Artificial intelligence may help cars detect potholes before you hit them

GOYANG, South Korea — It never feels good running over a pothole at high speed. It’s even worse for your car and could even lead to an accident on busy roadways. Now, researchers in South Korea are using artificial intelligence to help you dodge these car-killing craters.

A team from the Korea Institute of Civil Engineering and Building Technology (KICT) say they’ve developed an “AI-based automatic pothole detection system” that scans the road surface for problems in real-time. That information can help local governments get out and pave over these problem areas before an accident occurs.

In South Korea alone, potholes are a major issue during the country’s rainy season. In August 2020, drivers reported over 7,000 potholes during a record downpour in the city of Seoul. Across all of Korea between 2016 and 2018, there were over 650,000 potholes on roadways. Moreover, estimates put the price tag for the damages potholes caused in South Korea during that time at nearly $4 million.

“It is essential to maintain road facilities in good condition in the coming era of autonomous vehicles. This AI-based technology will make effective road surface management much easier,” says Dr. Seungki Ryu in a media release.

Improving how we fix broken roadways

Keeping the roads drivable starts with quickly finding damage, like potholes, using laser scanning and other image recognition technology. Dr. Ryu’s team developed a system that detects potholes using real-time photographing of the road surface while driving with a vision sensor on a car’s windshield. The AI model then segments damaged portions using an encoder-decoder network.

A common problem with image-based detection, however, is that the images can vary in pixel quality depending on the surrounding environment. Specifically, even AI programs can have trouble spotting potholes as the brightness on road surfaces over time.

To fix this, study authors created a new AI network called convolutional neural network that combines image preprocessing and the older segmentation model to improve the performance of pothole detection. The technology uses a mobile app to gather data and sends it to a map-based cloud server platform to identify potholes. Several local governments in South Korea are already testing the technology and Dr. Ryu’s team hopes the rest of the country will try this new-age approach to road management soon.

The study appears in the journal Electronics.

Follow on Google News

About the Author

Chris Melore

Chris Melore has been a writer, researcher, editor, and producer in the New York-area since 2006. He won a local Emmy award for his work in sports television in 2011.

The contents of this website do not constitute advice and are provided for informational purposes only. See our full disclaimer