Pavement Crack Extraction Based On Digital Image Processing

Road construction and development occupies a pivotal position in the transportation infrastructure in the world.However,during engineering,various damages,collapses and diseases will inevitably occur,which will bring threats and losses to social economy and traffic safety.Crack is the most common disease phenomenon on the pavement of a road.Because of the causes,shapes,and types of cracks,the damage in the pavement structure is also very different.Therefore,the detection and identification of cracks has become a hot issue,especially in the context of the rapid development of information processing and machine learning and other information technologies.Crack feature extraction based on crack images has become an important issue in both engineering and scientific research One of the most active topics.The research goal of this paper is to extract and classify cracks using road crack images.We have designed a total of three steps to achieve this goal.(1)First,after analyzing the characteristics of pavement images,the paper preprocesses the crack image as follows: For the crack image shrink,we propose a distance-weighted image shrink algorithm;for crack image smooth processing,we propose An image smoothing method based on Fractional calculus;for crack image enhancement,we study an image enhancement method that improves Fractional calculus templates.The algorithm proposed in this paper is compared with the traditional algorithms of image shrink,image smoothing and image enhancement to verify their superiority.(2)Secondly,in view of the problem of crack line shape extraction,a method for extracting feature points of crack midline based on the Steger algorithm and a method for connecting crack midline based on hydrelics are proposed,which is also experimentally compared with traditional algorithms.(3)Finally,a crack classification method based on Support Vector Machine(SVM)is proposed for crack classification,which divides crack targets into four categories.In order to assist the classification of cracks,two methods of constructing cracks image features are used: the construction method of counting features of cracks targets and the construction method of morphological features of cracks targets.Experiments show that the crack image pre-processing algorithm studied in the paper has better processing effect on crack images than that by the traditional algorithm;the proposed method of crack midline extraction can accurately extract the linear features of cracks;the designed crack classification method has high classification accuracy,and the average test accuracy concentrated in the test can reach more than 90%.