“可见光-近红外”行人数据集

在开放环境的监控环境中下，当行人处于黑暗环境中，可见光图像并不适用。因此，在低照度的室内场景和24小时室外监视系统中，使用黑暗环境中可成像的近红外图像是很有必要的。可见光与近红外行人图像的跨模态匹配问题在实际应用中亟待解决。但是，现有的数据集只包含可见光图像而没有近红外图像，难以支持“可见光-红外”跨模态行人重识别的研究。为解决此问题，本课题采集了一个同时包含可见光图像与近红外图像的数据集，名称为SYSU-MM01。

In open-world surveillance scenarios, visible-light images are inapplicable when pedestrians are in dark environments. Therefore, it is necessary to use near-infrared (NIR) images capable of imaging in dark environments for low-light indoor scenes and 24-hour outdoor surveillance systems. The cross-modal matching problem of visible-light and near-infrared pedestrian images urgently needs to be addressed in practical applications. However, existing datasets only contain visible-light images without near-infrared counterparts, making it difficult to support research on visible-infrared cross-modal person re-identification. To address this issue, this study collected a dataset containing both visible-light and near-infrared images, named SYSU-MM01.

搜集汇总

背景与挑战

背景概述

“可见光-近红外”行人数据集（SYSU-MM01）是一个专为行人重识别研究设计的跨模态数据集，包含可见光和近红外图像，旨在解决低照度环境下可见光图像不适用的问题。该数据集由中山大学创建，数据量1.5GB，包含45871个文件，支持“可见光-红外”跨模态匹配的研究，弥补了现有数据集缺乏近红外图像的空白。