Exercise and Performance Learning Assistant System for Self-Practice Dynamic Yoga by OpenPose and Fuzzy Based Design

Abstract

In this study, a learning assistant system based on OpenPose and fuzzy technologies is studied for self-practice dynamic Yoga at home. The skeleton information of the human body is extracted through OpenPose. In the pre-processing, the initial frame and key frames of the instructor are selected with Euclidean distance. In the system processing, the initial frame and key frames are used to determine the difference between the instructor and the user. The vector and the angle are calculated with the key points of the instructor and the user. The angle differences between them are used as the inputs of the scoring system based on fuzzy theory to calculate each score. This system synchronizes the time domain of the instructor and the user to effectively, and it correctly selects the user's corresponding frame, and then the system greatly reduces the amount of calculations without losing accuracy. The experimental results show that the proposed system can effectively detect and score the corresponding video frames between the user and the instructor.