The RBO Dataset of Articulated Objects and Interactions

关节对象和交互的 RBO 数据集是 358 个 RGB-D 视频序列（67:18 分钟）的集合，这些视频序列是人类在不同条件（光线、透视、背景、交互）下操纵 14 个关节对象。所有序列都用刚性部件的姿势和通过运动捕捉系统获得的关节物体的运动学状态（关节状态）的基本事实进行注释。我们还提供这些对象的完整运动学模型（运动学结构和三维纹理形状模型）。在 78 个序列中，还提供了操作过程中的接触扳手。

The RBO Dataset for Articulated Objects and Interactions is a collection of 358 RGB-D video sequences totaling 67 hours and 18 minutes, in which humans manipulate 14 articulated objects under varying conditions including lighting, viewpoint, background, and interaction settings. All sequences are annotated with the poses of their rigid components and the ground-truth kinematic states (joint states) of the articulated objects acquired via a motion capture system. We also provide complete kinematic models for these objects, encompassing their kinematic structures and 3D textured shape models. For 78 of these sequences, contact wrenches collected during the manipulation process are additionally provided.

搜集汇总

背景与挑战

背景概述

该数据集包含358个RGB-D视频序列，记录了14个关节对象在不同条件下的交互过程，并提供了部件姿态、关节状态等标注。由柏林工业大学于2018年发布，部分序列还包含接触扳手数据。