An artificial intelligence (AI)-based automatic pothole detection system has been developed by researchers at the Korea Institute of Civil Engineering and Building Technology (KICT).

The KICT system reportedly detects potholes in real-time on the surface of roads via a vision sensor that is affixed to the vehicle’s windshield. As the car moves, the road’s surface is photographed and an AI inference model segments surface damages using an encoder-decoder network derived from fully convolutional neural (FCN) network architecture.

Overall flow of the developed automatic pothole detection system. Source: KICT

A mobile app gathers data via the AI model while a map-based cloud server platform identifies potholes according to that data, thereby improving road surface management.

Currently, image-based detection methods for maintaining road surfaces can vary — even at the same location — in the pixel unit information according to changes in the external environment. For instance, the AI inference model may not be able to identify damage to the road surface because the brightness of the road surface transforms over time.

As such, the researchers developed a convolutional neural network (CNN) model for image pre-processing and combined it with the segment model to demonstrate detection performance with images of the road surfaces captured under various brightness conditions.

The study — CNN-Based Road-Surface Crack Detection Model That Responds to Brightness Changes — appears in the journal Electronics.