Research On Application Of Traffic Sign Recognition Based On Improved Convolutional Neural Network

With the rapid development of China's economy,the number of domestic cars is also increasing rapidly.It is not difficult for every family to own a car.However,at the same time,traffic accidents also increase with the increase of cars,which is a great danger to the country,society,and every family.To avoid traffic accidents as much as possible,traffic signs are usually installed on highways.Under the guidance of traffic signs,drivers can drive safely and avoid most traffic accidents.However,due to the weather,visibility,or the driver's own inattention and fatigue,some situations make the driver not see the traffic signs,and thus lead to a series of traffic accidents.The emergence of Traffic Sign Recognition Systems(TSRS)has made it possible to completely solve this problem.In a real scene,the scene is very complex and varied due to light intensity,shielding,weather and other reasons,and it will bring some difficulties and problems to the recognition system of traffic signs.In recent years,with the development and application of Deep Neural Networks(DNN)and the appearance of Convolutional Neural Networks(CNN),the problem of image recognition and classification has been greatly developed,making the solution to the problem that detection of traffic signs possible.In this paper,the image recognition classification algorithm based on convolutional neural network is applied to the problem of traffic sign recognition.The relevant key technologies are studied,and several improvement measures are proposed on the basis of it again.Based on the convolutional neural network,the spatial transformation network(STN)and Color Transformer Layer modules are added in front of the convolutional layer so that the convolutional neural network can be used during training and prediction.It is possible to ignore the geometric transformation and color transformation due to the shooting angle and other reasons,thereby improving the detection accuracy of traffic signs.In the process of traffic sign recognition,the sample RGB color space is first adjusted through the color conversion layer,and then the geometric transformation is adjusted through the space transformation network so that the sample can ignore the adverse effects of these two factors as much as possible and then pass through the convolutional neural network to perform feature extraction and classification,and finally get sample categories.The main work of this paper is to improve the convolutional neural network by adding space transformation network and color transformation layer for traffic sign recognition problem,and to improve the classic convolutional network model for traffic sign recognition.The improved recognition system was finally implemented and applied in the German Traffic Sign Recognition Benchmark(GTSRB)data set.The experimental data obtained by detecting the recognition rate and efficiency of the algorithm through many real shooting samples proves that the recognition accuracy of the recognition algorithm in this paper reaches 99.45%,and the time for a single sample identification is 1.5ms.Compared with the traditional traffic sign system,the recognition rate is higher,and it is more resistant to geometric and other noise interferences under complex backgrounds,which satisfies the high accuracy requirements of traffic sign recognition.