Design And Implementation Of Pedestrian Re-identification For Security Monitoring

With the rapid development of intelligent security monitoring technology,a lot of monitoring information has been generated.Person Re-Identification methods based on deep learning can intelligently analyze monitoring information.Most person reidentification methods only focus on visible images,but the monitor images in poorly illuminated scenes.are near-infrared images.Different images have modality differences which is difficult to complete the task of person re-identification in monitorting scenes all day.Most researches focus on the person images,while few pieces of research on person detection for security monitorting.This paper combines visible images with nearinfrared images to study person re-identification methods for security monitoring systematically.The main work content is as follows:(1)The cross-modality person detection algorithm based on the fusion of visible images and the near-infrared image is studied.On the back of the ALFNet target detection network,the mid-term fusion two-way network structure is designed which fuses different modality images.Besides,an kind of illumination information weighted method is proposed.which weights the classification and regression results of person detection using illumination information weights on multiple scale feature maps The experimental results show that the cross-modality person detection method designed in this paper can effectively reduce the loss rate of person detection.(2)The cross-modality person re-identification algorithm of visible image and the near-infrared images is studied.Combined with the existing researches,a two-way network model is designed to extract the characteristics of different modality images,and bi-directed center ranking loss function is proposed for the problem of modality different existing in visible image and the near-infrared image.To capture the distinguishing characteristics of person in a specific field,the Generalized-Mean Pooling layer is used to replace the Max pooling layer in the network structure.Experiments verify that the cross-modality person re-identification method effectively improves the accuracy of person matching.(3)The above-mentioned cross-modality person detection algorithm and crossmodality person re-identification algorithm are integrated to design and implement a person re-identification application software for security monitoring.Which is tested by monitoring images to verify the design.The result shows that software can complete the task person re-identifying for monitorting images all day.