Yoga Posture Recognition and Quantitative Evaluation with Wearable Sensors Based on Two-Stage Classifier and Prior...

Area of Science:

• Biomechanics and Human Movement Analysis
• Artificial Intelligence in Sports Science
• Wearable Technology for Health and Fitness

Background:

• Self-taught yoga practitioners often lack standardized performance feedback.
• Imprecise yoga form can limit benefits and increase injury risk.
• Objective quantitative evaluation of yoga postures is needed for effective learning.

Purpose of the Study:

• To develop a full-body posture modeling and quantitative evaluation method for yoga.
• To recognize and assess yoga postures, offering personalized learner guidance.
• To improve the accuracy and standardization of yoga practice through technology.

Main Methods:

• Utilized Back Propagation Artificial Neural Network (BP-ANN) for initial posture classification.
• Employed Fuzzy C-Means (FCM) for refined classification within categories.
• Constructed a Bayesian network using multidimensional Gaussian variables for quantitative evaluation of body part alignment.

Main Results:

• Achieved 95.39% accuracy in yoga posture recognition on a dataset of 211,643 frames.
• Effectively distinguished standard from non-standard body parts using Bayesian network probabilities.
• Demonstrated significant reduction in joint angle errors when guidance was provided by the system.

Conclusions:

• The proposed method accurately recognizes and quantitatively evaluates yoga postures.
• The system provides actionable, language-based feedback for posture correction.
• This technology offers a promising tool for enhancing yoga training and self-practice.