Automated Hit-frame Detection for Badminton Match Analysis

During my junior year, I had the opportunity to join the Software Security Laboratory, where I embarked on an exciting project aimed at advancing badminton sports analysis. This project focused on automatically detecting hit-frames from raw match videos, leveraging cutting-edge AI techniques. Through this endeavor, I gained extensive knowledge and hands-on experience in various advanced technologies.

Key Achievements and Learning Outcomes:

Trimming Module Development: I designed and developed a rally trimming module using convolutional neural networks (CNNs) to segment rally frame sequences from videos. This module achieved an impressive 81% accuracy, demonstrating the effectiveness of CNNs in sports video analysis.

Detection with RCNN: I implemented a keypoint region-based convolutional neural network (RCNN) to detect the player's joints. To meet our specific needs, I customized and finetuned the RCNN model to detect the border key points of the court, enabling precise identification of on-court players. This customization was crucial for accurate player tracking and movement analysis.

Novel Transformer for Shuttlecock Direction Prediction: I trained a novel transformer model that achieved over 92% accuracy in predicting shuttlecock direction sequences based on player joint sequences in the rally frames. This transformer-based approach provided robust and reliable predictions, enhancing the analysis of player strategies and movements.

Hit-frame Detection Accuracy: By evaluating our work with a publicly available badminton dataset, we achieved 96% accuracy in detecting hit-frames based on shifts in the shuttlecock's direction of motion. This high accuracy underscores the potential of AI in automating and improving sports analysis.

Awards and Recognitions:

Research Grant: National Science and Technology Council, R.O.C. awarded me a research grant for undergraduate students, recognizing the significance and potential impact of our project.

International ICT Innovative Services Award: Our project won first place in the Asia-Pacific category and third place in the AUO corporation-sponsored AI category. This recognition highlighted the innovation and practical application of our research in the field of sports analysis.

This project was a transformative experience that taught me how AI-powered services could assist with manual labor, making tasks more efficient and accurate. It ignited my passion for combining technology with sports and inspired me to explore the potential of starting my own business in the future. The knowledge and skills I acquired through this project have equipped me to contribute to the advancement of sports analysis and AI applications.

I am excited to continue exploring the intersection of technology and sports, contributing to advancements that can help athletes and coaches achieve their best performance. Feel free to explore my research and connect with me to discuss potential collaborations or insights into this exciting field!

• PyTorch
• Deep Learning
• Video Analysis
• Automated Badminton Analysis
• CNN
• RCNN
• Hit-frame Detection
• Transformer
• Shuttlecock Direction Prediction
• Keypoint Detection