AI-Powered Badminton Shot Classification

AI technology has catalyzed new frontiers across numerous domains, including sports analytics. Due to the diversity of sports, certain areas remain under-explored. This work will focus on bringing AI-driven analysis to the sport of Badminton. By leveraging computer vision techniques and ML models, we can analyze athlete performance by identifying shot selection. By examining their stroke preparation for conducting a type of shot, which differs subtly between shots, we can gain insights to their strengths and weaknesses. We developed two ML models for shot classification using official match data from BWF, categorizing shots into ‘lob’, ‘smash’, and ‘net’. Our results show that the Keras-Mediapipe model outperforms the YOLO-NAS model in shot classification, however, still requires further improvements to be applicable.