Deep Learning Based Technical Classification of Badminton Pose with Convolutional Neural Networks

Abstract

This research aims to identify and categorize badminton strategies using a Convolutional Neural Network (CNN) model combined with BlazePose architecture and Mediapipe Pose Solution tools, yielding understandable and practical results. The challenge of finding the best mobility strategy for badminton serves as the primary motivation for this study. The research employs an image recognition and supervised learning approach to classify poses in badminton training videos. The training data comprises various photos and images representing different badminton techniques, such as Service Technique and Smash Technique. After data processing, the CNN model is trained using the training data to identify and classify poses in badminton training videos. Testing is conducted using test data, and classification accuracy is evaluated using the CNN method. The results show that the CNN model implemented alongside BlazePose and Mediapipe Pose Solution achieves significant classification accuracy, ranging from 80% to 90%. Thus, this research presents an effective and practical method for classifying badminton strategies based on poses in training videos.

Keywords

• badminton
• classification
• convolutional neural networks
• deep learning
• estimation pose