RRIS40 Test Dataset

This dataset is a set of spatially aligned and temporally synchronized records of random human movements from a marker-based motion capture system and eight video cameras. The main purpose of this dataset is to benchmark the 3D accuracy of any multi-camera human motion capture method.

Specifications

• 10 subjects
• total of about 0.1 million frames (8 images per frame) with raw 3D marker positions.
• 3D data from a marker-based motion capture system (Qualisys)
  • 200 fps
  • positions of 40+6 markers in TSV format (Labels and definitions of all markers can be found here)
• Data from RGB Video Cameras
  • 8 synchronized global shutter cameras
  • 1920 x 1200 resolution
  • 50 fps
  • h.264 encoding
  • intrinsic calibration parameters
  • extrinsic calibration parameters (in the Qualisys coordinate reference frame)
• Frame-level hardware synchronization is used between Qualisys’ Miqus sync unit and all the RGB cameras.
• The formulation to interpolate 3D marker positions at the center of frame exposure time and projection to the video frame is provided in the data viewer source code.

Organization

• Data from each subject is stored in one zip file such as SN123g.zip or FT321g.zip. Each zip file contains one or more record folders inside.
• Each record folder has the following structure
  • YYYY-MM-DD-HH-mm-ss/ is called a record folder which contains data from only one continuous record.
    • qualisys.tsv contains data from the marker-based motion capture system
    • metadata.pkl contains metadata on the video record
    • systemCalibration.pkl contains calibration parameters of all the video cameras
    • keepMarker{camera_id}.h264 contains encoded video content.
    • keepMarker{camera_id}.seek Do not remove this file.
    • frameInfo{camera_id}.dat Do not remove this file.

Download

You can download them here or with mirror link. Password can be found in the paper.

WARNING: All the files are ~21GB in total.

Want to visualize the dataset quickly?

The runnable Python source code is available here. It simply playback a video with an overlay of 2D marker projections. This code allows you to understand our file structure to continue your work.

Cite Us

If you gain something from our dataset, please cite our publication. Volume and number will be updated after the official release. Note that you can cite an early access article with this IEEE guideline.

@ARTICLE{10591328,
  author={Jatesiktat, Prayook and Lim, Guan Ming and Lim, Wee Sen and Ang, Wei Tech},
  journal={IEEE Journal of Biomedical and Health Informatics}, 
  title={Anatomical-Marker-Driven 3D Markerless Human Motion Capture}, 
  year={2024},
  volume={},
  number={},
  pages={1-14},
  keywords={Three-dimensional displays;Solid modeling;Motion capture;Feature extraction;Deep learning;Cameras;Accuracy;Anatomical landmarks;biomechanics;data collection;deep learning;markerless motion capture},
  doi={10.1109/JBHI.2024.3424869}
}

Additional Paper Supplementary

Qualitative Results

• Videos of qualitative results on assistive outfits are accessible through this link or this mirror link.

Contact Us

• Prayook Jatesiktat. prayook001[at]e[dot]ntu[dot]edu[dot]sg
• Guan Ming Lim. guanming001[at]e[dot]ntu[dot]edu[dot]sg

We are from Nanyang Technological University.