AI-Powered 3D Markerless Motion Tracking

The rapid advancements in computer vision and human pose estimation algorithms have facilitated the transition to using ordinary video cameras for analyzing human movement kinematics. Leveraging the widespread availability of smartphone cameras, I developed a video-based markerless motion capture application that is significantly more cost-effective than traditional optical marker-based or inertial motion capture systems. This application is scalable to accommodate any number of smartphone cameras, with a minimum requirement of two cameras. In this setup, six affordable smartphone cameras were calibrated using a checkerboard pattern with known cell dimensions to determine their intrinsic and extrinsic parameters, enabling the calculation of projection matrices to map image points to their corresponding three-dimensional coordinates. The system captures human body movements simultaneously across multiple camera views and feeds to a fast and efficient pose estimation algorithm to extract the two-dimensional human joint centers in each camera view. The three-dimensional joint coordinates are then reconstructed using the calculated projection matrices. These coordinates are subsequently utilized to perform detailed kinematic analyses of human movements.