Automated Evaluation of Karate Practitioners’ Style during the First Kata Performance using Deep Learning and Pose Estimation

Automated motion analysis of human body movements has attracted a lot of interest recently since it has the ability to offer unbiased evaluations of performance in a variety of sports and martial arts. This study focused on analyzing the manner in which karate practitioners perform the first Kata, a critical component that reveals their command of fundamental motions and technique proficiency. This study suggests an automated method for analyzing karate practitioners’ movements from recorded video using pose estimation and a deep Convolutional Neural Network (CNN) model. The experimental findings show that our approach achieved a high level of accuracy 97%, and 94% particularly when employing different evaluation window sizes. This strategy provides fair and accurate decisions in sports arbitration by removing subjective biases that may result from human observers.