Research On Traffic Sign Detection And Recognition Method Based On Deep Learning

With the rapid development of artificial intelligence and information technology,image detection and recognition technology is gradually applied to various fields of society,which plays an important role in improving people’s living and travel efficiency.With the popularity of 5g network and the improvement of chip computing power,and the continuous investment in new infrastructure construction,it will provide a suitable hardware environment for the application of image detection and recognition in the field of automatic driving,which will become a hot research direction of intelligent transportation in the future.Traffic sign recognition technology has important value in automatic driving and auxiliary driving.In this paper,we use deep learning technology to study the image of traffic signs,and focus on the detection and recognition methods of traffic signs.The common target detection and image recognition algorithms are sorted out.In order to solve the problem that the detection accuracy of traffic sign recognition is not high due to the complex natural environment such as light,weather,image shooting angle and sign fading,an improved traffic sign recognition method based on regression is proposed.The SSD algorithm is improved to improve the accuracy of traffic sign location.The depth residual network is used as the front-end network structure of the prediction model,and different scale fusion feature layers are introduced.The traffic sign data set is transmitted to the trained and optimized network to locate traffic signs.In the identification of traffic signs,this paper studies an improved convolution neural network model.A convolution feature extraction module(FE-Module)is added to the traditional CNN network.The cascaded convolution layer is used to extract the features of the image input into the network model.The problem of insufficient image feature extraction of single convolution check in the traditional convolution neural network is solved.With the deepening of the network structure,the main feature information of the image can be extracted by the convolution kernel at the end of the cascade This network model can reduce the parameters in the network,so the whole network model will become lightweight and reduce the training time.Comparing the recognition effect on the gtsrb traffic sign data set is more efficient than the traditional CNN network,because the parameters of the model are less,the consumption of space resources is reduced,and the portability of the model is also improved.