Cracking Classification Using Minimum Rectangular Cover–Based Support Vector Machine

Abstract

Cracking characterization is one of the most important tasks in automated pavement data analysis. Although cracking detection and segmentation algorithms have become more reliable in recent years, accurate cracking classification remains a constant challenge to pavement engineers. Conventionally, manual recognition uses cracking orientation and topological features to classify cracking into different types such as alligator cracking and transverse cracking. However, the rules to classify cracking are often complicated and subjective, which compromises the reliability of computerized implementation. This study develops a support vector machine (SVM)–based method to intelligently identify cracking types in an automated manner. Pavement cracks are grouped using a minimum rectangular cover (MRC) model. Using the relative location, orientation, and size of the MRC, as well as the cracking characteristics such as cracking density and cracking connectivity, three SVM models are compared in this study. It is found that an 88.07% accuracy is achieved for 10,134 MRCs collected from four highway sections using the two-phase SVM model. The proposed methodological framework would improve overall accuracy in cracking classification.