UAV123
收藏资源简介:
从低空无人机捕获的视频与流行的跟踪数据集 (如OTB50,OTB100,VOT2014,VOT2015,TC128和ALOV300) 中的视频本质上不同。因此,我们提出了一个新的数据集 (UAV123),其序列来自空中视点,其子集用于长期空中跟踪 (UAV20L)。我们新的UAV123数据集包含总共123个视频序列和超过110K帧,使其成为仅次于ALOV300的第二大对象跟踪数据集。所有序列都用直立的边界框完全注释。数据集可以很容易地与视觉跟踪器基准集成。它包括无人机数据集的所有边界框和属性注释。还请使用包含序列和跟踪器配置的修改后的文件 “configSeqs.m” 和 “configTrackers.m” 下载修改后的跟踪器基准。另外,请注意,文件 “perfPlot.m” 已根据本文中描述的属性进行了修改以进行评估。
Videos captured by low-altitude unmanned aerial vehicles (UAVs) are fundamentally different from those in popular object tracking datasets such as OTB50, OTB100, VOT2014, VOT2015, TC128, and ALOV300. Therefore, we propose a novel dataset (UAV123), whose sequences are collected from aerial viewpoints, with a subset dedicated to long-term aerial tracking named UAV20L. The newly proposed UAV123 dataset contains a total of 123 video sequences and over 110,000 frames, making it the second-largest object tracking dataset after ALOV300. All sequences are fully annotated with upright bounding boxes. This dataset can be easily integrated with visual tracker benchmarks. It includes all bounding box and attribute annotations for this UAV dataset. Please also download the modified tracker benchmarks using the modified files "configSeqs.m" and "configTrackers.m", which contain sequence and tracker configurations. In addition, please note that the file "perfPlot.m" has been modified based on the attributes described in this paper for evaluation.

- UAV123数据集首次发表,由Matlab团队在计算机视觉与模式识别会议上正式发布,标志着无人机目标跟踪领域的一个重要里程碑。
- UAV123数据集首次应用于目标跟踪算法评估,多个研究团队开始使用该数据集进行算法性能测试和比较。
- UAV123数据集被广泛接受为无人机目标跟踪领域的标准基准数据集,成为评估和改进跟踪算法的重要工具。
- UAV123数据集的扩展版本UAV20L发布,增加了更多的场景和目标类型,进一步丰富了数据集的内容和应用范围。
- UAV123数据集及其扩展版本在多个国际顶级会议上被引用和讨论,推动了无人机目标跟踪技术的发展和创新。
- 1UAV123: A Benchmark and Simulator for UAV TrackingUniversity of Stuttgart, Germany · 2016年
- 2Tracking-by-Detection with Contextual Information for UAV TrackingUniversity of Amsterdam, Netherlands · 2018年
- 3A Two-Stage Kalman Filter for Object Tracking in UAV VideosUniversity of California, Berkeley, USA · 2019年
- 4Deep Learning for Robust Visual Tracking in UAV VideosStanford University, USA · 2020年
- 5Multi-Object Tracking in UAV Videos Using Deep Reinforcement LearningMassachusetts Institute of Technology, USA · 2021年



