Research On Traffic Sign Recognition Based On Deep Learning

With the increasing number of cars in China,people pay more and more attention to road traffic safety.As an important part of the car's advanced driving assistance system,traffic sign recognition can enable cars to obtain road information actively,so as to remind or stop dangerous driving behavior.Traffic sign recognition includes two parts: traffic sign detection and traffic sign classification.The former is used to obtain the location information of traffic signs,and the latter is used to determine the category of traffic signs.At present,some related scholars have made progress in the research of traffic sign recognition methods.However,due to the small target,many types of traffic signs,and the complex environment,the accuracy of the recognition still has a lot of room for improvement.In view of the above problems,the study summarizes the existing research results at home and abroad,and uses deep learning to design the traffic sign recognition algorithm.The main work is as follows:(1)In the part of traffic sign detection,the detection algorithm is designed based on the idea of candidate region.Firstly,Res Ne Xt with multi-scale convolution kernel is constructed as the backbone network of the detection algorithm,and a multi-layer feature fusion method is used to improve the detection accuracy of small target traffic signs.Secondly,in order to further avoid missed and false detection,the anchors of Region Proposal Network are designed by clustering traffic sign size and scale features.Finally,the Non-maximum Suppression algorithm is improved by filtering candidate regions,so as to improve the processing efficiency of the overall detection algorithm.(2)In the part of traffic sign classification,a traffic sign classification algorithm with two-way convolutional network structure is constructed.The two-way network structure of Mobile Net+Res Net50 is used to enhance the information expression capability of features,thereby improving the accuracy of the classification algorithm.Then,the training data of the classification algorithm is constructed and the data is preprocessed.The traffic signs with a small sample size are expanded to improve the classification accuracy.(3)The overall algorithm performance analysis part combines the traffic sign detection algorithm with the traffic sign classification algorithm,and conducts anexperimental analysis of the entire traffic sign recognition algorithm on different data sets,and finally tested by actual road condition data.The experimental results show that the overall traffic sign recognition algorithm has strong robustness and universality,which solves the problem of difficult recognition of small target traffic signs,and can effectively identify traffic signs in different lighting,complex background and other environments.