Validation, Methods & System Performance
DITECT HAS-U2

Inter-system differences in modelling approaches for shoulder joint kinematics between markerless and marker-based motion capture during baseball pitching

Sports Biomechanics
•
2026
•
Original research
Theia3D version:
v2024.1.24
More Articles

Study at a glance

In six collegiate baseball pitchers, Theia3D's shoulder joint angles agreed closely with marker-based motion capture at the instant of maximum external rotation, but inter-system differences grew substantially by ball release, with external rotation showing the largest and most variable discrepancy between systems.

Summary

Disclaimer: The following summary details independent academic research. Theia3D is intended for research, sports performance and educational use only. It is not cleared as a medical device and is not intended to be used for the diagnosis, treatment, mitigation or prevention of any disease, injury or medical condition.

Editor's note: Theia3D version v2024.1.24 (Theia3D Apollo), as stated in the paper.

Study Overview

Design

A within-subject, concurrent comparison of shoulder joint kinematics measured simultaneously by Theia3D markerless motion capture and a marker-based optical motion capture system during fastball pitching, focused on the instants of maximum external rotation and ball release as well as the continuous arm-acceleration phase between them.

Participants

Six male university baseball pitchers (mean age 20.8 years; height 174.7 cm; weight 77.8 kg), right-handed, with 117 pitching trials included in the final analysis after excluding trials affected by marker dropout, marker displacement, or tracking failures.

Data Collection

Participants threw fastballs at approximately 80 percent of their maximum ball velocity (mean 35.2 m/s across trials) toward a net at regulation pitching distance. Markerless data were captured using eight synchronized Theia3D (HAS-U2, DITECT Corporation) cameras at 400 Hz, down-sampled to 360 Hz for analysis; marker-based data were captured using eight infrared cameras at 360 Hz, with 47 reflective markers placed on anatomical landmarks. Both systems shared a unified global coordinate system and used the same trunk and arm coordinate-system definitions.

Analysis

Shoulder joint angles (external rotation, horizontal abduction/adduction, and elevation) and joint centre positions were compared at maximum external rotation and ball release using intraclass correlation coefficients (ICC 3,1), and across the 200-frame arm-acceleration window using statistical parametric mapping (SPM). A nested Bland-Altman approach, accounting for multiple trials per participant, separated between-participant and within-participant sources of variance in the limits of agreement.

Key Findings

  • Agreement at maximum external rotation was good to excellent for all three shoulder angles. External rotation showed a negligible mean difference between systems (bias of -0.3 degrees, ICC of 0.77), while horizontal adduction and elevation showed excellent and good reliability respectively (ICC of 0.92 and 0.89), despite horizontal adduction carrying a larger fixed offset of -19.1 degrees between systems.
  • Agreement declined substantially by ball release, most notably for external rotation. External rotation showed poor reliability at ball release (ICC of 0.33) with an increased bias of -8.5 degrees, while horizontal adduction retained excellent reliability (ICC of 0.93) and elevation showed moderate reliability (ICC of 0.61).
  • Statistical mapping confirmed that the two systems produced significantly different shoulder angle trajectories throughout the arm-acceleration phase, for external rotation, horizontal abduction/adduction, and elevation alike, indicating the disagreement observed at ball release reflects a sustained divergence rather than an isolated endpoint difference.
  • Differences in joint centre position, rather than a single fixed error source, appear to drive much of the inter-system disagreement. The markerless system positioned the shoulder and wrist joint centres more anteriorly than the marker-based system at maximum external rotation, while the elbow joint centre showed the opposite pattern; along the vertical axis, the markerless system positioned every joint centre systematically lower than the marker-based system at both events.
  • The two systems showed essentially opposite patterns of segment-length consistency. The marker-based system's upper-arm and forearm segment lengths stayed consistent between trials but elongated within a single trial, a pattern consistent with soft-tissue motion under the forces generated during arm acceleration, while the markerless system's segment lengths stayed constant within a trial but varied between trials, consistent with trial-to-trial differences in markerless pose estimation.

What This Means for Movement Science Researchers and Sports Performance Teams

For teams using or considering Theia3D for overhead-throwing analysis, this study's central message is one of phase-specific caution rather than a blanket verdict: agreement with marker-based data was strong near maximum external rotation but degraded meaningfully by ball release, particularly for external rotation itself, a measure commonly used in assessing throwing mechanics. The authors also found that the markerless system's horizontal abduction waveform did not reproduce a late-phase decrease seen in the marker-based data, which previous work has linked to excessive horizontal abduction as a parameter of interest in throwing-mechanics assessment, a difference worth understanding specifically before drawing conclusions from markerless data about horizontal-plane shoulder motion late in the pitching motion. More broadly, the study is a useful reminder that apparent inter-system disagreement in fast, complex movements can often be traced back to specific, interpretable differences in joint centre estimation and modelling approach, rather than treated as an undifferentiated accuracy gap.

The Role of Theia3D in Movement Analysis

Theia3D was directly evaluated throughout this study, generating shoulder joint angles and upper-extremity joint centre positions from synchronized video during high-velocity baseball pitching, compared continuously against a marker-based reference across the full arm-acceleration phase from maximum external rotation to ball release. This detailed, event- and phase-specific comparison let the researchers identify not just how much the two systems disagreed, but when during the pitching motion and in which joint centres that disagreement arose.

Study Limitations and Future Directions

The authors note that only one shoulder joint centre definition was used for the marker-based reference system, and that different definitions reported elsewhere in the literature could influence the specific kinematic values obtained. The marker-based joint centre positions were calculated as simple midpoints between markers without constraining segment lengths to remain consistent across frames, an approach susceptible to soft-tissue motion artefact that a more constrained modelling approach might reduce. Pitching trials were performed at approximately 80 percent of each participant's maximum ball velocity rather than full-intensity pitching, so the findings may not fully generalize to maximal-effort throwing. The small sample of six participants limited the statistical power of the reliability analysis, producing wide confidence intervals that warrant caution in interpreting the specific ICC values reported, and it remains unclear whether similar patterns would hold in athletes with different anthropometric profiles. Because the joint centre estimation algorithm used by the markerless system is proprietary and not publicly documented, the authors could not attribute the observed inter-system differences to specific components of the markerless pipeline, and they recommend future work validate markerless joint centre positions against a biplane fluoroscopy reference to independently assess accuracy.

Citation: Uchida, T., Matsuo, T., Nakamura, T., Nishizawa, I., Sakata, J., Matsuzawa, K., Suzuki, T., & Sakai, T. (2026). Inter-system differences in modelling approaches for shoulder joint kinematics between markerless and marker-based motion capture during baseball pitching. Sports Biomechanics. https://doi.org/10.1080/14763141.2026.2711093

Link: Read the full study

How Well Does Markerless Capture Track the Shoulder Through a Pitch?

This study offers a detailed, phase-by-phase look at where Theia3D's shoulder kinematics during baseball pitching agree closely with marker-based motion capture, and where joint-centre modelling differences cause the two systems to diverge. If your work involves markerless analysis of overhead-throwing mechanics and you want to understand these phase-specific agreement patterns for your own application, talk to our team about what this study could mean for your throwing-biomechanics research.

Recent Posts