This project is a deep learning-based application that can classify common yoga poses from video input and provide real-time feedback to users on how to improve their form. The application utilizes computer vision techniques to track body landmarks and a convolutional neural network (CNN) model to classify the poses. The demo for this app can be found at : demo
- Pose Classification: The application can accurately classify three yoga poses: Tree Pose, Warrior Pose, and Downward-Facing Dog Pose, with an impressive 95% accuracy.
- Real-time Feedback: In addition to classifying the pose, the application provides real-time feedback to users on how to improve their form based on the detected body landmarks.
- User-friendly Interface: The application has a simple and intuitive user interface, making it easy for users to start the webcam and receive pose classification and feedback.
- Deep Learning: A CNN model was trained on a dataset of yoga pose images to achieve high classification accuracy.
- Computer Vision: The MediaPipe library was used for real-time body landmark detection and tracking.
- Python: The application was developed in Python, leveraging popular libraries such as OpenCV, TensorFlow, and NumPy.
- Clone the repository:
git clone https://github.com/nilaygaitonde/yoga-pose-classifier.git - Install the required dependencies:
pip install -r requirements.txt
- Run the application:
python capture.py - The application will open a window displaying the webcam feed.
- Perform one of the three supported yoga poses (Tree Pose, Warrior Pose, or Downward-Facing Dog Pose) in front of the webcam.
- The application will classify the pose and provide real-time feedback on how to improve your form based on the detected body landmarks.
- The yoga pose dataset used for training the CNN model was sourced from https://www.kaggle.com/datasets/nilaygaitonde/yoga-classification.
- The MediaPipe library for body landmark detection was developed by Google.