Automated Detection of Road Defects Using LSTM and Random Forest

Road networks represent an essential factor in economic prosperity and well-being. Assorted damages and defects on the road surface mainly impact the transportation efficiency of road networks and driving safety. In current traditions, it may take weeks or even months before competent government departments repair such road conditions due to their lack of timely awareness. This research proposes a framework for collecting data and analysis using supervised machine learning (ML) techniques, more specifically Long/Short-Term Memory (LSTM), Random Forest (RF), and networked sensor-enhanced vehicles for effective and efficient monitoring and assessment of road surface conditions. Data collected in four different scenarios and multiple vehicle data help to discover abnormal defects and verify automated detection. ML techniques were developed to identify bumps and potholes automatically, which are the two significant types of road defects. The experiment results of both LSTM and RF models can automatically identify bumps and pothole defects with an accuracy rate of up to 99.8% and a good detection effect. Cross-validation was set to validate the accuracy of the two decision-making strategies, LSTM and RF. Automation technology is expected to provide qualitative and quantitative information about road defect conditions, enabling timely maintenance to improve transportation efficiency and driving safety.