Pavement Crack Detection and Localization using Convolutional Neural Networks (CNNs)

Abstract

Regular inspection and maintenance of roads are required to ensure safe transportation. While examining the state of structural health, cracks are considered as the primary indicators. In the past decades, researchers have been working on various image-based pavement crack detection techniques for non-destructive evaluation. The main advantages of these techniques over manual inspection are accuracy, efficiency and cost. However, the problems associated with the existing methods are their dependence on the handcrafted features, which may not give accurate results due to insufficient feature selection. In this paper, an automatic image-based crack detection algorithm for pavement crack detection using Convolutional Neural Network is proposed. The data set was obtained from various road surfaces of United Arab Emirates (UAE) by using an unmanned aerial vehicles (UAVs) and was used in training and validation of the proposed system. The collected data was also used to create a composite view of the road by creating a continuous mosaic. From the experimental results, it was found that the proposed system has an accuracy of 92% in the validation stage and 90% in the testing stage and can be used for crack detection of road surfaces.