Published Paper Details:

Yoga is a popular practice that involves various body postures and breathing exercises to improve physical and mental health. In this project, we have developed a system that detects yoga poses in real-time using a webcam and machine learning algorithms. The system uses a total of eight yoga poses including Bhadrasan, Shavasan, Gomukhasan, Vajrasan, Sarvangasan, Shirsansan, Chakrasan, and Dhanurasan. The system works by first collecting a dataset of yoga poses with corresponding virtual coordinates for each pose. We manually annotated the poses with 501 virtual coordinates on the body and face of the yoga practitioner. The dataset was preprocessed by normalizing the coordinates and splitting them into training and testing sets. We used OpenCV and MediaPipe libraries to extract features from the virtual coordinates for each pose. Once the model was trained and validated, we used OpenCV and MediaPipe libraries to capture live video feed from the webcam and extract virtual coordinates from the video frames. The extracted coordinates were then passed through the trained model to predict the corresponding yoga pose. The system displays the name and accuracy of the predicted pose on the screen in real-time. In addition to pose detection, we implemented a user registration and login system using the tkinter library and an SQLite database. This system allowed users to register their information, which was stored in the database for future logins. The login system ensured secure access to the real-time pose detection system. The proposed system achieved an average accuracy of 95% in detecting the eight yoga poses in real-time. The system can be used by yoga practitioners to monitor and improve their practice, as well as by instructors to monitor their students and provide personalized feedback. Future work includes expanding the dataset to include more yoga poses and incorporating feedback mechanisms to improve the accuracy of the system.