Study at a glance
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 2023.1.0.3160, as stated in the paper.
Study Overview
Design
A concurrent, three-system comparison study evaluating the accuracy of an in-stadium markerless system (Hawk-Eye) and a portable, lab-based markerless system (Theia3D) for baseball pitching kinematics and kinetics, using simultaneously recorded marker-based motion capture as the reference standard.
Participants
Eighteen NCAA Division I and Division II collegiate male baseball pitchers (height 1.83 m; mass 86 kg) who provided informed consent to participate.
Data Collection
After a self-supervised warm-up, each pitcher threw 10 maximum-effort fastballs off a Major League Baseball stadium mound while three systems recorded simultaneously: a 5-camera in-stadium Hawk-Eye markerless system permanently mounted around the field, and an 18-camera portable hybrid system combining 10 markerless Qualisys Miqus cameras and 8 marker-based Qualisys Arqus cameras positioned on tripods within a 4-meter radius of the mound, with reflective markers placed on each pitcher following a standard marker set. All systems recorded at 300 Hz.
Analysis
Video data from the in-stadium system were processed using Hawk-Eye's own pose-estimation software, and video data from the portable markerless cameras were processed through Theia3D, with both sets of output aligned to the marker-based reference system's coordinate frame. Agreement was assessed using mean per-joint position error, statistical parametric mapping of kinematic waveforms, root mean square error, Bland-Altman analysis, and concordance correlation coefficients, across both joint kinematic measures (joint angles and positions) and kinetic measures (joint torques and forces) relevant to pitching performance.
Key Findings
- Both markerless systems showed measurable discrepancies from marker-based motion capture, with mean per-joint position error of 56.6 mm for the in-stadium system and 52.0 mm for the portable Theia3D system, and errors were consistently smaller at lower-body joints (hip, knee, ankle) than at upper-body joints (shoulder, elbow, wrist) for both systems.
- Stride length showed the strongest agreement with marker-based motion capture for both systems (concordance correlation above 0.85), while shoulder rotational measures showed the greatest variability and weakest agreement for both systems.
- The portable Theia3D system showed smaller kinematic waveform errors than the in-stadium system for knee flexion, pelvis rotation, trunk rotation, pelvis rotational velocity, and shoulder rotational velocity, while the in-stadium system performed comparably for elbow flexion and shoulder internal-external rotation and showed lower error for trunk rotational velocity specifically.
- For kinetic measures, both systems underestimated peak shoulder distraction force by as much as 38 newtons, with moderate agreement overall; shoulder external rotation torque and elbow valgus torque showed considerably weaker agreement for both systems, marking this as the first study to evaluate markerless-derived pitching kinetics, not just kinematics, against marker-based motion capture.
- The authors explicitly frame these results as reflecting each system's distinct operational design, portable and flexible camera placement versus fixed and distant in-stadium placement, rather than establishing one markerless system as categorically more accurate than the other.
What This Means for Movement Science Researchers and Baseball Performance Assessment Teams
For researchers and performance staff working in baseball, this study offers a specific, practically important comparison between two genuinely different ways of deploying markerless motion capture: an always-available, in-stadium system built for in-game tracking at scale, and a portable, purpose-configured lab system offering more flexible camera placement. The portable Theia3D system showed somewhat smaller average errors across many measures, consistent with the added flexibility of closer, adjustable camera placement, while the in-stadium system still performed comparably for several specific, practically relevant measures despite its fixed, more distant camera configuration, highlighting that in-stadium systems can offer meaningful ecological validity for certain measures even without the setup flexibility of a dedicated lab system. This distinction is a useful, practical consideration for teams and researchers deciding between the convenience of in-game data capture and the accuracy profile of a dedicated, portable setup for a given application, including when kinetic measures like shoulder or elbow torque are of specific interest, given both systems' comparatively weaker agreement in those variables.
The Role of Theia3D in Movement Analysis
Theia3D was one of two markerless motion capture systems directly evaluated against marker-based motion capture in this study, specifically as the portable, lab-configured alternative to an in-stadium markerless system, during the fast, complex, whole-body movement of baseball pitching. The study's direct, concurrent comparison of two markerless systems' designs against the same marker-based reference offers a distinct and practically valuable contribution to markerless validation research, extending beyond the more typical single-system-versus-marker-based comparison to show how specific system design choices, not just the underlying markerless approach in general, shape measurement accuracy.
Study Limitations and Future Directions
The authors note that a subset of pitches was excluded due to incomplete marker or image data across the marker-based and portable configurations, which may have introduced some selection bias toward more stable tracking conditions, though no apparent differences in pitch velocity were observed between retained and excluded trials. They note that marker-based motion capture itself, while serving as the reference standard here, is subject to known error from soft-tissue artifact, particularly during highly dynamic movements like pitching, and remains the most practical available standard in the absence of a true gold-standard method such as bi-planar fluoroscopy. The authors also note that pose-estimation accuracy for any markerless system depends heavily on the quality of keypoint detection, which can be influenced by clothing, lighting conditions, and how well test participants resemble the people represented in each system's underlying training data, and that greater camera-to-subject distance, as used by the in-stadium system, could introduce additional challenges that high-resolution, zoomed camera settings may or may not fully offset. Because each system was evaluated in its standard operational configuration, the authors note that observed differences reflect the combined effects of camera hardware, camera count, and pose-estimation algorithm rather than any single isolated factor. They identify systematically varying camera configuration and standardizing image data across markerless systems to isolate their individual effects, and extending this kind of validation to other baseball-specific tasks such as hitting and fielding, as well as across other system configurations and competitive levels, as important next steps.
Citation: Aguinaldo, A. L., Cardinale, T., La Salle, D. T., Escamilla, R. F., & Buffi, J. H. (2025). Assessing the accuracy of in-stadium and portable multi-camera markerless motion capture for baseball pitching kinematics and kinetics. Journal of Sports Sciences. Advance online publication. https://doi.org/10.1080/02640414.2025.2595411
Link: Read the full study
How Does Portable Markerless Capture Compare to In-Stadium Tracking for Baseball Pitching?
This study of 18 collegiate pitchers compares Theia3D, a portable, lab-based markerless system, against an in-stadium markerless system, finding that both show measurable discrepancies from marker-based motion capture during pitching, with Theia3D producing somewhat smaller average errors overall while each system has its own specific strengths. If your work involves baseball or overhead-throwing biomechanics, talk to our team about applying Theia3D to your own research.