Research Of Person Re-identification Based On Global And Local Information Joint Learning

Since the beginning of the 21st century,with the development of society,the mobility of society has been increasing,and the subsequent security issues have been paid more and more attention by the government and people."Smart Cities" and "Safe Campuses" and other security projects are under construction,and a lot of monitoring equipment is deployed in the streets of the city.In the face of massive surveillance video data,the traditional processing method is weak,so building an intelligent monitoring system that can automatically supervise,alarm and track has become the future development trend.Person re-identification technology is the core algorithm of the system,and its status is very important,which has attracted the attention of domestic and foreign researchers and has invested many researches.This article first briefly introduces the necessity,current situation,difficulties and challenges of person re-identification research;then for the problems of pedestrian image misalignment,occlusion and irrelevant information interference,this paper mainly does the following algorithm innovation and research work:1.In-depth study of the development process and theoretical knowledge of person re-identification.First,a detailed overview of a variety of person re-identification algorithms,analysis and comparison of the difficulties and deficiencies solved by different algorithms;according to the current status and challenges facing,elaborated on the characteristics and advantages of deep learning-based person re-identification algorithm.Aiming at the problems to be solved in this paper,a new person re-identification algorithm based on deep learning is proposed based on the existing algorithms.2.A method of supervising joint learning by using global feature information and local feature information is proposed.Firstly,for the problem of pedestrian misalignment,the global feature and local feature of the image are used to construct a pedestrian feature matrix to achieve pedestrian alignment;for the occlusion problem,the correlation between the spatial information of the picture and the channel information can be used to provide a certain area feature.The picture is segmented according to the human space structure to obtain local features,and the local features are used to assist in the supervisionand learning of global features,and the more robust features are learned.Use global features and local features for supervised learning twice to better express features.3.Combined with other image processing technologies in the field of computer vision,image segmentation technology is used to process pedestrian images.At present,most of the image segmentation for person re-identification is to segment the image into strips and extract the features of each part.This can improve the performance of the algorithm,but it cannot fundamentally solve the impact of irrelevant information on feature extraction.In this paper,the semantic information of the image is used to perform foreground segmentation to eliminate the influence of background and other irrelevant information on pedestrian feature extraction,and improve the performance of the algorithm.Designed an actual monitoring system,apply some of the innovations in this article to the actual system,improve the performance of the system,and quickly and effectively complete the tasks related to person re-identification.