无人机车辆数目估计及定位图像数据集

UAVVC 数据集包含500 和385 个用于训练/测试的图像。这些图像是从城市地区多个地点用无人机平台拍摄的超过10 个小时的视频中选择的，代表各种常见场景，包括广场、主干道、收费站、高速公路、十字路口和丁字路口。图像的分辨率为1080 × 540 像素。UAVVC 数据集包含50 种不同的计数场景。该数据集有四种视角，即高纬度的前视图、低纬度的前视图、侧视图和顶视图。在观看高度不受限制的情况下，帧内和帧间的比例会发生显著变化。在多种不同的天气条件下采集帧。所有这些拍摄条件增加了数据集的多样性，相比现有数据集，提出数据集更接近真实的交通状况。本数据仅供研究使用，使用该数据集，请引用论文： Y. Yang, G. Li, Z. Wu, L. Su, Q. Huang, N. Sebe, “Reverse Perspective Network for Perspective-Aware Object Counting”, CVPR 2020.

The UAVVC dataset contains 500 images for training and 385 for testing. These images are selected from more than 10 hours of video footage captured by drone platforms at multiple sites across urban areas, depicting a wide range of common traffic-related scenarios including squares, main roads, toll stations, expressways, intersections and T-junctions. All images have a resolution of 1080 × 540 pixels. The UAVVC dataset includes 50 unique object counting scenarios, and offers four shooting viewpoints: high-altitude front view, low-altitude front view, side view and top view. When the shooting altitude is unconstrained, the scale of objects varies considerably both within individual frames and across consecutive frames. Frames are collected under diverse weather conditions. All these shooting conditions enhance the dataset's diversity, making the proposed UAVVC dataset much closer to real-world traffic scenarios than existing datasets. This dataset is for research use only. Please cite the following paper when using this dataset: Y. Yang, G. Li, Z. Wu, L. Su, Q. Huang, N. Sebe, "Reverse Perspective Network for Perspective-Aware Object Counting", CVPR 2020.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于车辆数目估计和定位的无人机图像数据集，包含885张图像（500张训练、385张测试），覆盖城市中多种交通场景如广场和高速公路，图像分辨率为1080×540像素。数据集具有50种计数场景和四种不同视角，拍摄条件包括多样高度和天气，增强了数据多样性，使其更接近真实交通状况，适用于计算机视觉研究。