new capabilities without replacing every camera in the field.

We also need to be clear about the role of AI. I see it as a tool to augment people, not replace them.

A transport control centre operator might be responsible for dozens of camera feeds. Much of the time, those cameras show vehicles moving normally or simply empty pavement. Asking people to monitor every feed continuously is inefficient and leads to fatigue.

AI can instead help direct attention towards events that matter, potentially identifying a developing situation and bringing it to an operator’s attention. The human remains responsible for the decision but has better information on which to act.

Plan for the next decade

There is also an economic argument for getting the architecture right. Cloud computing, data transport and AI processing all carry costs, particularly as systems scale.

The answer is not to move everything to the edge. Cities will continue to need backend infrastructure connecting different locations and supporting wider analysis. The challenge is deciding what needs to happen centrally and what can be handled locally. It is an “and”, not an “or”.

That matters as city leaders look five or ten years ahead. Much of today’s transport infrastructure was designed around assumptions made years or even decades ago, while population growth and changing travel patterns are putting increasing pressure on networks.

Technology offers an opportunity to get more from those existing assets, but cities need to look beyond individual projects and procurement cycles. What problems will become more acute over the next decade? What computing and data infrastructure will be needed? And can investments made today provide a foundation that evolves as AI capabilities improve?

Cognitive transport infrastructure is not about handing control of our streets to AI. It is about giving the people responsible for those streets a better ability to understand what is happening and act sooner.

We have spent years getting better at sensing our transport networks. Now we need to get better at using what they tell us.

Main image: Michaelurmann – Dreamstime