Implementation of the MoveNet Method for Movement Accuracy Detection in Badminton

Abstract

Badminton is a sport that requires mastery of precise movement techniques. Conventional training methods that rely heavily on coaches’ visual observations are often subjective and have limitations in monitoring fast and detailed movements in real time, especially for players who train independently. This study aims to design and develop an Android-based mobile application that utilizes the MoveNet algorithm to detect and evaluate the accuracy of basic badminton movements. The system employs computer vision and pose estimation to track 17 human body keypoints and analyze seven fundamental movements: smash, drop shot, drive, serve, lob, backhand, and netting. The application is developed with TensorFlow Lite integration to enable efficient processing on mobile devices. Testing is conducted using the black-box testing method, and model performance is evaluated using a confusion matrix. The results show that the model is able to classify movements with an average accuracy of 78%, precision of 78%, and recall of 78%. The system is proven to provide real-time visual feedback in the form of a body skeleton and movement evaluation, making it effective as an objective and data-driven tool to support independent training.