Detection And Recognition Research Of Road Traffic Signs In Natural Enviroments

With the development of our country's urbanization and the proportion of private cars straight up, traffic problem has become a social problem. To solve the traffic problem used by Intelligent Transportation System(ITS) has become the future trend of development. Traffic sign recognition system(TSR)is an important branch in the field of ITS. The traffic accidents caused by fatigue driving, bad weather and other reasons will be avoided by real time date got by traffic sign recognition system. Therefore, traffic sign recognition is of great theoretical significance and application prospects.This thesis mainly focuses on rapid detection algorithm and identification algorithm of the road traffic signs in natural enviroments, the details are as follows:(1)Aiming at the problem that there exists motion blurred image, the method to restore motion blurred image is proposed. We mainly deal with the motion blurred image caused by the approximate uniform linear motion. First of all, we get the point spread function(PSF) of uniform linear motion, and then according the degradation model to restore the original image. Wiener filtering and other several kinds of method to restore motion blurred image are analyzed and tested, the method how to reduce the noise effects are also discussed.(2)Concerning on the problem of brightness and contrast of the photos is too low caused by lack of sun?difference of recording and so on, this paper proposed an algorithm that the histogram equalization and Gabor filter is applied for image enhancement and a method for light compensation based on reference white.(3)Focusing on the problem of time consumption and accuracy of traffic sign detection system,an improved algorithm for traffic sign detection based on improved color enhancement combined with support vector machine()SVM is proposed. First the improved color enhancement method is used to select the coarse candidate areas, after which a SVM classifier is trained using HOG feature for further detection and shape judgment.(4)In order to solve the problem of traffic sign identification, we carried on the comparison to the identification speeed and accuracy of the common classification methods in target recognition. This thesis uses the Probabilistic Neural Network(PNN) which need less time to train and has other competitive advantage for classification. HOG features are chosen as training feature too. The dimension reduction of the training sample is done by the Principal Component Analysis(PCA) in order to avoid curse of dimensionality and increase recognition speed.In this thesis, the main work is some exploration and attempt based on the existing sign derection algorithm.The experimental results show that the algorithm can fast detect signs and determine the shape,it can also deal with various cases such as images with low brightness,rotation and partial occlusion with high accuracy,basically meet the demand of practical application.