Tools
Tools
Ghongane, Aishwarya (2022) Hierarchical Classification of Yoga Poses using Deep Learning Techniques. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (2MB) | Preview |
Preview |
PDF (Configuration manual)
Download (11MB) | Preview |
Abstract
Yoga has grown incredibly in recent years due to the physical, mental, and spiritual benefits it provides to those who practice it. Many of these yoga poses entail performing complex body postures. However, failing to perform these poses with great care and attention may prove ineffectual to one’s health. To avoid unfortunate incidents raises the need to develop a system that will correctly classify the yoga poses and ultimately assist the Yogis (yoga practitioners) in understanding the difference between yoga poses to avoid injuries. Existing research to develop such a system has employed various Machine Learning algorithms and Deep Learning techniques. However, the typical approach followed by them is that the classification of the yoga poses is based on the final posture attained by the Yogi and not the intermediate steps. The limitation of this approach is the misclassification of yoga poses due to similar intermediate steps. Therefore, this report proposes a novel deep learning framework to classify the yoga poses hierarchically using the knowledge of intermediate steps involved in the yoga pose. To achieve the aim of this research, two frameworks are broadly designed and tested using the Yoga-82 dataset. Firstly, the state-of-the-art model using DenseNet-201 architecture was implemented to form a baseline for comparative study. Secondly, a deep learning framework using a modified ResNet-50 architecture was used. The modifications include classifying the yoga poses at three levels: coarse level 1, coarse level 2, and fine level. The data augmentation techniques were also used to analyze the performance of the DenseNet-201 and ResNet-50 classifiers. The model performance is evaluated using top-1, top-3, and top-5 accuracy metrics. Amongst all the approaches that were followed, the modified DenseNet-201 architecture along with image augmentation proved to perform better for hierarchical classification. In addition, the use of data augmentation techniques improved the model performance by 7%-8%.
| Item Type: | Thesis (Masters) |
|---|---|
| Subjects: | Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science R Medicine > RC Internal medicine > RC1200 Sports Medicine 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: | 24 Jan 2023 17:46 |
| Last Modified: | 03 Mar 2023 12:03 |
| URI: | https://norma.ncirl.ie/id/eprint/6127 |
Actions (login required)
 |
View Item |