Person Re-identification System Integrating GFP-GAN and Yolov4 Models

Person re-identification (ReID) technology can find specific persons in different scenes on various conditions. In this an improved ReID model is designed and implemented combined with a generative adversarial model (GAN). Specifically, this paper uses yolov4 detection framework to detect the human body in the video, then employs the improved Aligned ReID++ network model to calculate the similarity of human body pictures, and finally uses the GFP-GAN model to generate a prior knowledge of the face area which effectively improves the amount of detail information of human body pictures. As the experiments show, our model could improve the detection accuracy of ReID and suppress the error detection rate on body areas with large image resolution. All experiments are evaluated on the Market-1501 dataset. At last, the ReID system has been implemented with PyQT which combine efficiency and convenience.