Person Re-identification Based On Content Consistency And Pedestrian Attribute

Person re-identification is in a video surveillance system,when a target pedestrian is captured by the surveillance system,and using modern technology combined with the existing video surveillance system to recognize the same camera at different times,the same location of different cameras or different locations Pedestrian-specific process.This task can quickly and accurately find the criminal suspects,can effectively improve urban security,and escort public safety,so it has received great attention in recent years.However,due to the realistic and complex background,this task will encounter difficulties in that some areas of the pedestrian are blocked by debris,the pedestrian is exposed to light,the posture of the pedestrian changes with time,and the pedestrian image resolution is relatively low.By dividing the pedestrian image to obtain local features,the above problems can be solved better.However,the problem that the pedestrian area does not belong to its own divorce value after division is not effectively solved.At the same time,pedestrian attributes can provide additional auxiliary information is also conducive to solving the above difficulties.Therefore,this article explores pedestrian re-identification by providing auxiliary information from pedestrians dividing local areas and pedestrian attributes.Firstly,in order to solve the problem of the divorce value that does not belong to itself after dividing the local area of the pedestrian,resulting in the problem of reducing the expressibility of pedestrian features extracted by the model,a pedestrian re-identification algorithm for the consistency of the content of the local area of the pedestrian is proposed.The algorithm regenerates new local areas based on the consistency of the local content of pedestrians,increases the differences between classes,and reduces the differences between classes,making the extracted pedestrian features more distinguishable.Then for the differences between the pedestrian attribute feature space and pedestrian identity feature space,and the difference in the amount of information contained between attributes,a maximum pedestrian attribute response to personre-identification network is proposed.The model can effectively distinguish between pedestrian attribute feature space and pedestrian identity feature space,increase the richness of pedestrian feature extraction,and adjust the proportion of its loss function according to the maximum value of the attribute response,so that the attributes with different amounts of information become more Contrast.In this paper,experiments conducted on the two largest datasets for pedestrian re-identification,Market-1501 and Duke MTMC-re ID.Experimental results verify that the content consistency algorithm and the maximum value of pedestrian attributes proposed in this paper respond to the person re-identification network,which increases the robustness of extracting pedestrian features and improves the accuracy of model recognition.