Concurrent Validity And Reliability Of A Markerless Motion Capture System In Estimating Joint Kinematics In The Tennis Serve

PURPOSE: Marker-based (MB) motion capture systems are the standard in biomechanics, but markerless (ML) systems, such as Theia3D, offer accessible and time-efficient alternatives with potential for practical use in dynamic sports analysis. The purpose of this study was to assess the validity and reliability of a ML system in capturing both continuous joint angular velocity profiles and peak angular velocities during the tennis serve. METHODS: Nine collegiate tennis players performed five serves each, with joint kinematics recorded simultaneously by both MB (Qualysis) and ML (Theia3D) systems. Time-series data for the left knee, right knee, pelvis, trunk, and shoulder were normalized from the trophy position to foot strike, comparing ML data with MB data using Statistical Parametric Mapping (SPM), Concordance Correlation Coefficients (CCC), and Bland-Altman plots. RESULTS: SPM{t} analysis identified significant differences in angular velocities for all joints and segments at different phases (p < 0.05); however, these differences were limited, with significance at peak noted only in trunk rotation (p < 0.001). CCC for peak trunk rotation showed moderate agreement (CCC = 0.539, 95% CI [0.407, 0.650]), with Bland-Altman analysis indicating a mean bias of -100.7 deg/s. SPM{t} analysis did not identify the peak differences for all other joints and segments as significant (p < 0.05). These findings indicate a systematic and consistent offset, particularly in peak values, suggesting that while differences exist, the ML system closely follows the same movement patterns as the MB system and may be viable for practical applications where a predictable bias can be adjusted for. CONCLUSIONS: The ML system effectively captured both continuous joint velocity profiles and peak values with a high degree of precision and accuracy, demonstrating potential as a practical tool in sports biomechanics. The limited and consistent nature of observed differences further supports the ML system as a reliable, efficient alternative for dynamic assessments in tennis and similar sports settings.