BON

BON数据集是由IEEE Signal Processing Society创建的一个大型公开办公室活动识别数据集，使用胸戴式GoPro Hero相机在巴塞罗那、牛津和内罗毕三个地点收集。该数据集包含18种常见的办公室活动，分为人-对象交互、人-人交互和本体感受活动三类。数据集通过5秒的视频片段进行标注，总计2639个视频片段，涉及25名参与者。BON数据集旨在通过提供多样化的办公室环境数据，支持更复杂的模型训练，如深度学习，以解决办公室环境中的活动识别问题，特别是在人机交互、辅助计算机视觉和机器人技术领域。The BON Dataset is a large-scale public office activity recognition dataset created by the IEEE Signal Processing Society. It was collected using chest-worn GoPro Hero cameras at three locations: Barcelona, Oxford, and Nairobi. The dataset contains 18 common office activities, categorized into three groups: Human-Object Interaction (HOI), Human-Human Interaction (HHI), and Proprioceptive Activities. All samples are annotated as 5-second video clips, with a total of 2639 clips involving 25 participants. The BON Dataset aims to support the training of more complex models such as deep learning for solving office activity recognition tasks by providing diverse office environment data, particularly in the fields of human-computer interaction, assistive computer vision, and robotics.

搜集汇总

背景与挑战

背景概述

BON是一个用于办公室活动识别的第一人称视角视频数据集，数据收集自巴塞罗那、牛津和内罗毕三个地点，涉及多个参与者和办公室环境。与先前版本相比，视频片段数量增加了六倍，达到2639个，旨在支持位置无关的框架开发。数据集发布于2021年，文件大小为37.37 GB，适用于人工智能、计算机视觉和机器学习研究。