Research Of Traffic Sign Recognition Based On Ensemble Convolutional Neural Network

Traffic sign recognition is an important part of intelligent transportation systems and driverless technology.The traffic sign recognition system needs to identify the traffic signs collected by the on-board camera in the actual driving environment.The quality of the collected traffic sign images is easily influenced by many factors such as light intensity,speed and damage of traffic signs in the actual driving environment,which brings great difficulties to the subsequent identification.Therefore,it is a great challenge and has great significance to research on the identification methods of traffic signs.In this thesis,the traffic sign recognition is deeply researched with the combination of the convolutional neural network,which takes the German traffic sign recognition benchmark(GTSRB)as the research object.In addition,the traffic sign recognition which combines the convolutional neural network and ensemble learning together is also further researched.The main work of this thesis is as follows:1.According to the characteristics of traffic sign images collected in the actual driving environment,in order to reduce the recognition burden of the convolutional neural network and improve the recognition rate,this thesis performs a series of preprocessing operations on the GTSRB data set.The steps mainly include image enhancement,image graying,size normalization and data set expansion.After preprocessing,the GTSRB set is more helpful for the identification of the subsequent convolutional neural network.2.This thesis studies the convolutional neural network and designs a convolutional neural network model CNN-SVM to identify traffic sign images.The proposed convolutional neural network uses ReLU as active function,and the Dropout strategy is utilized to prevent over-fitting happening in the training process.Then,the batch normalization method is applied to speed up the network training.In order to verify the validity of the proposed CNN-SVM model,the detailed comparison experiments were performed on convolution layers,Dropout parameters and classifier categories of the model.The experimental results show that single convolutional neural network model achieves a recognition rate of 97.37% and an average recognition speed of 0.12 ms on the GTSRB data set.3.With the foundation of research on the traffic recognition based on convolutional neural network,an ensemble convolutional neural network model is proposed which combines the convolutional neural network and ensemble learning.The model consists of six base classifiers and uses bagging ensemble algorithm.The difference among these classifiers are ensured from three aspects included training data set,convolutional neural network model structure and network parameter initialization.The experimental results show that the proposed ensemble convolutional neural network model achieves a recognition rate of 99.33% and an average recognition speed of 0.76 ms on the GTSRB data set,which has good recognition performance.