Research On Roda Cracks' Recognition Based On Image Processing

As China's rapid economic development, transportation's importance in the national economy and people's lives become more and more obevious. Highway Traffic is in an important position, and it's also been paid more attention with the rapid development of China's economy. In order to improve the road service life, road maintenance work has been paid more and more attention, relying on traditional manual methods of detecting the road damage can not adapt to the needs of the rapid highways development, so the reseach and development in the system of automatic detecting road surface damage has been particularly urgent. This article describes the detection of road cracks based on the image processing.First introduce the architecture of the system in detecting road cracks, describe the methods and reasons in choosing the hardware and the working principle of the system.second,make a de-noising pretreatment to the road crack image after analying the ferture of the image.Nomorly the images we collect have a lot of noise for the complex road conditions and the device,the noise bring a lot of difficulty in detecting the road crack.In order to make the follow-up work easier,do a de-noising pretreatment to the image.According to the feature of the image,use the median filter to do the de-noising word,and compare the result to the one that use mean filter,find out the validity of using median filter and the advantage for follow-up work.Third,edge detection and image segmentation on the image from the second processing make a binary image which make the recognition possible.Comparing the prewitt operator to the sobel operator,verify the validity and advantage of sobel operator,extend the sobel operator to make the edge response more intense, and that can let us use the double-humped method in image segmentation.There will be some"isolated'spot after the image segmentation, so the filter on the bool image is needed. we use the morphological method to do the work, and prove it works. we can find that the bool edge is filled up after using the morphological method. The fill-extent is related to the matrix'size. The crack whose size is bigger tan the matrix is not filled up, and the number of crack pixels is not changed.This feature can be used in measureing the size of crack and distinguish the cracks from the other road damage.finally,after the processing above,we should recognize the crack through the feature of the bool image.Use the crack pixels'number projecting on X,Y axis in bool image and the number of the crack pixels for the imput.after having choosen the imput,the article design a 3 layers BP Neural Network,using the 3 imput and design the 4 output indicate 4 different crack method.Then compare and analyse the network with 30,60 and 150 neuron,find the one with 60 neuron works better.