Reproduce “BST: Badminton Stroke-type Transformer for Skeleton-based Action Recognition in Racket Sports” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jing-Yuan Chang
Researchers at National Tsing Hua University developed the Badminton Stroke-type Transformer (BST), a new architecture combined with a specialized video clipping strategy, to classify player stroke-types in racket sports. This approach effectively integrates player skeletal pose, position, and shuttlecock trajectory, achieving superior accuracy in distinguishing subtle and rapid movements compared to previous methods.
Key insights
- Shuttlecock trajectory is not merely an auxiliary feature but a primary, non-deceptive, and indispensable source of information for robust stroke classification in racket sports, providing unequivocal data about the actual stroke.
- A strategically designed video clipping method that captures inter-stroke context, including partial opponent actions, is crucial for providing sufficient temporal information for accurate stroke recognition in fast-paced racket sports.
- Integrating player court position with skeletal poses and employing mechanisms to clean noisy trajectory data and attribute it to the 'aiming' player significantly enhances the model's ability to differentiate between fine-grained stroke types.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.