Underwater Diver Activity (UDA) dataset
收藏资源简介:
UDA数据集是由明尼苏达大学研究团队创建的首个水下潜水员活动数据集,旨在解决水下人机协作场景中活动识别数据稀缺的难题。该数据集包含2640张经过语义分割标注的高清图像(分辨率1920×1080),涵盖六类关键交互活动,数据来源于封闭水域环境中真实的人机协作试验视频片段。数据集通过SAM模型生成像素级标注并经过人工校验,精确标注了潜水员、机器人及目标物体的边界信息。该数据集主要应用于水下机器人感知与决策领域,为训练深度学习模型提供基础数据支持,助力实现智能水下协作系统的开发。
The UDA dataset is the first underwater diver activity dataset created by the research team at the University of Minnesota. It was developed to solve the problem of scarce activity recognition data in underwater human-robot collaboration scenarios. This dataset includes 2640 high-definition images with a resolution of 1920×1080, all annotated with semantic segmentation labels, covering six key interactive activities. The data is derived from real human-robot collaboration test video clips taken in confined water environments. Pixel-level annotations for this dataset were generated using the SAM model and manually verified, with precise boundary annotations for divers, underwater robots and target objects. This dataset is mainly applied in the field of underwater robot perception and decision-making, providing foundational data support for training deep learning models and facilitating the development of intelligent underwater collaborative systems.
数据集名称
Underwater Diver Activity (UDA) Dataset
数据集概述
该数据集是首个专门用于水下潜水员活动识别的数据集,旨在帮助自主水下航行器(AUV)理解并识别潜水员的当前活动,以促进有效的人机协作。
数据集内容
- 图像数量:超过2400张经过语义分割的图像。
- 场景:多个人类-机器人协作环境下的水下场景,展示了潜水员的各种活动。
- 标注信息:图像具有像素级别的标注,标注对象包括潜水员、机器人和感兴趣的目标物体。
数据用途
用于训练和评估基于Transformer的活动识别框架,该框架通过学习场景中的关键元素(如潜水员、机器人及其交互)的时空特征,实现对潜水员活动的识别。
获取方式
数据集可通过Google Drive链接下载: https://drive.google.com/file/d/1mVapKpNWNM8bUro2kssiElckZDdccEtI/view?usp=sharing





