基于无人机的计算机视觉 VisDrone 数据集

VisDrone 是由无人机搭载相机采集的大规模航拍视觉基准数据集，包含 288 个视频片段、261908 帧视频及 10209 张静态图像，覆盖中国 14 个城市的城乡环境；含行人、车辆、自行车、三轮车等多类别目标，拥有超 260 万个人工标注框，标注包含类别、遮挡等属性，按五大任务划分子集，是无人机计算机视觉领域的核心研究资源

VisDrone is a large-scale aerial visual benchmark dataset collected by cameras mounted on unmanned aerial vehicles (UAVs). It includes 288 video clips, 261,908 video frames and 10,209 static images, covering urban and rural environments across 14 cities in China. The dataset features multiple object categories such as pedestrians, vehicles, bicycles and tricycles, with over 2.6 million manually annotated bounding boxes. Annotations cover attributes including object category and occlusion status, and it is divided into subsets based on five major tasks, serving as a core research resource in the field of UAV-based computer vision.

搜集汇总

背景与挑战

背景概述

VisDrone数据集是一个大规模无人机航拍视觉基准数据集，包含288个视频片段、超过26万帧视频和1万多张静态图像，覆盖中国14个城市的城乡环境，标注了行人、车辆等多类别目标，拥有超260万个人工标注框。数据集按图像目标检测、视频目标检测、单目标跟踪、多目标跟踪和人群计数五大任务构建，支持无人机场景下的深度学习模型训练与评测，广泛应用于交通监控、城市规划和安全评估等领域。