Research On Pedestrian Re-ID (Re-ID) Algorithm Based On Clothing Feature Transfer

Person re-identification(Re-ID),that is,the process of matching target people in non-overlapping cameras by using computer vision and machine learning methods.This technology can effectively help investigators find and track suspicious targets in massive videos,thereby increasing the case detection rate of public security organs,and can play an important role in the areas of intelligent security,public safety management,and pedestrian behavior analysis,so it has important research significance.This paper addresses the problem of the same target pedestrian wearing different clothes for re-identification in the query database and image library.When the target pedestrian changes clothes,the existing Re-ID algorithm fails,resulting in an unsuccessful matching process.Different from face recognition,gait recognition,and other person recognition technologies,person Re-ID relies heavily on clothing feature information in the matching process.For example,during the process of arresting the fugitive,the fugitive ran away after changing clothes in order to get rid of the tracking,making the clothes of the target person in the query image different from the clothes in the image library.This increases the difficulty of capturing,so clothing characteristics are important in the person matching process.In order to solve this problem,this paper proposes a person Re-ID algorithm based on clothing feature migration.The main idea is to give the image of the target pedestrian in the query database,and use the clothing feature migration algorithm to replace all pedestrian clothing in the image database with the query database Like the pedestrian clothes,the purpose is to eliminate the differences in clothing characteristics during the matching process,and finally to match the pedestrians by the characteristics of the human body.The main contents and innovations of this article are as follows:(1)In order to solve the problem of person Re-ID after changing clothes,a new data set PRDDC of target person wearing different clothes in the query database and image library is newly created in this paper.(2)This paper proposes a person Re-ID algorithm based on clothing feature migration,which is mainly divided into three modules,a pedestrian detection module,a clothing feature migration module,and a pedestrian matching module.The main idea is that given the image of the target pedestrian in the query database,through the clothing feature migration algorithm,the clothes of all pedestrians in the image database are replaced with the clothes of pedestrians in the query image to eliminate the influence of clothing characteristics in the matching process,and finally match by the characteristics of human limbs.After the data set PRDDC is processed by the clothes feature migration module and not processed,the pedestrian matching network is used to test the matching accuracy of 57.8% and 16.5%,respectively.The experimental results prove the effectiveness of the clothes feature migration algorithm.