Computer Vision for Infrastructure Health Monitoring: Automated Detection of Pavement Rutting from Street-Level Images

The increasing availability of low-cost sensors, such as smartphones and the cameras embedded in cars for driving assistance, provides new opportunities to enhance pavement health monitoring. While computer vision techniques have been successfully used to detect pavement distresses using 2D images, distresses involving depth measurements, such as rutting, remain a challenge. The objective of this study is to develop an object detection model to automatically detect instances of rutting in pavements. This study leverages pavement images and distress data collected from the United States Federal Highway Administration Long-Term Pavement Performance (LTPP) database to train a rutting detection model. Transfer learning is used to analyze the generalizability of the model. Results indicate that street-level images are highly suitable for detecting pavement rutting using computer vision algorithms. Future research is needed to quantify the severity of rutting using image processing techniques.