SRPRID: Pedestrian Re-Identification Based on Super-Resolution Images

Abstract

Intelligent security expects to avoid the occurrence of robbery, theft, and other undesirable situations through video surveillance. In video surveillance, images of human faces undimmed are not easily available, so pedestrian re-identification (person ReID) is an alternative technique which attracts a mount of researchers attention. Person ReID is a technique used to match pedestrian images across cameras. Due to the interference of shooting angle and camera quality, it is difficult to obtain the images of high resolution, no obstructions, simple backgrounds and similar posture, which brings great challenges to the research of person ReID. Most existing methods of pedestrian re-identification ignore the inconsistency of resolution, and they are based on the assumption that all images have similar and high enough resolution by default. In this paper, we propose a hybrid framework, Super-Recognition of Pedestrian Re-Identification (SRPRID), in order to strengthen pedestrian re-identification based on multi-resolutions images captured by disparate cameras. Particularly, residual dense block (RDB) and Integrated Attention (InnAttn) block are merged to SRPRID. It is worth mentioning that the rank_1 accuracy of our method outperforms the state-of-art method by 17.2 points (86.9%-69.7%) on CUKH03 dataset of extremely challenging.