Maintaining physical health through yoga is a widely acknowledged at-home exercise practice. However, successfully performing the 82 Yoga Asanas across multiple sessions can be challenging for many individuals. Finding a knowledgeable and affordable yoga instructor becomes a hurdle for those seeking guidance. This project addresses this challenge by leveraging Deep Learning (DL) techniques, modifying pre-trained models to detect yoga poses and classify them into different classes.
-
Methodology:
- Utilized two pre-trained Convolutional Neural Network (CNN) models.
- Employed ensemble modeling for accurate yoga pose detection.
-
Dataset:
- Comprised a total of 18,488 images, spanning 6 major yoga classes and 82 distinct poses.
-
Transfer Learning:
- Applied transfer learning to adapt pre-trained models for yoga pose detection.
-
Ensemble Model:
- Utilized an ensemble model to enhance the accuracy of yoga pose detection.
A detailed experiment was conducted, resulting in a 95% accuracy rate for yoga pose detection. The experiment focused on addressing the challenges associated with finding suitable yoga instructors by providing an accessible and automated solution for individuals.
For a comprehensive overview of the experiment results and performance metrics, please check the sample section.
- Yoga Poses
- Transfer Learning
- Posture Detection
- Ensemble Model
- Deep Learning