Research On Complex Pavement Crack Detection And Algorithm Evaluation Based On Improved Niblack And Radon Related Technologies

With the rapid development of highway, the pavement crack detection based on digital image processing technology has become a hot and difficult research. The crack detection work will encounter bad road condition frequently in actual application, then the acquired image will contain some complex background, such as uneven illumination, oil stain, shadow of obstacle, random noise, and other factor. In this case, the normal crack extraction algorithm can not meet the demand. Many important locations are missing because the useful data is not available.Aimed at solving these problems, this paper mainly describes the four key technologies,such as image enhancement, crack information extraction, crack connection and calculation of crack parameters, crack algorithms evaluation.(1) In view of the characteristics of the pavement crack image with many noise and uneven contrast, this paper use the algorithm based on fractional integration to enhance the pavement crack image.Crack images enhanced not only enhances the image contrast, retained crack image edge and texture details, but also do not produce an image blurring.(2) The crack extraction algorithm which based on improved Niblack and connected domain histogram characteristics for the images with the unevenness background and shadow is proposed. Analysis of the shadow image characteristics, the Niblack algorithm is improved by use of the statistical characteristics and the correlation between spatial information and gray information, then the crack is segmented; Next, the crude extracted crack images is extracted further through the connected domain distance-standard deviation feature.(3) The maximum entropy threshold and distance and direction of the cracks are used to connect the extracted cracks, and then this paper analyzes the length and width of the cracks after connection. Wherein, the crack length measurement is based on skeleton extraction and the width is based on analysis of second-order moment Ferret, and the method of measuring is more accurate.(4) The crack extraction algorithm proposed is evaluated. First the extracted cracks are classified using a combination of Radon transform and clustering method, the classification method is simple and rapid, and more accurate. Then the algorithm is evaluated, this section isdivided into two parts:(1) Because the proposed algorithm is generally applied to low contrast images, background non-uniform and even shadow images. Therefore, in order to further analyze the superiority of this algorithm, three representative images is selected. The proposed algorithm is compared with classic and popular algorithms to embody the advantages of the proposed algorithm;(2) According to the type of cracks, the proposed algorithm is evaluated by F, the results show that the proposed algorithm is good for uneven background and shadow images.