| dc.description.abstract | The use of three-dimensional (3D) pavement distress detection mechanisms has become a trend in pavement surface condition evaluation, including crack classifications and ratings. Inaccurate crack classification could affect indicators of crack detection, as represented by the area of each cracking region. Therefore, development of intelligent crack detection models are necessary and important. This paper focuses on irregular cracks, such as linear and netted types. The primary purpose of the research summarized in the paper is to develop a multiscale clustering model by taking advantage of 3D pavement surface images. A preprocessing procedure is carried out to obtain an initially identified crack image (by a CrackSeed-based approach). A cracking region is then characterized by a minimum circumscribed rectangle as the basic unit of the clustering. Then multiscale features are used as the crack unit. Linear crack categories and intensities are classified according to an orientation angle to the length-width ratio, while netted cracks are recognized on the basis of block features. Meanwhile, the optimized spatial distance between the crack subblocks is calculated to precisely locate pavement crack. Finally, a multiscale criterion is applied to quantitatively analyze the degree of road surface damage. In classifying transverse cracks, longitudinal cracks, block cracks, and alligator cracks in 320 crack images, the classification precision is 92.8%, 95.2%, 87.8%, and 97.6%, respectively. Experimental results indicate that the proposed method has good precision for 3D pavement crack detection and classification and could provide reliable suggestions for pavement protective maintenance operations to extend service life. | |
