Research On The Detection And Recognition Algorithm Of Traffic Signs Based On Computer Vision

According to statistics,the fatality rate about road traffic accidents in China had reached 18.2%in 2017.Therefore,the traffic safety has become the most common problem in our daily lives.One of the effective ways to solve traffic accidents is to identify the road traffic signs accurately and effectively.At present,the improvement of the artificial intelligence technology in the Intelligent Transportation System(ITS)field brings hope to the intelligent identification of road traffic signs.Therefore,the Traffic Signs Recognition(TSR)has attracted the attention of scholars from various countries,especially in Advanced Driver Assistance Systems(ADAS).The demanding for the ability to detect and identify traffic signs automatically has become increasingly.The focus of ADAS research is vision,and the premise of vision is perception,the perception of road scenes.However,there are huge challenges in detecting these signs in real road scenes.For example,the intensity of the light,the degree of blur of the signage,and the coverage and occlusion of other obstacles,these weather and scene factors will affect the detection and recognition effects and accuracy.This article mainly studies the prohibition,warning,direction traffic signs.To solve this problem,this paper proposes an overall framework for detection and identification.The main contributions are as follows:1.In the detection stage,based on the characteristics of HSV color space and RGB color space,this paper proposes a method based on RGB color threshold segmentation.Then,in order to solve the problem of coverage and ambiguity,based color segmentation,this paper proposes a shape fitting algorithm based on the least squares method.It used to determine the ROIs.The experiment shows that the detection accuracy is improved greatly.2.In the stage of classification and recognition,this paper proposes the idea of using svm and cnn as two-level classification.Based on the multi-classification of svm and the color and shape of traffic sings,the traffic signs are classified into six subcategories for rough classification by svm.Firstly,the features of HOG and LBP algorithms are compared and analyzed.The method of appropriate feature extraction is selected.Then six different types of traffic signs were trained as their respective classifiers.The classifier used as a first-level classifier for candidate regions with an accuracy of 95%,3.In the secondary classification idea,this paper proposes and designs a shallow convolutional neural network as a second-level classifier.Based on the first-level classifier,this network is used to identify the specific content of traffic signs.That is to say,the patches after SVM classification are input into the network after being grayed and identified the results by fine classification.ExPeriments show that the network has a good performance in real-time and accuracy.