🧘‍♀️ Yoga Pose Detection Using GANs and Real-Time Pose Estimation

📌 Introduction

This project implements an advanced Yoga Pose Detection system using a combination of Generative Adversarial Networks (GANs) for data augmentation, state-of-the-art pose estimation models, and traditional machine learning classifiers for pose classification. The system supports both image-based input and real-time webcam feed for live yoga pose detection and accuracy evaluation.

🎯 Objective

The primary objective of this project is to develop an intelligent system that can:

• Recognize and classify various yoga poses from static or real-time images.
• Evaluate the correctness and accuracy of the detected pose.
• Provide feedback to users practicing yoga at home or in training environments.

🧠 Key Components

1. Input Image / Real-Time Webcam Feed

• Users can input images or use webcam integration for real-time yoga pose detection.

2. GAN Augmentation

• Utilizes GANs to generate diverse augmented images to enrich the dataset.
• Helps to overcome dataset limitations and improve model generalization.

3. Key Points Detection

• Extracts body keypoints using advanced models like:
  • OpenPose
  • PoseNet
  • PIFPAF

4. Pose Estimation

• Predicts 2D and 3D human poses using:
  • YOLOv8 for 2D pose estimation
  • SSD (Single Shot Detector) for 3D pose estimation

5. Pose Classification

• Pose classification is done using machine learning models:
  • Support Vector Machine (SVM)
  • K-Nearest Neighbors (KNN)
  • Naïve Bayes
  • Logistic Regression
  • Random Forest

6. Evaluation Metrics

• Models are evaluated using:
  • Accuracy
  • Precision
  • Recall
  • F1-Score

7. Pose Accuracy Feedback

• Real-time feedback on how accurately a user is performing a yoga pose.
• Displays:
  • Pose names
  • Detected confidence levels
  • Visual feedback

🖥️ Workflow Overview

1. Input image or webcam frame is processed.
2. Data augmentation is applied using GANs.
3. Keypoints are detected and passed to pose estimation models.
4. Classification using ML models.
5. Real-time feedback and pose accuracy are displayed.

📷 Sample Output

Real-time detection displays:

• Yoga pose name
• Pose confidence score
• Visual skeleton overlay

🔍 Tools and Technologies Used

• Python
• OpenPose / PoseNet / PIFPAF
• YOLOv8, SSD
• GANs for augmentation
• scikit-learn (SVM, KNN, Naïve Bayes, etc.)
• OpenCV for webcam integration
• Matplotlib / Plotly for 3D pose visualization

📊 Future Scope

• Incorporation of feedback loop for posture correction.
• Integration with mobile apps.
• Support for more yoga poses and multi-person tracking.