Bounding Box–Based Technique for Pavement Crack Classification and Measurement Using 1 mm 3D Laser Data

Abstract

Pavement crack classification is important for pavement distress condition rating and the subsequent maintenance and rehabilitation activities. Substantial research and field efforts have been dedicated to developing a usable solution for fully automated crack classification, but successful field applications are limited because the existing processing techniques have several deficiencies. One example is the inability of separating simply connected cracks into proper types. To overcome this and other limitations, this paper presents a four-phase bounding box–based image-segmentation technique for fully automated crack classification. First, 1 mm three-dimensional (3D) pavement data are continuously collected at high speed using an emerging pavement data collection system, on the basis of which cracks are identified with a multiseed fusion algorithm. Then a series of image processing techniques are introduced to produce terminal bounding boxes in combination with location references of wheel paths and lane markings. Finally, crack classification and severity measurement are conducted on the terminal bounding boxes. The case study demonstrates that automated results for crack classification using the proposed technique compare well with those from manual processing.