Unmanned Road Sign Recognition Algorithm

With the development of social economy,the continuous population growth and the increase in the number of cars,road traffic congestion,traffic safety,and air pollution have become issues that cannot be ignored at present.The emergence of driverlessness will be one of the research directions to solve this problem.Driverless cars are more sensitive,quicker and more stable than humans,and can largely avoid human-induced traffic accidents.Unmanned road sign recognition refers to the use of the on-board camera to obtain road scene images,and to identify the road signs and semantics on the images.It belongs to the important content of the unmanned driving current road indication.How to allow the car to automatically and accurately identify the road signs Important research significance.At present,one of the main research directions is to match and identify through the characteristics of road signs and color shapes,and the other is based on machine learning recognition.In addition,there is a deep learning that has developed rapidly this year and become a hot research topic.This paper does the following work on the basis of in-depth study of several classical methods:1.The principle of shape-based template matching algorithm is studied.The shape-based template matching under projection transformation is proposed.The original matching algorithm is added to the projection transformation.The experimental results show that the proposed algorithm improves the accuracy by 5.9%.2.The principle of image recognition based on image recognition is studied.The algorithm of convolutional neural network is applied to the recognition of landmarks.The structure of convolutional neural network is studied in depth,and the structure of modular convolutional neural network is designed.Based on TensorFlow’s open source framework,the algorithm design and experiment were completed.On the GTSRB data set,the data set was visualized and normalized.3.Through preliminary experiments to prove that the initial depth of the convolutional layer is 32,and the convolution kernel size is the time,the performance of the convolutional neural network model is better.On this basis,the input data is expanded by the data set.Degreed processing and histogram equalization have been used to achieve contrast enhancement,and then the Dropout layer has been added to prevent over-fitting.Under different parameter settings,the model was established to carry out the experiment,and finally the best model was established.The number of iterations was increased,and the recognition rate was increased to over 97%.3.Modularize the network architecture to facilitate the subsequent adjustment process.The preliminary experiment proves that the initial depth of the convolutional layer is 32,and the convolutional kernel size is better.The performance of the convolutional neural network model is better.On this basis,the input data is expanded by the data set,and the gray scale is realized.Processing and histogram equalization have been implemented to enhance contrast,and then the Dropout layer has been added to prevent over-fitting.Under different parameter settings,the models were established and experiments were carried out,and the best model was finally established.