Concurrent validity of a deep learning algorithm-based markerless motion capture system for biomechanical analysis

This study tested the concurrent validity of pose estimation from markerless motion capture against those collected using a marker-based technique. Kinematic data were recorded simultaneously while ten male participants walked over ground at their self-selected speed. Sagittal plane joint angles demonstrate agreement in pose estimation between methods, supported by mean ICC values between the markerless and marker-based data (ankle=94%, knee=98%, hip=96%).