Behavioral Analysis-based Labeling for Physical Exercise Posture Image Dataset using National Artificial Intelligence Computing Platform

Abstract

Abstract Recently, research on physical fitness posture estimation in the virtual space using artificial intelligence (AI) has been actively conducted. However, AI has been difficult due to a lack of fitness datasets and guidelines. To advance fitness posture estimation algorithm through analyzing fitness image provided by national artificial intelligence Platform known as AI-Hub (powered by National Information society Agency, NIA) in Korea and approaching it wider and deeper from the perspective of exercise prescription vacation and behavioral analysis. Through this advancement, this study intended to closely analyze fitness movements and guide correct exercise posture with screen and sound. Referring to image and labeling JSON (JavaScript Object Notation) file provided by AI-Hub, contents necessary and useful for the posture estimation algorithm from the perspective of exercise prescription were explained in writing, diagrams, and photos. The structure of data for 6 million consecutive and diverse fitness images and labeling data of scenes was analyzed. In addition to existing explanation, exercise state and posture of exercise in the data structure of 41 fitness images were analyzed and presented in detail. In addition, annotation and labeling characteristics were explained in detail with photo images. Conclusion: Detailed description and introduction of fitness image analyzed in this study would help developers of exercise coach web/ app services use fitness images by upgrading posture estimation algorithms using AI.