Research On The Technology Of Automatic Pavement Crack Detection Based On Digital Image Processing

With the development of economy, the highway is developed quickly in ourcountry. Thus, all of these reasons bring a higher acquirement to the work of highwaymaintenance. After the construction, some pavement cracks of different levels areusually caused by many reasons, such as climate, geological condition, traffic volume,load capacity and so on. Therefore, the regular inspection and maintenance forhighway should be made by relevant departments. In our country, the traditionalmanual inspection methods are used widely, which have many disadvantages such aslow efficiency, big errors, dangerous, negative influence to the traffic frequently. Inorder to save the maintenance cost and prolong the service life, the study of pavementautomatic detection technology based on digital image processing is focused on. As aconsequence, the service level of the highway is also improved.This thesis has mainly studied crack image processing technology, whichincludes image preprocessing, image segmentation and feature extraction. In imagepreprocessing, the original images have been reduced to25percent by using thenearest neighbor interpolation. The median filtering has been carried out in turn byusing four different types of structural elements so as to complete the image denoising.The grayscale correction algorithm, which is based on the image backgroundextraction, has been adopted to improve uneven illumination of the image. In imagesegmentation, the Ostu threshold segmentation algorithm has been used to segmentthe crack image, meanwhile the threshold value has been improved. The thresholddenoising algorithm of white pixel domain has been carried out to remove the noise ofthe binary images. The mathematical morphology and white pixel threshold denoisingalgorithm have been used alternately to extract the crack, finally, the iterativeelaboration method has been adopted in the thesis. In feature extraction, cracks pixelarea, vertical projection, horizontal projection and the rectangular degree have beenselected as the characteristics of different types of cracks. Cracks pixel area has beendemonstrated to be a proper value that makes an accurate judgment in the imagewithout cracks. In the thesis, four properties of the crack are considered as the eigenvalue vectorsof the Support Vector Machine (SVM). Ninety-five test sample have been identifiedby the SVM based on RBF, in which the―one-against-the rest‖multiple classificationalgorithm has been adopted. As a result, the accuracy is85.26percent. At the last ofthesis, a brief analysis of the error reason has been provided. In addition, acomparison has been made between Back Propagation Neural Network (BPNN) andSVM under the same sample set. It is shown that SVM has higher accuracy thanBPNN.In order to meet the crack parameters needs of road maintenance, the length oftransverse and longitudinal cracks and the minimum circumscribed rectangle area ofblock and reticular cracks have been selected and calculated. According to theseparameters, the degree of pavement damage level is obtained by calculating the PCI,so that the pavement damage has been given and several related suggestions havebeen put forward.