Symbol Detection And Recognition On Traffic Panel From Nature Scene

With the improvement of living standards,the number of cars is more increased,which brings the serious traffic problems,such as traffic jam.Developing intelligent transportation system is an effective means to solve the traffic problems.Traffic sign detection and recognition is one of the most important parts of intelligent transportation system,and it is also widely used in automatic driving,which has gotten attention from many experts and scholars.At present,the detection and recognition of traffic signs are mainly focused on the ban and warning signs.There are few studies on rectangular traffic panels containing characters and symbols,but rectangular traffic panel is also an important part of traffic signs,which plays an important role in ensuring road safety and traffic efficiency.Therefore,studies on rectangular traffic panels have great application prospects and practical value.Symbol detection and recognition on traffic panels are studied in this dissertation,and the main work of this paper includes the following three aspects:1.Image preprocessing of rectangular traffic panels.Because light condition in natural scene is not under control and shooting angle is not fixed,Contrast Limited Adaptive histogram equalization is adopted to correct light condition.Line detection algorithm is used for distortion correction by detecting the rectangle traffic panel outer border.The experimental results show that the effects of the change of the light and shooting angle are reduced and image quality is improved after processing.2.Symbols detection on traffic panels.In order to boost the symbol detection speed of traffic panels,two levels detection algorithm from coarse detection to precise locating has been proposed.Firstly,extract regions of interest from traffic panel images,filter the candidate region based on rules.Secondly image gradient feature and texture feature is combined,put the combined feature into linear Support Vector Machine for precious symbol detection.In order to reduce the calculated amount further,Principal Component Analysis(PCA)algorithm is adopted to reduce the feature dimension.3.Symbols recognition on traffic panels.Because symbols on traffic panels have obvious gradient and direction characteristics,an improved Gabor feature is proposed.Meanwhile,image gratitude features are extracted on the improved Gabor feature maps.Linear support vector machine and random forest are used for symbol recognition,and the experimental results are compared.To test the performance of the algorithm,a dataset containing 1350 traffic panels which were collected in China is built.In this dataset,symbols on traffic panels are divided into of 25 categories,and 450 pictures are randomly selected as test images in this paper.Symbols detection accuracy is 90.9%,and the recognition accuracy is 98.1%.The experimental results show the effectiveness of our proposed algorithm.