Integrated Processing of Image and GPR Data for Automated Pothole Detection

Abstract

A pothole is a severe pavement distress that can compromise pavement rideability and safety and can be the cause of expensive damage claims. The detection and evaluation of potholes are predominantly manual and time-consuming. Although sensing technologies such as global positioning systems (GPS), stereovision systems, and ground penetrating radar (GPR) now can be combined to collect pavement condition data for assessment, the raw data returned by these sensors are often processed individually and separately. This isolated approach to data processing hinders the potential efficiency and effectiveness of multisensor systems. This paper proposes a method to integrate the processing of two-dimensional images and GPR data to automate accurate and efficient pothole detection. First, the images and GPR scans are preprocessed to filter out noise and enhance the essential clues related to potholes. Second, a novel pothole detector was designed by investigating the patterns of GPR signals reflected by potholes. Third, the position and dimension of the detected pothole can be estimated from GPR data and mapped to the image to enable a localized shape segmentation. The proposed method was validated through 50 experiments. The precision, recall, and accuracy achieved were 94.7, 90, and 88%, respectively. The mean and standard deviation of error percentage in pothole shape extraction were 12.8 and 6.5%, respectively. The method and results reported in this paper demonstrate that integrated and complementary processing of multisensory data can be achieved by channeling data streams and linking data processing according to the merits of the individual sensors.