A Kinematic Consistency-Based Method for Refining Axial Twist Errors in Skeleton-Based Motion Data

Optical motion capture is vital for the digitization of human movement, yet marker occlusion often causes unnatural axial twist errors in limb segments. Current refinement methods primarily target marker trajectories or general smoothing, leaving skeleton-based axial rotations inadequately addressed when raw marker data are absent. This study proposes a kinematic framework to automatically detect and refine these twist errors by analyzing the kinematic relationship between parent and child joints. We utilize swing-twist decomposition to isolate the parent joint’s axial rotation and estimate the azimuth rotation to identify kinematic contradictions, such as counter-rotating joint behavior. Validation using synthetic level-shift errors injected into the CMU Motion Capture Database demonstrated a Frame Success Rate (FSR) of 94.81%. The proposed method significantly reduced the Mean Absolute Error (MAE) from 124.61° to 8.91°, effectively removing 92.85% of injected errors while preserving the global orientation of the motion. This approach ensures the kinematic integrity of skeleton animation, providing a robust solution for high-precision motion analysis tasks such as motion retargeting.