A dual-view videos and two-dimensional pose annotation dataset of Taekwondo Poomsae (Taegeuk 1–8) for human motion analysis
- 1Department of Recreation Management; Corresponding author.
- 2Department of Computer Science & Information Engineering
Abstract
Computer vision in sports increasingly supports judging, analytics, training, and injury prevention, yet Taekwondo’s high-speed kicks introduce motion blur, self-occlusion, and large joint excursions that challenge models trained on everyday actions. We release a synchronized dual-view dataset of standardized Taekwondo Poomsae (Taegeuk 1–8), recorded concurrently from frontal and lateral viewpoints in Full HD (1920 × 1080, 30 fps). Each frame includes normalized 2D coordinates and confidence scores for 12 major joints (shoulders, elbows, wrists, hips, knees, ankles), distributed as frame-aligned CSV files paired with the videos. Dual viewpoints provide complementary evidence for fast, large-amplitude lower-limb motion, enabling robust 2D pose model training/evaluation, cross-view generalization tests (e.g., train-frontal/test-lateral), and reproducible benchmarks for sports motion analysis. This resource addresses the scarcity of sports-specific, high-speed data and supports fair, replicable evaluation pipelines in computer vision.
Related
- DermArtifactDB: a multi-label dataset for artifact annotation in public dermoscopic image repositories.
- Short-term Rhodiola rosea supplementation improves premenstrual symptoms, reaction time, and perceived exertion in female taekwondo athletes: a randomised double-blind placebo-controlled trial
- Calibration and processing sensitivity of time-embedded three-dimensional electrocardiographic descriptors: a two-dataset benchmark
- Caffeine Intake and Performance Outcomes in Taekwondo Athletes: A Systematic Review and Meta-Analysis
- Optimizing hydration in Chinese youth taekwondo athletes: evidence-based daily and training session-specific fluid intake recommendations.
BibTeX
@article{huang2026dualview,
title = {{A dual-view videos and two-dimensional pose annotation dataset of Taekwondo Poomsae (Taegeuk 1–8) for human motion analysis}},
author = {Chien-Chung Huang and Qing-Wei Zheng},
journal = {Data in Brief},
year = {2026},
month = aug,
doi = {10.1016/j.dib.2026.113324},
url = {https://doi.org/10.1016/j.dib.2026.113324},
note = {CC BY},
}
RIS
TY - JOUR
TI - A dual-view videos and two-dimensional pose annotation dataset of Taekwondo Poomsae (Taegeuk 1–8) for human motion analysis
AU - Chien-Chung Huang
AU - Qing-Wei Zheng
PY - 2026
DA - 2026/08/01
JO - Data in Brief
DO - 10.1016/j.dib.2026.113324
UR - https://doi.org/10.1016/j.dib.2026.113324
AB - Computer vision in sports increasingly supports judging, analytics, training, and injury prevention, yet Taekwondo’s high-speed kicks introduce motion blur, self-occlusion, and large joint excursions that challenge models trained on everyday actions. We release a synchronized dual-view dataset of standardized Taekwondo Poomsae (Taegeuk 1–8), recorded concurrently from frontal and lateral viewpoints in Full HD (1920 × 1080, 30 fps). Each frame includes normalized 2D coordinates and confidence scores for 12 major joints (shoulders, elbows, wrists, hips, knees, ankles), distributed as frame-aligned CSV files paired with the videos. Dual viewpoints provide complementary evidence for fast, large-amplitude lower-limb motion, enabling robust 2D pose model training/evaluation, cross-view generalization tests (e.g., train-frontal/test-lateral), and reproducible benchmarks for sports motion analysis. This resource addresses the scarcity of sports-specific, high-speed data and supports fair, replicable evaluation pipelines in computer vision.
SN - 2352-3409
ER -