Abstract: Cross-modal person re-identification (cm-ReID) technology aims to identify the same person across images from different modalities, such as visible light and infrared. It has important applications in intelligent systems and equipment for human-machine collaboration, the Internet of Everything, cross-domain integration, and ubiquitous intelligence. This paper proposes a data augmentation method for cross-modal person re-identification that performs data augmentation in the wavelength domain while preserving the structural information of visible images to bridge the gap between different modalities. Based on this, an edge AI terminal was designed and implemented using Rockchip's RK3588 chip, and the cross-modal person re-identification algorithm was deployed on it.

0 Introduction

Person re-identification (Re-ID) is widely used in various applications such as video surveillance, security, and smart cities. It aims to retrieve a person of interest across multiple non-overlapping cameras deployed at different locations. Due to its importance in intelligent video surveillance, it has attracted increasing attention from the computer vision community [1-3]. Currently, a large number of person re-identification models focus on visible-to-visible person image matching, which is the most common single-modal person re-identification task. As shown in Figure 1, given a detected person image, it is matched with a set of images captured by other non-overlapping cameras, where this gallery set contains data from all visible cameras. Due to variations in illumination, camera models