Pavement distress image recognition using k-means and classification algorithms

Road pavement maintenance today relies mainly on manual pavement condition inspection and distress rating; however, this manual method is costly, labour-intensive, time-consuming, and dangerous to inspectors and may affect traffic flows. Moreover, such method is very subjective and may have a high degree of variability, being unable to provide meaningful information. Additionally, since using the manual method can only sample a small area of the road surface, it may result in relatively low accuracy of pavement distress information. Hence, an automatic inspection system for pavement distress images is needed in hope of resolving the above problems. This paper presents a novel method to classify pavement distress images. The system was installed on a pavement inspection vehicle with image acquiring devices. First, pavement images were processed to show only black-and-white pixels that can render true pavement cracks. Then, the pavement images were transformed into a set of clusters in order to capture the distress location of each crack. Next, the distress types, i.e., horizontal, vertical, alligator-like, or man-hole-like, were obtained by applying a decision tree algorithm. Finally, the system stored the data and images into a database and provided spatial query functions for users to retrieve crack information. The present results show that our method can successfully recognize various types of pavement distress. Our system also provides information regarding pavement crack lifecycle, i.e., the times when a crack was identified and when it was fixed, etc., so that public road authorities can define maintenance plans in accordance with real pavement conditions.