Real-time Traffic Sign Recognition System Based On Support Vector Machine

With the development of social economy,road traffic jam,congestion,safety and other issues have become a common concern of the society.The real-time identification system of traffic signs can strengthen the relationship between vehicles,roads and drivers,so as to form an effective strategy to ensure traffic safety and solve road congestion.As an important part of intelligent transportation system,traffic sign identification system is of great significance to solve urban traffic congestion,guarantee road traffic safety,and save energy consumption.Meanwhile,it plays a great role in the promotion of unmanned driving technology.Therefore,this paper studies the traffic sign identification system,which is of great significance to the actual traffic operation by timely warning,instruction and reminding drivers.In the natural scene,the traffic sign recognition technology is not very mature when the accuracy and real-time requirements are high.In this paper,the real-time and accuracy of traffic sign detection and recognition system to study,first we image preprocessing for road traffic signs,histogram equalization is used to improve image quality,and through smooth sharpening processing of image enhancement,we finally on the mathematical morphological image processing to enhance the image quality to reduce amount of calculation of the system.In order to ensure that the time of image preprocessing does not affect the real-time performance of the system,it is ready for the subsequent detection and positioning of traffic signs.Because the traffic sign image has specific color and shape information,we can detect and locate the traffic sign according to the information.We are comparing the real-time performance of RGB difference color segmentation and HSV color segmentation algorithm,and select the RGB difference segmentation algorithm with higher real-time performance and better segmentation effect to preliminarily determine the range of traffic signs.In order to deal with because of shade and speed lead to traffic sign image shape is not complete,Gramham scanning method is used for traffic signs contour convex hull processing,re-use traffic sign image shape information is put forward based on the shape of the least squares fitting method of fixing the traffic sign image segmentation to extract further to the target area of interest.The experiment proves that the real-time and accuracy of traffic sign detection and segmentation in natural scene are very good.In the process of traffic sign recognition,the first step is to extract the feature of traffic sign image.After analyzing and comparing the performance of image HOG feature and LBP feature,we select the HOG feature of the image.HOG feature in an image at the same time,we use PCA technology dimension,compared with the direct use of HOG feature classification,after the dimension reduction for system recognition accuracy and recognition speed have good increase,in the classification of traffic signs recognition phase,we use is traffic signs in Germany standard database(GTSDB),USES the commonly used SVM algorithm of support vector machine(SVM)is used to identify the training traffic sign recognition model is established.The experiment proves that the proposed algorithm has low time complexity and achieves good results in real time traffic sign recognition.