Research On Vehicle Re-identification

With the development of modern industrialization,the number of residents of motor vehicles continues to increase,bringing great convenience to people's lives,but it also brings serious challenges to traffic management.As a result,the intelligentization of Intelligent Transportation Systems(ITS)has become a research hotspot.In ITS,the study of the reidentification method of a given vehicle is one of the keys.This paper proposes a vehicle re-identification method based on deep learning network for detection and matching of vehicle window identification area because the license plate cannot be used as a basis for vehicle re-identification due to factors such as occlusion,insult,and shooting angle.The difficulty in recognizing a vehicle is the resolution of different vehicles of the same style and the same color.Because the appearance of the vehicle is exactly the same at this time,a taxi in a city is a typical example.Through observation,it was found that the difference between different vehicles of the same color and the same color is only the placement of the annual inspection mark on the window,and the difference in the items placed in the window.Therefore,this paper constructs a two-channel convolutional neural network,a channel to extract the vehicle's style,body color,positive and negative orientation of these three appearance features;another channel is to extract the vehicle's windows,annual inspection,the location of the body paste and Local fine-grained features of the type.When training the deep learning network model,the Batch Hard Triplet Loss loss function was selected to enhance the feature differences between different vehicles and enhance the fine-grained features of the network model.This article will be the network model output color,style,orientation of the global appearance of features,rough classification of the vehicle,can greatly reduce the number of maps to be matched;then through the annual inspection,the window,the fine-grained local characteristics of the relative position of the body paste It is obviously not the image of the same car.Finally,through deep learning,the features of the annual inspection and local subgraphs of the car window are extracted,the network's attention is focused on the area with the ability to identify,and the re-identification of the vehicle is completed based on the similarity of the local fine-grained features.The experimental results on the VehiclelD vehicle standard test data set verify the effectiveness of the algorithm and the MAP up to 70%..