Karate First Kata Performance Analysis and Evaluation with Computer Vision and Machine Learning

Karate kata is a martial art that involves performing a sequence of movements with accuracy and speed. In this paper, we present a system capable of automatically recognizing and evaluating karate performances using computer vision and machine learning. Our system comprises four main steps: pose estimation, movement detection, feature extraction, and classification. We utilize the ensembling of two models and 14 different binary classification models in the classification step. We have successfully developed a method that compensates for camera shake in movement detection within our karate performance recognition system. Subsequently, we compute features such as skew, skewness, kurtosis, and covariance. Lastly, a machine learning model is employed to classify the first karate kata video. We curate a dataset of karate kata videos and achieve an impressive accuracy of 91.99% in our classification task.