Research On Traffic Sign Recognition Based On Vision

With the rapid growth of the number of motor vehicles,traffic safety has been widely concerned at present,and traffic management system is becoming more and more complex,so the traditional traffic management system can't be suitable for social development.Under this background,the concept of Intelligent Transportation System(ITS)is presented.ITS is a kind of real-time,accurate and high-efficient integrated management system which integrates the advanced information,communication,sensing,control and computer technologies into the whole transportation management system.Traffic sign recognition(TSR)system is an important part of intelligent transportation system and plays an important role in auxiliary driving and automatic driving.Although the scholars have been studying traffic sign recognition for many years and some research results have been achieved,the research of traffic sign recognition has not yet reached maturity because of the complicated and changing traffic environment.In view of current situation,this thesis does researches on image restoration of complex weather,traffic sign detection and traffic sign classification.Firstly,an improved image restoration algorithm aiming at image restoration in rainy weather is proposed based on analyzing the characteristics of rain imaging and the advantage of L0 gradient minimization algorithm.The algorithm solves the problem of rain blocking traffic sign.Since rain is the most common weather condition,the traffic sign recognition system should be able to adapt to this situation.The experimental results show that the algorithm proposed in this thesis can improve the image clarity and provide better input image for the next traffic sign detection.Secondly,an improved algorithm is presented based on the color detection algorithm on the basis of the research on traffic sign detection.The binary image is generated by color segmentation.Light occlusion and mark damage were processed by ellipse fitting to form a closed image.The holes of the closed pattern are extracted,and a region of interest is generated using a number of limiting conditions.The experimental results show that the proposed algorithm is more accurate and can deal with the problem of mild occlusion and connected signs.Lastly,a classification algorithm of first coarse classification and then sub-classification is designed aiming at the traffic sign classification.The algorithm adopts the design pattern of HOG feature and SVM classifier.Shapes are adjusted based on logo features after the traffic signs are roughly classified into some sub-categories,and then sub-classification give the classification results.Experimental results show that the proposed algorithm has high accuracy and can adapt to various disadvantages.