AI improves safety, efficiency of traffic counting

Artificial intelligence is making a big change in traffic counting and classification.

After state, regional and local agencies spent years counting traffic using rubber hoses and classifying the types of vehicles by hand, electronic camera and computerized systems have made the chore easier, safer and more effective, officials say.

The Southwestern Pennsylvania Commission recently purchased four of the iTHEIA traffic-counting systems from International Road Dynamics, one of several vendors. After waiting for several months due to supply-chain issues, the commission deployed the systems — automatic cameras linked to a central computer — at six locations last month as a test run for using them regularly next year.

“We were able to get them out before the time expired for the year,” said Cort McCombs, manager of traffic counting for the SPC. “We were able to get a good test run this fall, so we’re ready to go full bore next spring.”

Under federal regulations, states must conduct traffic counts every three to five years on all roads that are eligible to receive federal funds. In this area, the state Department of Transportation has SPC do the count for a 10-county area, but other agencies and PennDOT itself do the counts in other areas from March to November every year.

Previously, counts were done in six-hour increments because personnel would need two hours to set up and take down the hoses. While the system was in place, the rubber hoses would count the number of vehicles, but one or two employees would be just off the roadway marking down the kinds of vehicles driving by, placing them in one of 13 categories from motorcycles to seven-axle trucks.

“Before we had these cameras, we had people on the side of the road,” said Jeremy Freeland, division manager for planning and research at PennDOT’s central office. “That was tedious. That was dangerous. We decided we wanted to avoid any injuries to our staff.”

When IRD and other vendors developed the automated systems, PennDOT reviewed them and decided to buy two a couple years ago. Then it held demonstrations at several locations across the state for other agencies.

“We were immediately interested and thought, ‘This is awesome,’” McCombs said. “This is how we should count traffic in the future.”

The agency waited for the next generation of the automated equipment before it spent about $60,000 on four of the systems. The new system takes about a half-hour to set up and can record traffic counts and the vehicles that pass through an area for 24 hours.

Developers have uploaded thousands of images of vehicles for computers to use to identify the type passing a particular location.

McCombs and Freeland cited several advantages: Tests have shown the computer recognition of the type of vehicle is about 3% more accurate than humans; the longer deployment gives a more complete picture of traffic in an area; and employees can be assigned other, safer work while the machines count traffic.

That information on traffic and vehicle use is invaluable when officials look for funding to improve road conditions, McCombs said.

“That’s the first thing that’s asked: What’s the traffic look like?” he said. “It’s the groundwork for everything else.”

Freeland said the state also uses the types of vehicles that use roads to determine the best surface to use. A road with mostly car traffic might get a different surface than an area with heavy commercial and tractor-trailer traffic.

Also, traffic patterns can help crews determine the best time of day to do work.

McCombs said the initial cost of about $15,000 for the AI system is a big increase from $500 to $600 for the previous systems, but the agency believes the new system is worth it.

“It’s initially prohibitive, but then it pays for itself,” he said.

Freeland agreed.

“It’s very new,” he said. “You can see it is a great tool to have.”