The Society of Convergence Knowledge Transactions ISSN:2287-8920(Print) 융복합지식학회논문지

The non-face-to-face and online markets are growing as social distancing has been intensified due to the recent COVID-19 impact. In particular, sports industry markets are also showing products that can be exercised alone at home due to the development of the IT industry. In this study, in order to meet the rapidly increasing demand for home training, we developed an “AI yoga trainer device” that can help healthcare. Morninghol yoga pose estimation using an inexpensive $99 Movidius Neural Compute Stick 2 instead of an expensive GPU. By implementing the algorithm, an experiment was conducted to recognize six poses of the sun worship posture, which is a basic yoga posture. As a result of the experimental analysis, the accuracy of the algorithm was 92.4%, which was measured to be a level that can be introduced in real life, and it was confirmed that the speed of deep learning calculation of visible and specific numerical values ​​using the Movidius Neural Compute Stick 2 was improved. Through this, we confirmed the feasibility of implementing IoT technology and deep learning-based home training technology.