BadminSense: Enabling Fine-Grained Badminton Strokes Evaluation on Single Smartwatch

Paper Title

BadminSense: Enabling Fine-Grained Badminton Strokes Evaluation on Single Smartwatch

Publication Info

• Topic area: Wearable sensing for sports performance analysis
• Keywords: Badminton, smartwatch, stroke evaluation, wearable sensing, IMU, acoustic signals, skill analysis, racket sports, user study, machine learning

Background and Problem

• Problem / challenge: Amateur badminton players lack accessible tools for precise skill evaluation and improvement, relying on subjective feedback or limited external resources like videos.
• Significance: Providing quantitative feedback on stroke quality and impact location can enhance self-training efficiency and skill development.
• Motivation and related work: Prior systems, such as racket-mounted sensors and camera-based setups, offer precise tracking but suffer from practical limitations like invasiveness, setup complexity, and interference with gameplay. Existing smartwatch-based systems focus on basic metrics but lack fine-grained stroke evaluation capabilities.

Solution

• Proposed approach: BadminSense, a smartwatch-based system leveraging IMU and acoustic signals to analyze badminton strokes, including classification, quality rating, and shuttle impact location estimation.
• Novelty:
  1. Development of a fine-grained badminton stroke sensing system using a single smartwatch.
  2. Creation of an annotated dataset with 848 strokes, including expert-assessed quality ratings and impact locations.
  3. Implementation of stroke segmentation, classification, quality rating, and impact location estimation algorithms.
  4. Usability study demonstrating real-world applicability and user satisfaction.
• Procedure and key techniques:
  • Interviews with experienced players to identify design requirements and implementation insights.
  • Data collection using IMU and acoustic signals from a smartwatch, annotated with stroke type, quality rating, and impact location.
  • Machine learning models (e.g., SVM, SVR) for stroke classification and regression tasks.
  • Evaluation of system components (segmentation, classification, rating, location estimation) using cross-validation and user-independent strategies.

Results

• Concrete findings:
  • Stroke classification accuracy: 91.43% (user-independent).
  • Stroke quality rating error: 0.438 (mean absolute error, user-independent).
  • Impact location estimation error: 12.9% (normalized MAE).
  • Stroke segmentation accuracy: 99.41%.
• Advantage over baselines: Outperformed other machine learning models (e.g., Random Forest, Linear Regression) in classification and regression tasks; demonstrated unobtrusive sensing compared to racket-mounted or camera-based systems.
• Experiments / evaluation:
  • Dataset: 848 strokes from 12 players, annotated by 21 experts.
  • Metrics: Accuracy, MAE, false positive rate.
  • Usability study: 12 participants rated system reliability (M = 4.14), usability (M = 4.27), and engagement (M = 4.24) on a 5-point Likert scale.
• Limitations and future work:
  • Limited stroke types supported; plans to expand dataset and explore transfer learning.
  • Effects of racket string tension and handedness on performance need investigation.
  • Lack of shuttle speed estimation; future work to combine physical modeling and data-driven approaches.
  • Rule-based feedback mechanism; potential for context-aware advice using large language models.
  • Longitudinal studies to assess skill improvement over time.

Summary

BadminSense is a smartwatch-based system designed for fine-grained badminton stroke evaluation, addressing gaps in accessible self-training tools. It achieves high accuracy in stroke classification (91.43%) and impact location estimation (12.9% error) using IMU and acoustic signals. Usability studies confirm its reliability, usability, and engagement in real-world practice. Future directions include expanding stroke types, adapting to diverse racket attributes, estimating shuttle speed, and enhancing feedback mechanisms. BadminSense demonstrates the potential of wrist-worn devices for precise skill analysis in racket sports.