Research On Vehicle Retrieval Method In Expressway Scene Based On Deep Learning Neural Network

In recent years,with the construction and promotion of intelligent transportation and highway intelligent management,automatic vehicle identification and automatic vehicle information perception technology have become an important part of intelligent transportation construction.At present,the expressway vehicle identification method mainly based on on on-board OBU reading IC card information has the disadvantages of low recognition rate,high failure rate,poor recognition effect of fake and fake cards and so on.In view of the above problems,this paper has carried out the research on vehicle retrieval method in Expressway scene based on deep learning.The research results and innovations are as follows:Firstly,the vehicle image retrieval sample set of Southeast University and the vehicle image retrieval test set of Southeast University are constructed.The image set includes four categories:vehicle image,vehicle face image,window image and license plate image;Four vehicle image feature extraction methods based on SIFT,orb,LBP and LDP are analyzed,four image feature matching methods based on visual word bag model,cosine distance similarity,European distance and Hamming distance are studied,and the vehicle image retrieval comparison experiment based on traditional machine learning is carried out.The experimental results show that the vehicle image retrieval method based on traditional machine learning has the problems of low accuracy and high miss rate.Secondly,the vehicle image retrieval models in three highway scenes based on vgg16,resnet50 and densenet121 networks are constructed,and the comparative experimental research is carried out based on the vehicle image retrieval sample set of Southeast University and the vehicle image retrieval test set of Southeast University.The experimental results show that the performance of the vehicle image retrieval model based on densenet121 network is better than the other two models The retrieval accuracy of car face image test set and window image test set is better than 90%;Finally,a vehicle image retrieval method in Expressway scene based on deep learning fusion model is proposed.The method includes vehicle image retrieval fusion model based on single feature multi network fusion and vehicle image retrieval fusion model based on multi network multi feature.Comparative experiments are carried out based on vehicle image retrieval sample set of Southeast University and vehicle image retrieval test set of Southeast University;The experimental results show that the performance of the vehicle image retrieval fusion model used in the expressway scene is better than the single network model.In the retrieval of vehicle image and license plate image in the expressway scene,the retrieval accuracy of the single feature multi network fusion retrieval model reaches 99.42% and 97.28% respectively.In the retrieval of vehicle face image and window image in the expressway scene,The retrieval accuracy of multi feature and multi network fusion retrieval model reaches 98.11% and 98.06% respectively.