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

OBJECTIVES

This study examined how camera geometry influences the agreement of two-camera markerless joint-angle estimates with an eight-camera markerless reference system.

METHODS

Twenty-three female soccer athletes completed pre-season Y-balance, L-hop, countermovement jump, broad jump, and bodyweight squat tasks. Movements were recorded using an eight-camera Theia3D markerless system as the reference. All 28 two-camera combinations were processed using a custom Pose2Sim-based pipeline to estimate lower-limb joint angles. Camera geometry was quantified from calibration data, including camera height, distance to origin, azimuth, field of view, and horizontal angular separation. Agreement with the reference system and consistency among two-camera configurations were assessed using root mean square error (RMSE), mean absolute error (MAE), bias, limits of agreement, and coefficient of multiple correlation (CMC).

RESULTS

Agreement varied by camera geometry, anatomical plane, and task. Front and Back configurations, with moderate horizontal angular separation values of 68.59 ± 17.68° and 80.95 ± 22.35°, generally showed more favorable agreement, particularly for sagittal-plane hip and knee kinematics (RMSE < 8°, CMC ≈ 1.00). Same Quadrant configurations had the smallest angular separation (28.24 ± 3.81°) and poorer agreement, especially for frontal- and transverse-plane angles (RMSE > 10°, CMC < 0.60). Diagonal configurations had the largest angular separation (160.56 ± 13.40°) but did not consistently improve agreement. Y-balance and L-hop showed reduced agreement and higher dropout due to self-occlusion and participants moving outside the field of view.

CONCLUSION

Camera geometry influenced two-camera markerless motion capture agreement. Front and Back configurations with moderate angular separation were most suitable for sagittal-dominant bilateral tasks, whereas Same Quadrant configurations should be avoided. For Y-balance and L-hop, camera placement should prioritize foot/ankle visibility and full movement-path coverage rather than angular separation alone.