NORMA eResearch @NCI Library

Real-Time Yoga Pose Detection using Machine Learning Algorithm

Sunney, Jothika (2022) Real-Time Yoga Pose Detection using Machine Learning Algorithm. Masters thesis, Dublin, National College of Ireland.

[thumbnail of Master of Science]
PDF (Master of Science)
Download (6MB) | Preview
[thumbnail of Configuration manual]
PDF (Configuration manual)
Download (1MB) | Preview


Yoga is an ancient art that provides physical and mental fitness. Yoga incorporates self-learning, but incorrect postures can cause serious muscle and ligament damage. During Covid-19, the importance of self-learning yoga practices has increased, and many people include yoga as part of their routines. A yoga pose detection system based on human pose estimation techniques and Machine Learning can assist people in practicing yoga correctly by themselves. The major challenge with current yoga pose detection methods is that most of them are computationally expensive and unsuitable for real-time applications. This research proposes a computationally inexpensive approach for real time yoga pose detection by combining the Mediapipe Framework and Classification algorithms. An artificial intelligence-based system was built based on Mediapipe’s Blazepose model and XgBoost Classifier to predict yoga postures in real-time. A publically available dataset of Five Yoga poses was analyzed in this study (down-dog pose, goddess pose, tree pose, plank pose, and warrior pose). In this Research, 3D landmark features in x,y,z directions were extracted from the dataset using the Blazepose model and then classified by four machine learning classifiers - Random Forest, Support Vector Machine, XgBoost, Decision Tree and two neural network classifiers - LSTM (Long Short Term Memory) and 1D CNN (Convolutional Neural Network). All models were evaluated based on performance metrics. In order to detect yoga poses in real-time, XgBoost classifier was determined to be the optimum model, with an accuracy of 95.14%, precision of 95.36%, Recall of 95.02% and F1 Score of 95.17%. The proposed model is computationally efficient with an optimum latency of 8ms and size of 513KB. The Framework has the potential to be integrated into mobile application which can be used by yoga practitioners to perform yoga in the comfort of their homes.

Item Type: Thesis (Masters)
Subjects: Q Science > QA Mathematics > Electronic computers. Computer science
T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science
Q Science > QP Physiology
Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions: School of Computing > Master of Science in Data Analytics
Depositing User: Tamara Malone
Date Deposited: 13 Mar 2023 15:43
Last Modified: 13 Mar 2023 15:43

Actions (login required)

View Item View Item