Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An All Deep System for Badminton Game Analysis

Po-Yung Chou, Yu-Chun Lo, Bo-Zheng Xie, Cheng-Hung Lin, Yu-Yung Kao

TL;DR

The paper presents an all-deep learning system for CoachAI Badminton 2023 Track1, tackling 11 precise badminton-event targets from video by coupling a robust object-detection stage (shuttlecock, players, court, net, racket) with specialized post-processing modules. They retrofit TrackNet-style shuttlecock detection with an asymmetric U‑NetTinySeeker to improve precision, followed by denoising and trajectory reconstruction; the system is augmented with X3D, EfficientNet, Transformer-based models, and VitPose-based pose cues for downstream tasks such as hit timing, location, ball type, and winner prediction. Evaluations show a 0.78/1.0 challenge score on the track and include quantitative comparisons (e.g., F1 gains from 0.93866 to 0.94255 for shuttlecock detection, sliding-window X3D for hit timing). The codebase is released on GitHub, enabling reproducibility and practical adoption for AI coaching and sports analytics; the work demonstrates that a deeper, data-driven approach can achieve precise, task-specific badminton analytics beyond raw detection visuals.

Abstract

The CoachAI Badminton 2023 Track1 initiative aim to automatically detect events within badminton match videos. Detecting small objects, especially the shuttlecock, is of quite importance and demands high precision within the challenge. Such detection is crucial for tasks like hit count, hitting time, and hitting location. However, even after revising the well-regarded shuttlecock detecting model, TrackNet, our object detection models still fall short of the desired accuracy. To address this issue, we've implemented various deep learning methods to tackle the problems arising from noisy detectied data, leveraging diverse data types to improve precision. In this report, we detail the detection model modifications we've made and our approach to the 11 tasks. Notably, our system garnered a score of 0.78 out of 1.0 in the challenge. We have released our source code in Github https://github.com/jean50621/Badminton_Challenge

An All Deep System for Badminton Game Analysis

TL;DR

The paper presents an all-deep learning system for CoachAI Badminton 2023 Track1, tackling 11 precise badminton-event targets from video by coupling a robust object-detection stage (shuttlecock, players, court, net, racket) with specialized post-processing modules. They retrofit TrackNet-style shuttlecock detection with an asymmetric U‑NetTinySeeker to improve precision, followed by denoising and trajectory reconstruction; the system is augmented with X3D, EfficientNet, Transformer-based models, and VitPose-based pose cues for downstream tasks such as hit timing, location, ball type, and winner prediction. Evaluations show a 0.78/1.0 challenge score on the track and include quantitative comparisons (e.g., F1 gains from 0.93866 to 0.94255 for shuttlecock detection, sliding-window X3D for hit timing). The codebase is released on GitHub, enabling reproducibility and practical adoption for AI coaching and sports analytics; the work demonstrates that a deeper, data-driven approach can achieve precise, task-specific badminton analytics beyond raw detection visuals.

Abstract

The CoachAI Badminton 2023 Track1 initiative aim to automatically detect events within badminton match videos. Detecting small objects, especially the shuttlecock, is of quite importance and demands high precision within the challenge. Such detection is crucial for tasks like hit count, hitting time, and hitting location. However, even after revising the well-regarded shuttlecock detecting model, TrackNet, our object detection models still fall short of the desired accuracy. To address this issue, we've implemented various deep learning methods to tackle the problems arising from noisy detectied data, leveraging diverse data types to improve precision. In this report, we detail the detection model modifications we've made and our approach to the 11 tasks. Notably, our system garnered a score of 0.78 out of 1.0 in the challenge. We have released our source code in Github https://github.com/jean50621/Badminton_Challenge
Paper Structure (11 sections, 4 figures, 2 tables)

This paper contains 11 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Object Detection
  3. Post-process Methods
  4. Number of shots, Hit frame and Hitter
  5. BallHeight, LandingX/LandingY, HitterLocationX/HitterLocationY, DefenderLocationX/DefenderLocationY
  6. Backhand
  7. RoundHead
  8. BallType
  9. Winner
  10. Demoenstraction Video
  11. Conclusion

Figures (4)

  • Figure 1: U-Net and Asymetric U-Net
  • Figure 2: The illustration of input data for the "Location" prediction.
  • Figure 3: The model of RoundHead task
  • Figure 4: The RoI for the input of the winner model