USVTrack
收藏资源简介:
USVTrack是一个用于复杂水面场景多目标跟踪的基准数据集,旨在促进无人水面车辆(USV)自主导航的研究。该数据集完全在船上采集,覆盖了多样化和复杂的水面场景,包括内陆和海洋表面场景,特点是包含大量小目标和多变的背景。具体来说,USVTrack包含27个视频序列,共计8,283个标注帧和超过10万个边界框。其中,3,223帧来自内陆场景(sequence_1到sequence_13),5,060帧来自海洋场景(sequence_14到sequence_27)。
USVTrack is a benchmark dataset for multi-object tracking in complex water surface scenarios, aiming to promote research on autonomous navigation of unmanned surface vehicles (USV). This dataset is fully collected onboard, covering diverse and complex water surface scenarios including inland and marine surface environments, and is characterized by a large number of small targets and variable backgrounds. Specifically, USVTrack contains 27 video sequences, totaling 8,283 annotated frames and over 100,000 bounding boxes. Among them, 3,223 frames are from inland scenes (sequence_1 to sequence_13), while 5,060 frames are from marine scenes (sequence_14 to sequence_27).
USVTrack数据集概述
数据集简介
USVTrack是一个专注于复杂水面场景下多目标跟踪的基准数据集,旨在推动无人水面艇(USV)自主导航技术的研究。该数据集完全基于船上采集,覆盖多样化的复杂水面场景,包括内陆和海洋场景,具有小目标占比高、背景变化大等特点。
关键特性
- 场景多样性:包含内陆和海洋两类水面场景
- 数据规模:27个视频序列,共8,283标注帧,超过100k边界框
- 内陆场景:3,223帧(sequence_1至sequence_13)
- 海洋场景:5,060帧(sequence_14至sequence_27)
- 对象类别:15个预定义类别
- 1-boat, 2-speedboat, 3-duckboat, 4-speedboat, 5-yacht
- 6-cruise, 7-oceanLiner, 8-aquaplane, 9-buoy
- 10-person, 11-disruptors, 12-reflection
- 13-building, 14-occlusion, 15-shine
数据集结构
{USVTrack ROOT}
|-- usvtrack
| |-- train
| | |-- sequence_1
| | | |-- img1
| | | | |-- 000001.jpg
| | | |-- gt
| | | | |-- gt.txt
| | | |-- seqinfo.ini
| | |-- ...
| |-- test
| | |-- ...
| |-- label.txt
| |-- information.txt
标注格式
遵循MOTChallenge标准格式:
<frame>, <id>, <bb_left>, <bb_top>, <bb_width>, <bb_height>, <conf>, <cls>, <vis>
评估方法
使用改进版TrackEval进行评估,要求跟踪结果格式为:
<frame>, <id>, <bb_left>, <bb_top>, <bb_width>, <bb_height>, <conf>, -1, -1, -1
数据获取
完整数据集将在ORCA-UBOAT Datasets发布。
更新时间
2025年5月10日:发布了USVTrack子集和评估代码。




