Influence of camera geometry on 3D joint angle estimation in markerless motion capture

Area of Science:

• Biomechanics
• Motion Capture Technology
• Sports Science

Background:

• Markerless motion capture offers a less intrusive method for analyzing human movement compared to marker-based systems.
• Understanding the influence of camera configuration is crucial for optimizing the accuracy of markerless systems.

Purpose of the Study:

• To investigate how different camera geometries affect the accuracy of lower-limb joint angle estimations using a two-camera markerless system.
• To compare the agreement of two-camera setups with a multi-camera reference system across various athletic tasks.

Main Methods:

• Twenty-three female soccer players performed dynamic tasks (Y-balance, hops, jumps, squats).
• Lower-limb joint angles were captured using an eight-camera markerless system as reference and 28 two-camera combinations via a Pose2Sim pipeline.
• Camera geometry parameters (height, distance, field of view, angular separation) were analyzed.

Main Results:

• Agreement varied significantly with camera geometry, anatomical plane, and task.
• Front and Back camera configurations (moderate angular separation) showed better agreement for sagittal plane kinematics (hip/knee).
• Same Quadrant configurations (small angular separation) yielded poorer results, especially for frontal/transverse planes; Y-balance and L-hop tasks suffered from self-occlusion and field-of-view limitations.

Conclusions:

• Camera geometry is a key determinant of two-camera markerless motion capture accuracy.
• Moderate angular separation (Front/Back configurations) is optimal for sagittal-dominant tasks, while Same Quadrant setups are not recommended.
• For complex tasks like Y-balance and L-hop, prioritizing full movement coverage and visibility of extremities is more critical than angular separation alone.