Yoga Pose Estimation and Feedback Generation Using Deep Learning

Area of Science:

• Integrative medicine
• Computer science
• Artificial intelligence

Background:

• Yoga, originating in ancient India, is a 5000-year-old practice for mind-body balance.
• Modern lifestyles increase stress, driving global interest in yoga for well-being.
• Self-learning yoga is popular but risks incorrect postures, potentially causing harm.

Purpose of the Study:

• To develop deep learning techniques for accurate yoga posture detection.
• To provide users with real-time feedback on pose correctness.
• To enhance the safety and effectiveness of self-taught yoga practice.

Main Methods:

• Utilized deep learning models to analyze user-submitted yoga practice videos.
• Developed algorithms to detect abnormal joint angles compared to correct poses.
• Implemented a system to provide specific feedback on pose correction.

Main Results:

• The proposed deep learning method achieved a high accuracy of 0.9958 in detecting incorrect yoga postures.
• The system demonstrated superior accuracy compared to existing state-of-the-art methods.
• The approach requires less computational complexity than comparable techniques.

Conclusions:

• Deep learning offers an effective solution for identifying and correcting yoga posture errors.
• This technology can significantly improve the safety of independent yoga practice.
• The system provides valuable guidance for users to refine their yoga poses accurately.