Smart pothole detection system using vehicle-mounted sensors and machine learning

Abstract

Road networks are critical assets supporting economies and communities. Despite budget and time constraints, road authorities strive to maintain them to ensure safety, ongoing service, and economic productivity. This paper proposes a virtual road network inspector (VRNI), which continuously monitors road conditions and provides decision support to managers and engineers. VRNI uses acceleration data from vehicle-mounted sensors to assess road conditions. It proposes a novel road damage detection method based on two adaptive one-class support vector machine models, which were applied on the vertical and lateral acceleration data. We evaluated this method on data from a real deployment on school buses in New South Wales, Australia. Experimental results show that our method consistently detects 97.5% of the road damage with a 4% false alarm rate that relate to benign anomalies such as expansion joints.