Road Crack Detection Using Visual Features Extracted by Gabor Filters

Abstract

Pavement management systems require detailed information of the current state of the roads to take appropriate actions to optimize expenditure on maintenance and rehabilitation. In particular, the presence of cracks is a cardinal aspect to be considered. This article presents a solution based on an instrumented vehicle equipped with an imaging system, two Inertial Profilers, a Differential Global Positioning System, and a webcam. Information about the state of the road is acquired at normal road speed. A method based on the use of Gabor filters is used to detect the longitudinal and transverse cracks. The methodologies used to create Gabor filter banks and the use of the filtered images as descriptors for subsequent classifiers are discussed in detail. Three different methodologies for setting the threshold of the classifiers are also evaluated. Finally, an AdaBoost algorithm is used for selecting and combining the classifiers, thus improving the results provided by a single classifier. A large database has been acquired and used to train and test the proposed system and methods, and suitable results have been obtained in comparison with other reference works. C 2013 Computer-Aided Civil and Infrastructure Engineering.

Review of remote sensing methodologies for pavement management and assessment

“…The identification and classification of pavement distress is then interpreted from the binary image [42,69,97,148]. Wavelet and Fourier transforms as well as segmentation algorithms are common image processing techniques used for pavement and concrete assessment.…”

Enhanced automatic detection of road surface cracks by combining 2D/3D image processing techniques

“…The following main procedures to process images from LRIS were presented by the authors in a previous work [3]. First, the images are preprocessed to correct the average brightness level of each column.…”

“…The first strong classifier is composed of 56 Gabor filters and is used for detecting transverse cracks, while the other strong classifier is composed of 9 filters and is used for detecting longitudinal cracks. This method was tested in [3] with a large set of images obtaining a balanced accuracy of 93.8% for transverse cracks and 90.9% for longitudinal cracks.…”

“…Image processing results obtained in previous works, [3,4], have been improved by incorporating geometric information obtained from the 3D profile of the road.…”

“…Oliveira and Correia [12] use the mean and standard deviation of grayscale values to train an unsupervised classifier; Cord and Chambon [13] apply the AdaBoost algorithm to textural descriptors; Zalama et al [3] use features extracted from filtered images by Gabor filters for supervised classification.…”