Lighting and clothing effects on Theia3D markerless joint-angle waveforms during walking

Markerless motion capture may enable three-dimensional gait analysis outside dedicated laboratory settings, but the resulting kinematics may depend on acquisition conditions such as lighting and clothing. This study investigated pose-estimation feasibility, waveform consistency, and absolute waveform differences in Theia3D-derived joint-angle waveforms under systematically varied lighting-clothing combinations in an indoor laboratory environment. Twenty healthy adults performed overground walking trials under 15 acquisition conditions combining five lighting and three clothing conditions. Video recordings from eight cameras (1,920 × 1,080 pixels, 85 Hz) were analyzed using Theia3D to derive lower- and upper-body joint-angle waveforms in the sagittal, frontal, and transverse planes. Waveform consistency across acquisition conditions was quantified using the intraclass correlation coefficient (ICC(C,1)), and absolute waveform differences were quantified using root mean square difference (RMSD). Joint-angle waveforms were generated for all participants in 11 of the 15 acquisition conditions, with reduced feasibility confined to extreme acquisition conditions. Across retained conditions, waveform consistency and absolute differences were joint- and plane-specific. Sagittal-plane lower-body joint angles showed high waveform consistency, with ICC(C,1) values of 0.983-0.998 and mean pairwise RMSDs of 1.24°-1.80°, whereas transverse-plane angles showed more variable consistency, with ICC(C,1) values of 0.505-0.986 and mean pairwise RMSDs of 1.37°-5.45°, particularly for the shoulder and elbow. These findings show that acquisition-condition effects on Theia3D-derived joint-angle waveforms are joint- and plane-specific and that feasibility is maintained across most of the tested lighting-clothing combinations. Clothing and lighting conditions should therefore be kept as consistent as practicable across comparative and repeated measurements.