MOR-UAV

MOR-UAV 数据集由 30 个视频序列组成，这些视频序列是从使用无人机平台在高速公路、立交桥、交通路口、市区、农业区等多个位置捕获的多个视频记录中收集的。这些视频代表了各种场景，包括遮挡、夜间、天气变化、相机运动、变化的高度、不同的相机视图和角度。视频以每秒 30 帧 (fps) 的速度录制，分辨率从 1280 × 720 到 1920 × 1080 不等。序列的平均、最小和最大长度分别为 364 .93 , 64 和 1 , 146。 MOR-UAV 数据集中大约 10, 948 帧用大约 89, 783 个表示移动车辆的边界框进行注释。这些车辆分为两类：汽车（80 个，340 个边界框）和重型车辆（9 个，443 个边界框）。边界框高度的平均、最小、最大长度分别为 29.011、6、181。类似地，边界框宽度的平均、最小、最大长度分别为 17.641、6、106。我们将 MOR-UAV 数据集中所有视频帧的大小调整为 608 × 608 × 3，以便在训练和评估中进行统一设置。The MOR-UAV dataset consists of 30 video sequences collected from multiple video recordings captured by unmanned aerial vehicle (UAV) platforms at various locations including highways, overpasses, traffic intersections, urban areas, agricultural zones, and others. These videos cover diverse scenarios, such as occlusion, nighttime environments, weather variations, camera motion, varying altitudes, different camera views and angles. The videos are recorded at 30 frames per second (fps), with resolutions ranging from 1280 × 720 to 1920 × 1080. The average, minimum, and maximum lengths of the sequences are 364.93, 64, and 1,146, respectively. Approximately 10,948 frames in the MOR-UAV dataset are annotated with around 89,783 bounding boxes representing moving vehicles. These vehicles are categorized into two classes: passenger cars (80,340 bounding boxes) and heavy-duty vehicles (9,443 bounding boxes). The average, minimum, and maximum heights of the bounding boxes are 29.011, 6, and 181, respectively. Similarly, the average, minimum, and maximum widths of the bounding boxes are 17.641, 6, and 106, respectively. We resize all video frames in the MOR-UAV dataset to 608 × 608 × 3 to ensure unified settings during training and evaluation.

搜集汇总

背景与挑战

背景概述

MOR-UAV是一个用于无人机视频中移动物体识别的基准数据集，包含30个视频序列，覆盖高速公路、市区等多种场景，并标注了10,948帧中的89,783个车辆边界框，分为汽车和重型车辆两类。数据集具有多样化的条件（如遮挡、夜间），视频分辨率从1280×720到1920×1080，所有帧统一调整为608×608×3以支持训练和评估，适用于目标检测和识别任务。