Research On Traffic Sign Recognition Technology Of Intelligent Driving System

With the rapid development of deep learning technology,self-driving technology based on deep learning technology has now become one of the most popular research among researchers.Traffic sign image recognition is the main problem to be solved in today's autonomous driving.With the commercialization of neural networks,convolutional neural networks are becoming the preferred method to solve such problems.Common convolutional neural networks such as Rich feature hierarchies for accurate object detection and semantic segmentation and Fast Rich feature hierarchies for accurate object detection and semantic segmentation do not have good fine-grained image processing capabilities and lack specific analysis of image feature points.This paper proposes that for this type of problem,we can start from the fine-grained perspective of the image and use the Recurrent Attention Convolutional Neural Network algorithm that is sensitive to fine-grained to enhance its ability to recognize complex images.In this paper,the Recurrent Attention Convolutional Neural Network algorithm based on attention mechanism and the optimized VGG19 network structure are used.On the basis of guaranteeing the original performance of VGG19,it is lightweight optimized and optimized to a 9-layer network structure to enhance The speed of model convergence,while using the Re LU function as the activation function of the network model.The training set and test set in this paper are selected from the common traffic sign images in the CCTSDB database.Before training and verification,the image preprocessing operations such as noise reduction and binarization are used to enhance the ability of image expression information.After preprocessing,the image is imported into the Tensor Flow platform for training,and finally the test set is used to test the actual expressiveness of the module.The experimental results show that the model finally trained in this paper is better than traditional support vector machine and other solutions.The accuracy of simple traffic sign recognition and classification,such as straight driving signs,can reach 99.14%,and the average recognition rate of other traffic signs can be Reaching 96.5%,Recurrent Attention Convolutional Neural Network based on attention mechanism is significantly better than traditional schemes in recognition of complex signs,and basically completes the goal of traffic sign recognition and classification.Then,this article also compares the currently developed more mature neural networks such as Hierarchical-CNN.The comparison results show that the current Recurrent Attention Convolutional Neural Network based traffic recognition system is still in the exploratory stage.The model in this paper is still a certain gap from the existing mature models.There is still room for optimization on the model structure.Finally,combined with the experimental results in this paper,using the Recurrent Attention Convolutional Neural Network network model based on the attention mechanism is completely feasible for identifying such problems.It can provide some reference in the field of image classification and recognition.