The Key Technology Research On Vehicle Re-identification

In recent years,the number and frequency of vehicles have been greatly increased,while monitoring equipments have been widely used.As one of the important technical means to obtain vehicle information from monitoring video,vehicle re-identification has gradually become a research hotspot.Vehicle re-identification aims to find samples belonging to the same target vehicle from multiple image sources.This thesis focuses on three key issues of vehicle re-identification.In vehicle re-identification based on single frame image,this thesis proposes a loss function combining logistic triplet loss and label-smoothing cross entropy loss to improve the accuracy of vehicle re-identification.On the basis of the traditional triplet loss,this thesis proposes logistic triplet loss to mine the internal information of hard samples and the inter-class differences.It utilizes label-smoothing method to enhance the robustness of traditional cross entropy.At the same time,it combines the advantages of fine-grained classification and metric learning.In vehicle re-identification based on sequence data,a feature fusion algorithm based on temporal attention is proposed,which improves the representation ability of vehicle sequence feature.In vehicle re-identification based on unsupervised learning,this thesis proposes a soft multi-label triplet loss to use the semantic characteristics of different levels of the model.At the same time,it uses the feature reference regularization method,which improves the generalization ability of the model.Experimental results on several datasets verify the effectiveness of the algorithm.