RGB-D Rigid Multi-Body Dataset

Description: 

The RGB-D Rigid Multi-Body Dataset consists of 3 RGB-D videos of objects with different sizes (chairs, box/watering can, small box/teacan). The datasets have been recorded using an Asus Xtion Pro Live camera in a resolution of 640x480 at 30 Hz frame rate. Ground truth for the camera pose has been obtained with an OptiTrack Motion Capture system. We also manually annotated the moving objects in frames at every 5 seconds. The datasets are stored in a format compatible to Juergen Sturm's RGB-D benchmark dataset (http://cvpr.in.tum.de/data/datasets/rgbd-dataset).

We also calibrated the optical frame to this MoCap-intrinsic camera frame. Its transform is specified in the file mocap_cam_diff.txt. The format is tx ty tz qx qy qz qw.

Each dataset contains 1100 frames.

Download: 

• Chairs [zip archive, RGB video]
• Medium Box/Watering Can [zip archive, RGB video]
• Small Box/Tea Can [zip archive, RGB video]

Citation: 

If you refer to our dataset, please cite:

[1]	Jörg Stückler and Sven Behnke, "Efficient Dense 3D Rigid-Body Motion Segmentation in RGB-D Video". Proceedings of the 24th British Machine Vision Conference (BMVC), 2013. [pdf]

Last updated: October 17th, 2013 by Joerg Stueckler (stueckler _at_ ais.uni-bonn.de)