Research On Automatic Pavement Crack Identification System Based On Digital Image Processing

As a common pavement problem, pavement cracks always attract high level of attentionfrom highway department of maintenance and preservation. By collecting details of pavementcracks rapidly and accurately, relevant plans of maintenance can be established. However,because of the weaknesses, such as inaccuracy, low-efficiency, high consumption of time andmoney, inconvenience and potential danger, the traditional manual road crack detectionmethod is not suitable for measurement implemented in the field of pavement transportationwhich develops very fast currently. Therefore, further study on the aspect of automaticdetection technology of pavement cracks becomes more and more important in our country.Digital image processing technology, which generally includes image enhancement,segmentation, feature extraction and so on, is the main study of this paper and the core part ofthis software. Based on the features of pavement images taken by CCD cameras, the uniformlight method, which is simple, is adopted to correct mistakes existing in uneven images in thispaper. Then the pavement crack image is processed by using the weighted average filter anddajin threshold segmentation method respectively. After that, the binary image with a smallamount of isolated noise or oil pollution is obtained. In order to identify and analyze theimage, grid black pixel threshold method and the biggest connected area of grid are used toeliminate the image noise. Thus, binary image with little noise is gotten. By using the thinningmethod, the characteristics of crack image can be extracted. In the step of characteristicextraction, the original image is divided into several sub-blocks, and four parameters:(x_max, y_max, sum, d)are obtained. At the same time, through the statistics of pixel, theparameter of all pavement cracks are calculated, which can provide the foundation for thepreservation by highway department.At last, the BP neural network is used to classify pavement cracks. The feature vector isused as the input to train the classifier, and the output of BP neural network includs4types ofpavement cracks, namely the transverse crack, longitudinal crack, netty crack and nubblycracks. In this paper,84different types of crack images are used to train the trained neuralnetwork. The classifier gets a better recognition. By using the classifier, classification of96test examples is realized and the recognition rate of87.5%is acquired, the experimentalresults show that the method is effective.