Pavement Distress Recognition Based On Image

Along with the further development on the digital image processing technology, this newtechnology is applied on the surface road damage detection which not only save labour forcesand resist the subjective interruption, but also it is able to provide a fast and accurate evaluationon the road status which is really useful. This article is the research on the methodology of theroad damage automatic identification and classification according to the road binary images. Itconcerns two parts: pick-up methodology on the characters of road damage image and selectionontheclassifiers.During the process of road damage identification, how to characterize the type of thedamage is the key to achieve an accurate and real-time image identification. Considering thestructure characters and statistics characters integrally, this article describes the improvement onthe current character extraction methodology. It chooses the road subblock images which showsdifferential vector from horizontal and vertical orientation, 3×3,5×5 density factors andrelative road image convolution. Support Vector Machine method is based on the theory ofVC-dimensional structure and the principle of risk minimization of statistical learning theory,According to the limited sample of the information to find the best compromise between thecomplexityandlearningabilityinthemodel,Accesstothebestabilitytopromote,itcaneffectivesolute the phenomenon of learning of neural networks.Model solution finally reflected in aquadratic programming issues.In theory,it can find the global optimal solution,that resovled thelocal extreme problems which can not be avoided in the neural network. For this reason, paperselection Support Vector Machine as classifier.through test choose radial basis kernel.The outputof the classfication are longitudinal cracks,converse cracks,allgitor cracks,block cracks and nocracksofthefivetypesofpavementdistresses.Based on two kinds of issues on damage character pick-up and classifier selection from theroad binary image, this article states the weakness on the traditional method and theimprovements which are selecting the actual and generated road images as debug and testsamples and using Matlab to fulfill the simulation test. First of all, it compared the differencebetween the tests under BP Neural Networks classifier and other character pick-upmethodologies. Second, it compared the different between choosing the support vector machineandBPNeural Networks byusingthe characterpick-upmethodologywhichis mentionedinthisarticle. All the tests proved that this method improved the identification accuracy, efficiencyand robustness.