MultiSenseBadminton: Wearable Sensor-Based Biomechanical Dataset for Evaluation of Badminton Performance.

Area of Science:

• Sports Science
• Biomechanics
• Data Science

Background:

• The sports industry increasingly uses synchronized sensors for personalized training.
• A lack of comprehensive badminton datasets hinders advanced analysis and feedback.
• Existing data collection methods do not capture multi-perspective player actions effectively.

Purpose of the Study:

• To introduce a novel multi-sensor badminton dataset for forehand clear and backhand drive strokes.
• To address the scarcity of detailed badminton action data for research and training.
• To provide resources for understanding player biomechanics across different skill levels.

Main Methods:

• Collected multi-sensor data (eye tracking, body tracking, muscle signals, foot pressure) from 25 players (beginner, intermediate, expert).
• Recorded 7,763 badminton swing instances, including video, sound, and ball landing details.
• Incorporated coach interviews to ensure dataset usability and relevance for training.

Main Results:

• Developed a comprehensive dataset with diverse sensor modalities and detailed annotations.
• Included stroke type, skill level, hitting location, and sound data for in-depth analysis.
• Validated the dataset's utility with a proof-of-concept machine learning model.

Conclusions:

• The multi-sensor badminton dataset is a valuable resource for advanced training and research.
• The data enables detailed biomechanical analysis and personalized feedback for players.
• This dataset can significantly advance the development of intelligent badminton training systems.