An Asphalt Pavement Crack Identification Key Technologies Based On Image Analysis

Asphalt pavement crack detection is a challenging work in the area of traffic information engineering and pattern recognition. This paper is based on the image data collected by the "CT-501A" high-speed laser road testing car of Chang'an University, and the thesis studied the method of asphalt pavement crack detection using the theory of the mathematical morphology and the method of image classification of support vector machine through the analysis of image features of asphalt pavement disease.According to the characteristics of pavement cracks, we choose to use different structural elements of different sizes and shape on the asphalt pavement images to analyze the influence of structural elements when applying the morphological edge detection operator. This paper adopts algorithms of improved morphological edge detection, has precisly detected edge details of the crack images by choosing structure element of 9x9 diamonds to deal with the batch of asphalt pavement images; adopts iteration threshold segmentation method to segment detecting results to obtain the binary images with smaller noise points; adopts 3x3 horizontal structure elements and the method of morphology to denoise the batch of binary images, and adopts projection and interval sampling to extract features of denoised binary images. The paper studies algorithm of support vector machine based on the asphalt pavement image classification. Experiments are carried out compare four kinds of kernel function of support vector machine. Choosing RBF kernel function which has the higher classification accuracy to implement asphalt pavement images classification based on support vector machine. It makes asphalt pavement images into two types:cracks and no cracks.Through experimental analysis of 260 asphalt pavement images, the proposed approach is proved to be effective for detecting crack edges, with strong ability to denoise, it can reach classification accuracy rate of 94.4278%. The scientific data provided is used for road evaluation and maintenance. The work is significant to increase the automation level of the road detection of our country.