Research On The Detection And Recognition Of Traffic Signs Based On Deep Learning

As China enters a well-off society in an all-round way and continuously realizes the great rejuvenation of the Chinese nation,China's economic strength and scientific and technological strength are also in the world's leading ranks.The number of people's car ownership has continued to increase,and travel safety and traffic safety have become increasingly important issues.The problem.On the other hand,with the continuous innovation and development of science and technology,the era of artificial intelligence is approaching,and intelligent transportation has become a focus of research scholars.One of the important conditions for achieving intelligent transportation is the detection and recognition of traffic signs.Among them,the research on traffic sign detection and recognition based on deep learning has the advantages of high accuracy and fast detection speed.However,under the real and complicated traffic environment in China,the traffic sign target detection algorithm still has problems such as high miss detection rate and low detection accuracy.The research content of this article focuses on how to solve these problems.The main research contents are as follows:In order to study the traffic sign target detection in the real traffic environment in China,comparatively analyze the traffic sign data sets at home and abroad,and make the CTSDE3 data set based on the CTSDB data set as the research object.At the same time,the current target detection algorithm based on deep learning is compared and analyzed.Considering the detection accuracy and speed of the algorithm,the SSD algorithm is selected as the basic network model for traffic sign detection and recognition research.Train,test,and verify the network model on the standardized data set,and compare and analyze the target detection results with other algorithms.The analysis results show that the traffic sign target detection based on the SSD algorithm has a poor detection effect on small targets,and there is a phenomenon of missed detection and false detection,resulting in a low average detection accuracy mAP value.In order to solve the shortcomings of the algorithm network model for the detection of traffic sign targets,this paper replaces the SSD algorithm basic network VGG16 with the ResNet50 network,and deepens the number of network layers to enhance the network model feature extraction.At the same time,a deconvolution operation is added to the network model to strengthen the contextual relationship between the feature extraction layers,strengthen the target semantic information,and improve the feature fusion of the network model,thereby greatly improving the detection accuracy of the algorithm.The improved algorithm network model considers improving the relationship between the feature extraction layers of each layer of the feature pyramid,increasing the number of feature maps extracted by the network model,so that the network model can detect more small targets.In addition,in order to make the algorithm network model better extract the feature information of small targets and increase the number of small targets in the training data set,this paper proposes two data enhancement methods-different picture pyramid images and the same picture combination image.The network model parameters are initialized by loading a pre-trained model,and the number of default boxes is reduced by adjusting the parameters in the aspect ratio set to improve the training and convergence speed of the network model.Finally,through experimental comparison and analysis,the average accuracy of the improved RD-SSD algorithm in traffic sign target detection and recognition is improved by 3.9%compared with the original SSD algorithm,which proves the feasibility of the improved method proposed in this paper.