NORMA eResearch @NCI Library

A Deep Learning Framework to Classify Yoga Poses Hierarchically

Rakate, Pooja (2021) A Deep Learning Framework to Classify Yoga Poses Hierarchically. Masters thesis, Dublin, National College of Ireland.

[img]
Preview
PDF (Master of Science)
Download (775kB) | Preview
[img]
Preview
PDF (Configuration manual)
Download (1MB) | Preview

Abstract

Yoga is beneficial for both the physical and mental well-being of individuals of all ages. Currently, machine learning and deep learning techniques are employed to perform flat classification of precisely 12 yoga poses using the human pose estimation (HPE) model ‘OpenPose’ (OP) followed by CNN. However, building a deep learning framework considering hierarchical postures details and increasing the scope by including many yoga poses is a challenging region of research. Thus, this study aims to improve the hierarchical yoga pose classification performance by three methods. Firstly, by introducing the OP model in the deep learning framework. Secondly, by adding a connectivity pattern to build the DenseNet201 Hierarchical-concat architecture, and finally, it proposes to utilize the VGG19 network to classify a total of 82 yoga poses hierarchically. The performance of all these models was compared based on top-1, top-3, and top-5 accuracy rates. The said objective was achieved by DenseNet201 Hierarchical-concat proposed as the second approach. It was found to perform the best amongst all giving enhanced results than the state-of-the-art with 73%, 63.54%, and 52% top-1 accuracy on hierarchical levels 1, 2, and 3 respectively. While other models were found to be not performing satisfactorily. Passage of lower-level features directly to a higher level was driven by the connectivity pattern. This approach demonstrated a significant increase in the classification performance on all levels which would be able to classify a given pose well on the basic posture levels 1 and 2. Thus, with the success of this study, the scope of the yoga pose recognition system was broadened by considering a total of 82 poses. This facilitates a yoga pose correction system that can correct a wide variety of poses. However, there is a scope to obtain better results with all the models by fine-tuning and training them on the entire dataset.

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 > QA Mathematics > Computer software
T Technology > T Technology (General) > Information Technology > Computer software

G Geography. Anthropology. Recreation > GV Recreation Leisure > Sports
Divisions: School of Computing > Master of Science in Data Analytics
Depositing User: Clara Chan
Date Deposited: 14 Dec 2021 11:40
Last Modified: 14 Dec 2021 11:40
URI: http://norma.ncirl.ie/id/eprint/5214

Actions (login required)

View Item View Item