Deep Learning-Based Knee Joint Analysis While Performing Adho Mukha Svanasana to Utkatasana

This project investigates a deep learning-based system designed to analyze knee joint movements during the transition from Adho Mukha Svanasana (Downward-Facing Dog) to Utkatasana (Chair Pose) in yoga. The study aimed to evaluate the feasibility and accuracy of a single-camera setup for motion analysis, offering a cost-effective alternative to traditional motion capture systems.

The proposed system employs BlazePose, a pose estimation algorithm, to track knee movement from video footage. Ground truth data was obtained using Kinovea software for comparison. The analysis focused on key biomechanical parameters: displacement, velocity, acceleration, and jerk. From the extracted knee coordinates, these parameters were calculated to assess the system's performance.

The results demonstrated the system's high accuracy, with mean squared errors (MSE) significantly low across all parameters: displacement (0.000306 pixels), velocity (0.000220 pixels/sec), acceleration (0.002897 pixels/sec²), and jerk (0.000103 pixels/sec³). These findings validate the precision of the BlazePose-based approach in tracking knee joint movements.

One of the key advantages of this system is its efficiency. By automating the analysis process, it substantially reduces the time required compared to manual methods, making it an attractive tool for researchers and practitioners. Moreover, the reliance on a single camera instead of costly motion capture setups enhances the accessibility of kinematic analysis for a broader audience, including yoga practitioners, physiotherapists, and researchers.

The study highlights the potential of integrating deep learning technologies with yoga practice analysis, offering a scalable and practical solution for biomechanical research. It opens new possibilities for exploring movement patterns and their implications on physical health and performance.