Study On Traffic Sign Recognition Method Based On Neural Network

With the progress of information technology,road safety has been greatly improved,and now more attention is paid to the safety problems caused by subjective errors in the driving process of drivers.Because traffic signs contain important road traffic information,they play an important role in traffic efficiency and safety.Therefore,automatic detection and classification of traffic signs is an important task for intelligent and self driving vehicles.For the complex traffic background,the traditional machine learning based traffic sign recognition(TSR)often has the problems of low real-time performance and high requirements for hardware computing performance.To solve these problems,an algorithm based on ROI extraction and convolutional neural network(CNN)is proposed.In this paper,traffic signs as the research object,from the traffic sign image preprocessing,feature extraction,detection and recognition technology and other aspects of in-depth research.This paper mainly does the following work:(1)The traffic sign recognition process generally consists of two phases:before testing and identification process,the GTSDB traffic sign detection(Germany standard)conducted a series of preprocessing data sets,including regional tailoring,gray processing and size normalization,and the traffic signs of processing image for further adaptive enhancement,especially emphasizes the effective characteristics of the traffic sign image information.(2)Study on traffic sign detection method based on convolution neural network.In this paper,the color of red edge traffic signs in GTSDB data set is enhanced,and then the feature of the enhanced image is extracted by combining with mser,and then the ROI is classified by CNN to complete the filter error detection of traffic signs,detect the location of traffic signs,and then the traffic signs are segmented from the background image and framed.(3)Study on traffic sign classification method based on convolution neural network.In this paper,the deep learning framework keras is used to build convolutional neural network,GTSRB data set is used for training and testing,and a deeper neural network model is built on keras to train and test GTSRB data set,and Lenet-5 convolutional neural network model is trained to classify regions of interest,and 30 types of traffic signs are trained by the model Classification.Then the detected traffic signs are input into the traffic sign classification model as the input image to get the accurate information of the traffic signs,to predict the detected traffic signs and complete the traffic sign classification,so as to achieve the purpose of traffic sign recognition.The experimental results show that the algorithm achieves 97%and 98%accuracy when it is used for detection and classification,and the speed of operation is greatly improved,and the real-time and robustness of traffic sign recognition are enhanced.respectively in the two convolutional neural network classifications,and it greatly improves the speed of operation and enhances the real-time and robustness of traffic sign recognition.