Research And System Implementation Of Improved Algorithm For Pedestrian Re-Identification Based On Image

Person re-identification is a technology to judge whether two images from non overlapping cameras are the same pedestrian.This technology has been widely used in the field of intelligent security monitoring.However,due to the significant changes of pedestrians in different cameras,the same pedestrian has large difference in different cameras and small difference in different pedestrians,which leads to misjudgment in pedestrian re recognition.Through consulting a large number of domestic and foreign related literature,learning and in-depth study of person re-identification theory and technology,it is found that the traditional person re-identification supervision learning method mainly relies on pre marked image data,but there are a lot of unmarked data in the actual security monitoring scene,which seriously limits the application of person re-identification technology in the field of security monitoring.To solve these problems,an improved person re-identification method based on generative learning and joint discrimination and a pedestrian re recognition method based on multi view attention mechanism based on view perception are proposedFirstly,the unlabeled data is generated by using the generative countermeasure network for semi supervised learning.In the process of using GAN to generate data,the realistic pedestrian image is generated by conditionally using the global and local information of pedestrian image.Aiming at the problem of significant intra class changes in different cameras,the robust feature representation is trained,so as to improve the pedestrian weight.The experimental results show that the proposed method is better than the benchmark method in the public dataset.Secondly,due to the lack of image data of pedestrians in each view,it is unable to deal with the problem of significant change of view angle.A pedestrian recognition feature extraction method based on multi view attention mechanism of view perception is proposed.Firstly,the fine pose information of pedestrians is extracted by using the PSE network.Secondly,the core area of pedestrian's different poses is selected by using the view perception attention model.At the same time,the Gan network is used for adversarial training to realize the effective inference of multi view features.Finally,by extending the cross neighborhood distance to rearrange the initial order,the accuracy of final person re-identification is obtained.The experimental results show that this method has a significant improvement compared with the benchmark method in the open data set.Thirdly,a person re-identification retrieval system based on resnet50 network is constructed.The interface adopts directui platform,and the core algorithm is implemented in Python.The system functions include basic function module(system login,system logout,system exit,system password modification,system operation supervision),user management module(user modification,user deletion,user log management),selection query image module,retrieval image visualization result module,retrieval image result XML file export module.