Research On Road Traffic Sign Recognition Method And Experimental System Development

Traffic signs provide drivers with the information of road conditions ahead in time through the relevant shape and text information as the carrier,which plays an important role in guiding vehicles to drive safely.The research of traffic sign detection and identification method is of great significance to improve the road traffic safety,ensure the safe driving of vehicles and the development of intelligent networked vehicles.This paper is based on the research of existing traffic sign detection and recognition methods.For static traffic sign recognition and dynamic traffic sign recognition,a traffic sign recognition method based on the improved Le Net-5convolutional neural network and a traffic sign recognition method based on the optimized Faster R-CNN convolutional neural network are respectively proposed.And based on this,a simulation experiment system for traffic sign recognition is designed.The main work includes the following aspects:Firstly,the unique color and shape feature information of traffic signs are analyzed.Combined with the existing research basis,this paper expounds the relevant methods of traffic sign detection and recognition,and reproduces the relevant traffic sign detection methods based on color and shape.By analyzing the advantages and disadvantages of each method,the advantages of this research method are determined,which provides a certain research basis for the later research.Secondly,a traffic sign recognition method based on the improved Le Net-5 convolutional neural network structure is proposed.Aiming at the problem of low recognition accuracy in traditional network model recognition,a new pooling layer and a fully connected layer are added to the back end of the model to deepen the feature extraction and data information classification of traffic sign images.In order to solve the phenomenon of gradient disappearance and overfitting during the training of the model,the Re LU activation function was used to replace the Sigmoid activation function in the original network structure,and the Dropout strategy was added to the structure.After the final experimental comparison and analysis,the recognition accuracy rate obtained by the improved model in the verification set reaches 99.45%,and the model’s recognition time for a single image is 35.34 ms,which meets the real-time recognition requirements.Tertiary,after analyzing the Faster R-CNN network structure,the network model structure is optimized for traffic sign recognition.In the training of the model,the data set is augmented and enhanced to enrich the diversity of the data set.In terms of feature extraction network,the Res Net50 network is used to replace the original VGG16 network for feature extraction.In order to reduce the complexity of the model,the L2 regularization method and the loss function are added to the model to reduce the large learning parameters during model training.After experimental comparison,the recognition accuracy of the improved model is increased by 0.79%compared with that before the improvement,and the recognition time of a single image is 31.6ms,and the model has been optimized to a certain extent.Finally,based on the above research results,a real-time traffic sign recognition simulation experimental system is designed and developed.The hardware part of the experimental system is designed with Solid Works software,and then developed and produced to provide hardware support for the experiment.In the software part,Python programming software,Tensor Flow framework and supporting Py Qt toolkit are used to design the system interface,build the visual interface of the experimental process,and realize the functions of model testing and real-time traffic sign recognition.To sum up,this paper studies the road traffic sign recognition method,proposes related improvement methods for static and dynamic traffic sign recognition,and establishes an experimental system for traffic sign recognition after experimental verification on the data set.Realize the whole process of traffic sign recognition.