Automatic Pavement-Crack Detection and Segmentation Based on Steerable Matched Filtering and an Active Contour Model

Cracks are an important symptom of pavement deterioration and deficiency. Accurate and complete information regarding pavement cracks is critical to determining pavement maintenance schedules, methods, and budgets. Two-dimensional (2D) pavement images are used in practice for crack detection and segmentation. Automatic crack detection and segmentation based on 2D images are challenging because of (1) low contrast between cracks and surrounding pavement; (2) complicated patterns of cracks; and (3) intensity inhomogeneity along cracks. To address these challenges, this paper presents a novel method to automatically detect and segment pavement cracks from 2D images. Specifically, the proposed method starts with the use of a steerable matched filter to generate a crack saliency map, which enhances the contrast between cracks and surrounding pavement and captures crack discontinuity and curvature. Analysis of the crack saliency map leads to a coarse crack region and rough estimates of crack properties. The coarse crack region is then fed into a region-based active contour model, and a level set evolution method is employed to implement the model for crack segmentation. The estimated crack properties provide information to automatically adjust the parameters of the active contour model for effective and efficient crack segmentation. The proposed method was tested using 65 pavement images with various cracks. The proposed method achieved average precision of 92.6%, recall of 85.1%, and F-measure of 88.7%. (C) 2017 American Society of Civil Engineers.