Traffic Sign Recognition Method Based On Deep Learning

With' the continuous advancement of artificial intelligence technology,driverless technology has made great progress.Recognition of traffic signs is an important part of driverless system,and it is of great significance for reducing traffic accidents and reducing casualties.The traditional traffic sign recognition system is mainly aimed at the traffic sign images under good environmental conditions,but in actual scenes,due to the unfavorable factors such as the jitter of the image acquisition device and the interference of natural factors,the collected image will have certain deformation and bluring,etc.This paper recognizes multiple types of traffic signs based on deep learning traffic sign recognition methods.The main research contents are as follows:(1)The processing of traffic sign data directly affects the ability of the recognition system.The image enhancement technology is applied to the traffic sign data set,and the image is de-noised through the equalization and normalization operations in the pre-processing.In order to improve the recognition ability of the network model,the image is flipped and symmetrical,and the contrast of the image is adjusted.Through these transformations,the data set is expanded to enable the network model to extract more abundant feature information.(2)Based on the LeNet-5 and AlexNet models,build a static recognition model of the convolutional neural network network on the TensorFlow computing framework.By comparing different activation functions,the ReLU activation function is selected for optimization,and then the data is flattened to reduce dimensionality,which simplifies the data.The fully connected layer and the downsampling layer added in the model improvement will also optimize the results to prevent overfitting during training.Through comparative testing,the accuracy rate of the static recognition model reached 97.7%.(3)Use FFmpeg to synthesize the video frames in the LISA data to obtain simulated test videos,integrate SSD-VGG,SSD-MobileNet_v1,SSD-MobileNet_v2 to build a dynamic recognition model,split the convolution operation to further reduce the parameters generated in dynamic recognition.Use Opencv to load the video and encapsulate it,input the encapsulated data into the traffic sign dynamic recognition system for testing,complete the traffic sign dynamic recognition,the recognition rate can reach 92%,and conduct experimental tests in actual scenarios.