Research On Cross-domain Person Re-identification Method And System Implementation

In the current society,smart security is increasingly highlighting its value,and it is an important means to maintain the long-term stability of society.The automatic video analysis system is a subsystem of the intelligent security system,mainly extracting key person information from the surveillance video content.Person re-identification is one of the core technologies of automatic video analysis systems,and has become a research hotspot in the field of deep learning.In recent years,deep learning-based person re-identification methods have shown superior performance on open source datasets,but the performance on the cross-domain person re-identification problem is significantly degraded.This paper provides an in-depth analysis of the problems of cross-domain person re-identification,and proposes an effective cross-domain person re-identification method using deep learning methods,and implements a prototype person re-identification system for offline video.The main work and innovations in this paper are as follows.(1)From the perspective of the cross-domain problem,a cross-domain person re-identification method that fuses multiple source domains is proposed.In view of the obvious differences in person image styles between different source domain datasets,a person re-identification method that can fuse the imaging styles of multiple source domains is designed,trained using several different datasets,and tested on the MSMT17 dataset that is closest to the actual application scenario.The experimental results show that the proposed method can alleviate the degradation of person recognition performance caused by cross-domain to a certain extent.(2)The person re-identification system is designed and implemented.The person detection algorithm was firstly improved using the target detection algorithm YOLOv3;then combined with the proposed cross-domain person re-identification method to implement the person re-identification system process;and finally the person re-identification process was visualised through an interactive interface.The system was tested on Windows 10 platform and the results showed that the system can extract person features from the imported video data and frame the target person,and display the final person recognition result in the interactive window.