Real-time Detection of Taikyoku Shodan Karate Kata Poses Using Classical Machine Learning and Deep Learning Models

The research discusses the necessity for a precise feedback karate training system utilizing machine learning (ML) and deep learning (DL) models to facilitate real-time pose detection by offering instantaneous and accurate feedback. This system has the potential to revolutionize training methodologies, improve performance, and make significant contributions to the field of martial arts research.Various ML & DL algorithms were explored to achieve optimal results. Among the classical ML methods, the Random Forest, attained an accuracy of 98%. Additionally, DL models were also examined, with the Long-Short term memory model achieving an accuracy of 97%. The research presents an overview of the problem, methods for data collection and preprocessing, model training, and evaluation. The results emphasize the system's potential for real-time pose detection among karate practitioners, ultimately benefiting their training and progress.