Research On Traffic Sign Recognition Algorithm

With development of social economy,there is great popularity of automobiles which consequently leads to increasing serious problems in traffic safety and congestion.So the study of driver assistance systems(DAS)has gained more attention in order to effectively improve the safety performance of automobiles.As an important part of DAS,traffic sign recognition is of noticeable research value.Affected by many challenges including the variability of natural weather,the complexity of road surroundings,and human factors,there are still of difficulties and challenges to achieve fast and accurate TSR.In this paper,large category set traffic signs are targeted for in-depth research.In order to solve above problems,two feature extraction methods are proposed innovatively and classification decision based on sparse representation is constructed.The main research work and contributions are organized as follows:Firstly,the method of image processing based on fusion of filtering and pruning strategy is proposed to decline the influence of noise and surroundings.Due to image preprocessing,noise pollution and the proportion of redundant background information can be decreased,which can effectively improve the image quality and classification rate.Secondly,the method of Gabor_MB-LBP feature extraction is proposed based on the in-depth integration of Gabor filter and MB-LBP feature,extracting local texture feature from multi-scale and multi-orientation filtered image,which can realize combination analysis of frequency and spatial domain.Through the research of K-Medoids clustering and SIFT feature,the method of K-Medoids_SIFT feature extraction is proposed,which can better describe the local feature of traffic sign image and enhance the description ability of feature.A high-dimensional joint feature is formed through linear combination of Gabor_MB-LBP feature,K-Medoids_SIFT feature,and Histogram of Oriented Gradient(HOG)feature.Thirdly,the cascade classification method of coarse level to fine level is proposed through the thorough analysis of characteristics of traffic signs.Accurately coarse level of classification is implemented fast by the combination of distance to border and SVM.Depending on the results of coarse level,a high-dimensional joint feature is shown in the next step.Based on the research of sparse representation,it can achieve the more compact description of joint feature construction.By integrating Bayesian decision into SVM,the SVM_NB classifier is constructed to develop more precise classification ability near the optimal classification plane.The joint feature is judged by the combination of sparse representation and SVM_NB classifier to realize fine level.