Traffic Sign Detection And Recognition Based On Convolutional Neural Netword

With the development of urban road traffic,Intelligent Traffic System(ITS)comes into being.As an important component of the intelligent transportation system,traffic marking plays a crucial role in traffic safety,so how to quickly and accurately locate and classified traffic signs are widely studied.Traffic signs in natural scenes have prominent color and shape features,which provide favorable conditions for the detection and recognition of traffic signs.However,due to the change of illumination,similar background interference and traffic signs occupy a smaller proportion of scene images,and feature extraction is insufficient.To a certain extent,it affects the detection and recognition accuracy of traffic signs.The convolutional neural network(CNN)achieves remarkable results in target segmentation and object recognition because of its good parameter learning and feature extraction capabilities adaptively.Therefore,this paper proposes a traffic sign detection system based on color information to extract candidate regions and an improved convolutional neural network to identify candidate regions.The main achievement of this article are summarized in the following three aspects.Firstly,when using color segmentation to get candidate regions,in order to remove the similar background interference,A difference R/B algorithm is proposed to segment the traffic sign candidate area.Finally,by applying this algorithm to the German Traffic Sign Detection Benchmark(GTSDB),compared with other color segmentation algorithms,experiments show the segmentation algorithm can reduce the number of false detections and has better real-time performance.In order to improve the real-time performance of the algorithm,a Retinex enhancement algorithm based on sky and nonsky region segmentation is proposed to enhance the image.Finally,the experiment shows that the real-time performance of the algorithm is effectively improved.Secondly,when using convolutional neural networks to discriminate candidate regions,due to the generally small size of the candidate regions,insufficient feature extraction problem,multi-layer feature fusion convolutional neural network(MLFFCNN)is proposed.First,we propose MLFF-CNN structure for feature extraction of traffic signs mapping output.Secondly,due to the loss of edge information,multi-scale sliding window pooling(MSP)is used to instead the single-layer spatial pyramid pool.Finally,the MLFF-CNN model was applied to the recognition of traffic signs.The experiment shows that the detection accuracy of small-scale traffic signs has been improved.Thirdly,using a hierarchical classification algorithm,the German Traffic Sign Recognition Benchmark(GTSRB)is roughly classified into six categories,and then classify each major category into specific categories.Finally,the experiment verifies that the classification algorithm has a higher recognition rate.And to show the results of each processing stage,a simple system detection and recognition interface was designed.