high altitude UAV multi-modal tracking dataset

The dataset consists of visible and infrared multimodal video captured by a drone equipped with a hybrid sensor at an altitude of 500 meters, mainly containing 150 visible and infrared video pairs, totaling more than 12,000 frame pairs. The dataset contains 9 target categories such as trucks and pedestrians, covering a variety of scenes (such as city, suburb, lake, construction site, viaduct, basketball court, lawn, plastic runway, residential area, square, street, etc.) and lighting conditions (day, night, fog, sunny, etc.), significantly increasing the diversity and challenge of the dataset. In addition to the 12 general tracking challenge attributes in the HiAl dataset, challenges specific to the drone perspective, such as drone rotation and small targets, were added. These challenge attributes comprehensively cover the complex scenarios and practical problems that may be encountered in high-altitude UAV target tracking missions. In order to ensure the high quality of the data set and the accuracy of the annotation, we carried out fine modal registration for the video of both visible and infrared modes, and manually labeled the target boundary boxes of all frames, adopting a unified modal annotation method to evaluate the performance of the multi-modal tracking method more accurately.

本数据集由搭载混合传感器的无人机在500米高空采集的可见光与红外多模态视频构成，共包含150组可见光-红外视频对，总计超过12000帧图像对。本数据集涵盖卡车、行人等共9类目标，覆盖城市、郊区、湖泊、建筑工地、高架桥、篮球场、草坪、塑胶跑道、居民区、广场、街道等多种场景，以及白天、夜间、雾天、晴天等多样光照条件，显著提升了数据集的多样性与挑战性。本数据集除包含HiAl数据集原有的12项通用跟踪挑战属性外，还新增了无人机视角特有的挑战项，如无人机旋转、小目标等。这些挑战属性全面覆盖了高空无人机目标跟踪任务中可能遇到的复杂场景与实际问题。为保障数据集质量与标注准确性，本数据集对可见光与红外两种模态的视频进行了精细模态配准，并对所有帧的目标边界框进行了人工标注，同时采用统一模态标注方案，以更精准地评估多模态跟踪方法的性能。

搜集汇总

背景与挑战

背景概述

该数据集是一个高空无人机多模态跟踪数据集，包含150对可见光和红外视频，超过12,000帧对，覆盖9种目标类别和多种场景与光照条件，增加了数据多样性。数据集引入了无人机视角特有的挑战属性，如无人机旋转和小目标，并结合精细模态配准和手动标注，旨在准确评估多模态跟踪方法在复杂高空任务中的性能。