How UNT is helping TxDOT use AI to detect debris, hazardous conditions on roadways
For the past four months, researchers at the University of North Texas have been using artificial intelligence images of traffic debris and carcasses to help detect road hazards.
"Every year, there's massive tons of debris and thousands of animal carcasses end up on the highway system," Dr. Yan Huang said. "So, at a very high speed, a single segment of tire will cause a vehicle to swirl and secondary accidents, and sometimes miles of traffic."
Huang is a professor of computer science in UNT's College of Engineering. She's working on the AI project and isn't a fan of traffic backups, which can be caused by debris on the road.
In partnership with the Texas A&M Transportation Institute, UNT is using a $400,000 grant to focus on how the technology could make commutes safer and, if needed, response times more efficient.
"So, the traditional systems require a human to report, like hazard, debris, either through emergency calls (or some crowd-source application, like Waze," Huang said.
How this AI project could help TxDOT
The two-year project will bring AI to a level of recognition that enables it to spot and assess risk for the Texas Department of Transportation. According to researchers, AI would send the information to TxDOT to make response decisions.
"It can recognize dogs, animals, they can recognize pedestrians versus vehicles or bicyclists, for example, and then it has a better understanding of that context," Minh Lee said.
Lee is with the TAMU Transportation Institute. Part of the data they are pulling from comes from the traffic app Waze.
"And what we have found for road debris, as opposed to accidents, road hazards are reported much more in those types of, in that technology versus people calling them in because it's not as major as an accident or if they're involved in an accident," Lee said.
How often is debris detected in North Texas?
Lee said in their review of Dallas, they found drivers reported debris and hazardous roadway problems three times more than similar traffic management center detects. Additionally, he said Waze detected the incidents 72% faster and nearly 16 minutes earlier.
"That's just one data source that we're using for this project. It's to help us to quickly identify where those potential debris are," he said. "We're going to use the other data sources too, and that's where the AI comes in. We're going to use the other data sources for the video analytics."
The AI is also being trained using images captured by cameras plugged into the system. Some may dismiss debris as a small thing, but Huang encountered an incident during her work.
"Yeah, I think I was driving to TxDOT, and then, I mean, just a, I think it's half of the tire from an 18-wheeler just blow off," she said. "Kind of rolling on the highway in front, and then there's one car ahead of me. That car was trying to dodge it."
How will it benefit drivers?
The work will connect to TxDOT's 3,500 cameras and begin zooming in on debris, even during rush hour traffic.
UNT Assistant Professor Hen Fan said there are a few limitations at the moment. One, deciding how molecular to go with debris. The other issues are visibility in bad weather and at night.
"One of the limitations for the system is it's very hard to detect the debris with limited vision," Fan said.
They anticipate progress with accuracy by this summer.
"We'll have a functional system that's basically AI get trained and tested, and we know some, we have some confidence about the accuracy of the system," Huang said.