Yoga Pose Recognition (YPR) using ML-DL and android application

Abstract

The study aimed to create a Human Activity Recognition (HAR) model for Yoga Pose Recognition and Classification
using datasets gathered through smart sensor technologies and imaging and filming devices to read various human
actions, recognize various poses, analyze them, and then predict and classify the Yoga pose with minimum error.
Pre-recorded data was fed to the model for the initial run and thereafter the model would learn and re-learn
new inputs and outputs by supervised learning methods. A collection of data from cameras present in smart
smartphones and other devices were used to create a dynamic dataset of posture photos and videos to predict the
most feasible output and add the mapping in the dataset to recognize particular Yoga poses. Yoga is a methodical
way of attaining balance and harmony both inside oneself and outside the body. It has its roots in ancient India.
Its history spans millennia, with the word “yoga” being first used in the Rig Veda, an ancient Indian scripture,
which dates back to around 1500 BC. The Atharva Veda, which was written about 1200–1000 BC, places a strong
emphasis on breath regulation. Indus-Saraswati seals and fossils depicting yoga sadhana practitioners have also
been discovered. These artifacts date back to 2700 BC (10). Nowadays, yoga is performed by millions of people
worldwide. It provides mental and physical health advantages, such as lowering stress, anxiety, and depression, as
well as physical benefits like better flexibility, strength, and posture. Yoga has grown popular as more individuals
try to live healthier lives.
The study investigated various human postures and actions to predict the possible Yoga pose performed by that
particular human through ML/DL (Machine Learning and Deep Learning) approaches. The proposed system or
model that learned and evolved by obtaining new data and through supervised learning. We have used single-user
pose recognition to create personalized datasets. Our aim was to provide a self-instruction system that allows
people to learn and practice yoga correctly by themselves. This development laid the foundation for building
such a system by discussing various ML and DL approaches to accurately classify Yoga poses on pre-recorded
videos and photos.