Using AI, cars can detect potholes in real time

The Korea Institute of Civil Engineering and Building Technology (KICT) has announced the improvement of an 'AI-based automatic pothole detection system'. The strategy is designed to beryllium installed connected the windshield of a conveyance to observe potholes connected the roadworthy aboveground successful real-time. Potholes tin harm cars and whitethorn adjacent pb to life-threatening accidents.

In particular, potholes tin origin problems successful the rainy seasons. In Seoul, erstwhile the grounds was acceptable for dense rainfall successful August 2020, much than 7,000 pothole reports were received successful August 2020 erstwhile a grounds downpour deed the city. In Korea, the fig of potholes reported crossed state from 2016 to 2018 was 657,993. Total harm compensation amounted to 4.6 cardinal won nationwide, and road repair costed 1.7 trillion won. If a conveyance passes implicit a pothole astatine precocious velocity without noticing it, the conveyance whitethorn interruption distant from the driving lane, hence a menace to the driver's life.

Road surface absorption starts with rapidly detecting damaged sections, and this involves vibration-, laser scanning-, and image recognition-based detection technologies. In particular, with the caller improvement of the detection exertion utilizing deep neural networks, representation recognition-based road surface absorption methods are receiving attention. In addition, the image-based method tin beryllium utilized with idiosyncratic devices, specified arsenic a conveyance oregon smartphone camera. This makes it easier for local governments, which person relied connected ocular inspections by humans, to employment the technology.

A probe squad astatine KICT led by Dr. Seungki Ryu developed this strategy that detects potholes successful real-time by photographing the roadworthy aboveground portion driving with a imaginativeness sensor installed connected the windshield of a vehicle. The AI inference exemplary semantically segments damages connected the roadworthy aboveground utilizing an encoder-decoder web based connected the FCN (fully convolutional neural network) architecture.

A communal occupation successful image-based detection is that adjacent astatine the aforesaid determination images tin alteration successful the pixel portion accusation depending connected changes successful the outer environment. In particular, it whitethorn beryllium challenging to place damages connected the roadworthy aboveground with the AI inference exemplary arsenic the brightness of the roadworthy aboveground changes implicit time. To lick this issue, a caller CNN (convolutional neural network) exemplary for representation preprocessing was developed and combined with the semantic segmentation exemplary to amusement robust detection show with roadworthy images taken nether antithetic brightness conditions.

This exertion consists of a mobile app for gathering information utilizing the AI exemplary and a map-based unreality server level to place potholes based connected information transmitted from the mobile app. Currently, respective section governments successful Korea, specified arsenic Gwangju Metropolitan City, Goyang-si, and Gimhae-si are piloting this technology. Dr. Ryu's probe squad aims to further grow and present this exertion to different section governments.

Dr. Ryu said, "It is indispensable to support roadworthy facilities successful bully information successful the coming epoch of autonomous vehicles. This AI-based exertion volition marque effectual roadworthy aboveground absorption overmuch easier."

---

---

More information: Taehee Lee et al, CNN-Based Road-Surface Crack Detection Model That Responds to Brightness Changes, Electronics (2021). DOI: 10.3390/electronics10121402

Provided by National Research Council of Science & Technology