Towards Machine Learning Framework for Badminton Game Analysis Using - Modified TrackNet - FRCNN and MoveNet-RNN - Hiba Abdulaziz Master of Science Thesis

Aug 4, 2026

Badminton offers strong value for players, coaches, and researchers through detailed game analysis, yet many existing methods require costly resources. This work presents an affordable machine learning framework for badminton analytics by applying three models: modified TrackNet for shuttlecock tracking, Faster-RCNN for player monitoring and shot detection, and MoveNet-RNN for pose-based shot type classification. A expanded dataset supports accurate shuttlecock localization and comprehensive shot-type labeling. Evaluation uses accuracy and computational efficiency, covering shuttlecock tracking, player tracking, and shot type classification. Results show modified TrackNet excels at shuttlecock tracking, while MoveNet-RNN achieves superior shot-type classification by leveraging recurrent modeling for sequential data, supporting an initial step toward a recommendation system for amateur play.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages41

Reading Time1 h 43 min

AccessFree to download

1 Introduction

2 Literature Review

2.1 Overview

2.2 Background

2.3 Related Work

3 Methodology and Design

3.1 Overview

3.2 Shuttlecock Tracking Using Modified TrackNet

3.3 Player Tracking and Shot Type Detection

4 Results and Discussion

4.1 Overview

4.2 Experimental Setup

4.3 Evaluation Metrics

4.4 Shuttlecock Tracking Experimental Results

4.5 Player Tracking and Shot Type Detection Experimental Results

5 Conclusion

5.1 Summary of Work

5.2 Limitations

5.3 Future Work

Bibliography

Appendix A

Towards Machine Learning Framework for Badminton Game Analysis Using: Modified TrackNet FRCNN and MoveNet-RNN by Hiba Abdulaziz A creative component submitted to the graduate faculty in partial fulfillment of the requirements for the degree of MASTER OF SCIENCE Major: Computer Science Program of Study Committee: Simanta Mitra, Major Professor Gurpur M Prabhu, Major Professor The student author, whose presentation of the scholarship herein was approved by the program of study committee, is solely responsible for the content of this creative component. The Graduate College will ensure this creative component is globally accessible and will not permit alterations after a degree is conferred. Iowa State University Ames, Iowa 2023 Copyright ©Hiba Abdulaziz, 2023 . All rights reserved. Abstract Badminton, a widely popular sport, holds significant potential for players, coaches, and researchers to gain valuable insights through game analysis. However, existing analysis methods often come with high costs and resource requirements. Machine learning techniques offer a promising avenue for automating and enhancing badminton game analysis while maintaining affordability and accessibility. This study explores the application of three machine learning models, namely modified TrackNet, FRCNN, and MoveNetRNN, for analyzing badminton games. The dataset utilized for training and evaluation is expanded to facilitate accurate shuttlecock tracking and comprehensive shot type classification. Performance evaluation is conducted based on accuracy and computational efficiency, encompassing shuttlecock tracking, player monitoring, and shot type classification using a custom badminton game dataset. The findings demonstrate the superior shuttlecock tracking capabilities of the modified TrackNet model, while the MoveNetRNN model exhibits exceptional performance in shot type classification. Leveraging recurrent neural networks (RNN) in MoveNet-RNN enhances the handling of sequencerelated data, resulting in improved shot classification accuracy. This study introducesan affordable analysis framework as an initial step toward the development of a recommendation system aimed at enhancing the gameplay of amateur players. Contents Abstract i List of Figures iv List of Tables v 1 Introduction 1 2 Literature Review 3 2.1 Overview .................................... 3 2.2 Background ................................... 3 2.2.1 Ball Tracking .............................. 3 2.2.2 Object Detection & Tracking ..................... 4 2.2.2.1 Traditional methods ..................... 4 2.2.2.2 Deep learning-based methods ................ 5 2.2.3 Action Recognition ........................... 6 2.3 Related Work .................................. 7 3 Methodology and Design 9 3.1 Overview .................................... 9 3.2 Shuttlecock Tracking Using Modified TrackNet ............... 10 3.2.1 Shuttlecock Data Collection ...................... 11 3.2.2 Modified TrackNet Training ...................... 12 3.3 Player Tracking and Shot Type Detection .................. 13 3.3.1 Method 1: Faster-RCNN ........................ 13 3.3.1.1 Player/Shot-Type Data Collection ............. 14 3.3.1.2 FRCNN Training ....................... 15 3.3.2 Method 2: Movement Analysis and Shot Type Classification using MoveNet and RNN ........................... 16 3.3.2.1 Human Pose Data Collection ................ 16 3.3.2.2 MoveNet-RNN model Training ............... 17 4 Results and Discussion 19 4.1 Overview .................................... 19 Contents iii 4.2 Experimental Setup .............................. 19 4.3 Evaluation Metrics ............................... 20 4.4 Shuttlecock Tracking Experimental Results ................. 21 4.4.1 Dataset ................................. 21 4.4.2 Modified TrackNet Results ...................... 21 4.4.3 Discussion ......................

FAQs

Which models are used for different badminton analysis tasks in this study?

Modified TrackNet is used for shuttlecock tracking, Faster-RCNN supports player tracking and shot detection, and MoveNet-RNN performs pose-based shot type classification.

How is recurrent modeling helpful for shot type classification in MoveNet-RNN?

MoveNet-RNN leverages recurrent neural networks to better handle sequential pose data, improving shot classification accuracy.

What evaluation criteria are applied to compare the models?

Performance is assessed using accuracy and computational efficiency across shuttlecock tracking, player monitoring, and shot type classification.

Aug 4, 2026

Upload by Olivia Brown