BadmintonGRF Dataset Enables Markerless Ground Reaction Force Estimation
A new dataset named BadmintonGRF has been introduced by researchers for estimating ground reaction force (GRF) in badminton without the use of markers. This dataset features eight synchronized RGB camera views at approximately 120 FPS, along with four Kistler force plates and Vicon motion capture (C3D) that operates without hardware genlock. The alignment process integrates human-validated events, automated quality checks, and time offsets for each camera, accompanied by uncertainty metadata. Tier 1 offers pose data, time-aligned GRF, and metadata under a CC BY-NC 4.0 license, facilitating a benchmark that correlates 2D poses to GRF. Tier 2 grants controlled access to raw RGB and C3D data. The public release includes 17,425 impact-segment archives from 10 participants, filling a gap in multimodal data for non-periodic court sports with high-quality sensing, aiding in markerless load estimation during realistic training scenarios.
Key facts
- BadmintonGRF records eight synchronized RGB views at ~120 FPS
- Includes four Kistler force plates and Vicon motion capture (C3D)
- No hardware genlock across modalities
- Alignment uses human-verified events, automated QA, and per-camera time offsets
- Tier 1 released under CC BY-NC 4.0
- Tier 1 benchmark maps 2D pose to GRF
- Tier 2 provides raw RGB and C3D under controlled access
- Public release contains 17,425 impact-segment archives from 10 subjects
Entities
Institutions
- arXiv
- Kistler
- Vicon