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

Automatic detection and recognition of traffic signs is an important research topic in the field of automatic driving.As traffic signs are closely related to the safety of passengers and road traffic,the automatic detection and recognition of traffic signs require extremely high accuracy.In addition,improving the accuracy of automatic detection and recognition of traffic signs will also help to improve the overall traffic efficiency of roads so as to build a more efficient and environmentally friendly road traffic system.In the early days of automatic detection and recognition of traffic signs,the methods of manual features were often used to carry out related research.Although certain achievements have been achieved,these methods rely on the designers’ prior knowledge to carry out tedious manual parameter adjustment.In recent years,with the rise of deep learning technology,convolutional neural networks such as VGG-Net and Res Net present better advantages compared with manual feature methods in the field of image recognition,so researchers have begun to study automatic detection and recognition of traffic signs by using deep learning technology.However,factors such as external weather,the distance of sample collection and the damage of traffic signs have a great impact on the automatic detection and recognition of traffic signs,which makes the accuracy of the current traffic sign automatic detection and recognition based on deep learning still lags behind the requirements of practical application.In order to further improve the accuracy of automatic recognition of traffic signs,this paper focuses on the following research based on deep learning technology.(1)For the problem that the traditional detection methods are not effective for small target detection,a traffic sign detection algorithm based on improved SSD model is proposed in this paper.Firstly,the feature pyramid FPN algorithm should be introduced into SSD network to deconvolute the deep network layer by layer and deeply splice it with the previous network layer.The high-level semantic information and shallow detail information can be combined to enhance the ability of feature representation.Then,the Kmeans clustering algorithm should be used to determine the size of the default frame window to obtain a proportion value suitable for the current dataset and improve the detection accuracy of the model.The experimental results on CCTSDB traffic sign dataset show that the m AP of the improved SSD model reaches 93.58%,which achieves better results in traffic sign detection.(2)For the problems of high-level information loss and insufficient feature extraction in sampling in network structure,the Res Net network structure has been improved and a traffic sign recognition method based on multi-scale features and attention mechanism has been put forward in this paper.Firstly,multi-scale features are used to fuse different levels of feature information to enrich feature semantic information and enhance the ability of feature extraction.Then,the features of different channels have been strengthened through the attention mechanism to improve the overall presence of traffic signs to achieve more accurate traffic sign recognition.The experimental results on GTSRB and Belgium TS traffic sign datasets show that the accuracies with the proposed methods reach 99.31% and 98.96%respectively,which achieves better results in traffic sign recognition.