Research On Person Re-Identification Based On Generative Adversarial Networks

Person Re-identification(re-ID)is an essential research topic in the field of computer vision,which refers to the recognition and matching of specific person identities from multiple non overlapping monitoring views.With the increase of monitoring network year by year,it is necessary to use computer vision technology to analyze and process massive video image data,and person re-ID technology has also attracted more and more attention.In this paper,the research background and significance of person re-ID are described,and the research status and main challenges at home and abroad are introduced.A person re-identification method based on generative adversarial network is proposed.This method can generate high-quality person images,increasing the number of training time samples and the diversity within the sample class,and keep the person identity features unchanged,which effectively improves the performance of person re-ID model.The most significant contributions are as follows:1.Using Mask R-CNN image semantic segmentation algorithm to segment high quality mask of the foregrounds of person images.2.Combined with mask and manual annotation of person attributes,an end-to-end Similarity Preserved Camera-to-Camera Generative Adversarial Network model is proposed,which can generate training images from original person images from one camera to others in the dataset,which can keep the foreground of the generated image unchanged and the background style consistent with the corresponding camera domains.3.A person re-ID model with good performance and discriminability is designed to verify the effectiveness of the algorithm proposed in this paper.Experiments on two public datasets,Market1501 and DukeMTMC-reID,demonstrate the effectiveness of the proposed method.Experimental results show that the proposed method,training with the generated person images,has higher accuracy of person identity matching and recognition than other advanced methods or other data augmentation methods based on the generative adversarial networks.