Page 357 - Kaleidoscope Academic Conference Proceedings 2024

VIRTUAL YOGA TEACHER WITH AI MODEL FOR YOGA POSTURE CORRECTION
FOR GOOD HEALTH

3
Swapna Yenishetti ; Dr.Ganesh Karajkhede Lakshmi Panat
1
2;
1,2,3 Centre For Development of Advanced Computing

growing in the coming years. The global yoga market size is
ABSTRACT forecast to reach $66.2 billion by 2027 and is expected to
have a compound annual growth rate of 9.6% from 2021 to
With the advent of preventive and promotive health in the 2027.
world, driven by the WHO’s Sustainable Development Goal
for Healthcare to ensure healthy lives and promote well- Practicing yoga in wrong way can lead to various negative
being for all ages it is a necessity to develop Artificial effects on the body. It may include muscle injury, joint pain,
Intelligence driven tools for Yoga. Yoga is the art and back pain, nerve compression, pressure on internal organs
science of healthy living and has a history spanning leading to injury, breathing problems, psychological effects
thousands of years. Hence there is genuine need of AI etc. To avoid these negative effects, it's crucial to practice
model for Yoga posture correction, to reap the benefits of yoga under the guidance of a qualified instructor. For
practice of Yoga. Developing AI models for yoga posture beginner or attempting advanced poses focusing on proper
correction has the potential to enhance the experience for alignment and breathing for a safe and effective yoga
yoga followers and acting as a Virtual Yoga teacher. In this practice.
study, we are focusing on identification of 10 important
yoga poses performed by the user with the help of Pre - Therefore, a need for a ‘Yoga Pose Detection’ algorithm
trained Move Net model and transfer learning. Using such arises, that can check and correct yoga poses, while also
deep learning algorithms, an individual can understand the eliminating the need of human instructors. Taking
gap between current and ideal way of performing the inspiration from the psychology of transfer of learning,
specific yoga asana, thus enabling correction in the yoga Transfer Learning is a section of Machine Learning that
pose. This process is to be completed in real-time and needs focuses on gaining knowledge from one problem, and using
to be interactive necessitating the usage of react.js library. the knowledge in another similar problem. This has proven
The system analyzes the difference between the actual and to be a significant milestone in Reinforcement Learning. In
ideal yoga pose and landmarks the image of human body Yoga, since we don’t have to deal with motion, Yoga poses
while performing the yoga pose with required correction. have been proven to be an excellent example of image
classification. There are different pre-trained models that
Keywords – MoveNet, Transfer learning, Yoga Posture can be used for Image Classification based on Transfer
Detection, Tensorflow Object Detection API, Deep Neural Learning. The first approach used in this project was
Network ResNet34. Then other approaches such as VGG,
EfficientNet, Image Net were also explored. Although each
1. INTRODUCTION of these approaches had their individual benefits, they also
came with their own drawbacks. After evaluation and
The history of yoga is spanning thousands of years and comparison of different pre-trained models, MoveNet was
evolving through various cultural, religious, and finalized and used in the project. Similarly we have
philosophical contexts. Archaeological evidence suggests prepared and finalized dataset. After exploring existing
that yoga practices may have existed as far back as the datasets of images, and benefits of the poses, we prepared a
Indus Valley Civilization (around 3300–1300 BCE). It is custom data set consisting of following ten poses:
one of the most appreciated scientific contribution of India Bhujangasana, Garudasana, Halasana, Natarajasana,
to world. Sukhasana, Makarasana, Savasana, Simhasana, Tadasana.
We intend to broaden the data set in the near future. By
The global yoga industry has experienced significant integrating following features, the application aims to be a
growth in recent years, driven by increasing awareness of catalyst for positive environmental, social, and economic
health-wellness and popularity of mind-body practices. change, empowering individuals to embrace sustainable
According to market research, the yoga market was valued living practices in their everyday lives.
at over $37 billion in 2020 and is projected to continue

978-92-61-39091-4/CFP2268P @ITU 2024 – 313 – Kaleidoscope