Pothole-Finding AI Could Make It Easier to Assess Roads

Potholes are a costly hazard for drivers, and finding them can be a costly enterprise for cities. A Canadian research team has developed an AI system that finds them automatically in photos.

University of Waterloo researchers have developed software that finds potholes automatically. (Image courtesy of Metropolitan Transportation Commission.)

University of Waterloo researchers have developed software that finds potholes automatically. (Image courtesy of Metropolitan Transportation Commission.)

Researchers have developed an AI software system that can automatically analyze photos of roads to locate potholes and cracks in the pavement, and help cities fill them faster.

Currently, there are two basic approaches to collecting pothole data. In smaller towns, workers simply drive around and inspect roads visually for signs of cracks or potholes. Larger cities often use truck-mounted cameras to take images of roads, which are then analyzed by teams of experts. And while the second approach is slightly more scientific than the first, it’s not perfect. “It’s time consuming and very expensive,” lead researcher John Zelek told Waterloo paper The Record. “You don’t get consistency because there is some subjectivity.”

The researchers started their project by using on-the-ground photos from Google Street View. As the project progressed, they used images from different sources, including some from a company that uses footage from vehicle-mounted cameras to identify pavement defects. In the real world, municipalities could use videos taken from onboard cameras on city vehicles that capture footage as workers drive around doing their daily business. The cameras do not have to be particularly hi-res; indeed, phone footage is enough for the software to detect potholes.

The researchers say that their program should lead to timelier and less expensive repair assessments, and it will be at least as accurate as human analysis, if not more so. “It is more consistent analysis because you’re not introducing the biases of different human beings who look at the data differently,” Zelek said. 

The researchers are very optimistic about their project’s possibilities: “If governments have that information, they can better plan when to repair a particular road and do it at a lower cost,” said Zelek, “Essentially, it could mean lower taxes for residents.” Indeed, potholes are costly for governments. According to the U.S. Congress’ National Surface Transportation Policy and Revenue Study Commission, the government sinks $68 billion per year into just maintaining public road, rail, and transportation systems. And that’s not the only place that potholes cost people: the American Automobile Association says that drivers are paying $6.4 billion per year in repairs specifically caused by pothole damage. 

The researchers also hope that their software can use drone footage of bridges, buildings and other infrastructure projects to check for any necessary repairs, or be used during the building process to identify any difficulties in construction.

As Zelek said, “It would make everybody’s lives a lot better.”