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: This study is distinctive in directly comparing two different Theia3D software versions (2020.6.0.1106 and 2023.1.0) against marker-based motion capture within the same dataset, rather than evaluating a single version. Both versions are reported here as stated in the paper.
Study Overview
Design
A concurrent, within-subjects comparison study evaluating agreement between marker-based motion capture and two different versions of Theia3D markerless motion capture software for lower- and upper-extremity kinematics during a simulated basketball throwing motion.
Participants
Thirteen healthy adults (six women, mean age 29.6 years; seven men, mean age 30.7 years) with no previous experience of the throwing task required and no physical limitations affecting participation.
Data Collection
Each participant performed five simulated (ball-less) basketball throws, performed without an actual ball due to laboratory space and ceiling-height constraints. Marker-based data were captured using a 47-marker Cleveland marker set recorded by 12 Qualisys Oqus infrared cameras (120 Hz), while markerless data were captured simultaneously using 10 Qualisys Miqus video cameras (120 Hz) and processed separately through two different Theia3D software versions, 2020.6.0.1106 and 2023.1.0.
Analysis
Knee, hip, shoulder, elbow, and wrist joint angles in the sagittal plane were extracted from the point of maximum elbow flexion ("ball elevation") through maximum elbow extension plus ten frames ("ball release") and normalized to a common time base. Root mean square deviation (RMSD, with a threshold of 5 degrees or less considered a minimal difference) was calculated for three pairwise comparisons: marker-based versus the older Theia3D version, marker-based versus the newer Theia3D version, and the two Theia3D versions against each other. Intraclass correlation coefficients (ICC) were also calculated for maximum flexion angle at each joint across the three comparisons, with statistical outliers excluded using a two-standard-deviation threshold.
Key Findings
- The newer Theia3D version (2023.1.0) showed closer agreement with marker-based motion capture than the older version (2020.6.0.1106) for elbow flexion, hip flexion, and wrist flexion, while knee flexion was nearly identical between the two versions and shoulder flexion was broadly comparable, suggesting that software updates meaningfully improved agreement for some joints more than others.
- Agreement varied widely by joint, with RMSD values ranging from about 7 degrees at the knee to as high as nearly 27 degrees at the wrist, depending on the specific joint and Theia3D version compared.
- Markerless motion capture consistently produced smaller joint angles than marker-based motion capture across all participants and joints, which the authors suggest may partly reflect marker-based systems' own difficulty precisely capturing joint positions near full extension, rather than solely reflecting a markerless tracking limitation.
- Wrist motion showed particularly weak correlation between systems in both Theia3D versions (ICC near zero), and some individual trials, especially using the older Theia3D version, showed physically implausible wrist rotation values that did not match the recorded video and were excluded as clear tracking errors.
- The authors recommend against directly comparing absolute joint-angle values between markerless and marker-based systems, instead favoring the use of relative changes in kinematics, and emphasize that the specific software version used should always be documented given how much it influenced the results observed here.
What This Means for Movement Science Researchers and Throwing and Overhead-Sport Performance Teams
For researchers and performance staff using markerless motion capture to study throwing or other fast, multi-joint overhead movements, this study offers two specific, practically grounded findings. First, software version matters: the newer Theia3D version tested here showed meaningfully closer agreement with marker-based motion capture for several joints, a useful data point for teams deciding whether and when to update their own processing pipelines, and a reminder to document software version consistently across a research program. Second, the study offers a specific, joint-by-joint caution: agreement was considerably stronger at the knee than at the wrist, suggesting that markerless throwing analysis built around hip and knee kinematics sits on firmer ground than analysis built around fine distal wrist motion, where teams should interpret absolute values cautiously and lean more heavily on relative, within-system comparisons over time.
The Role of Theia3D in Movement Analysis
Theia3D was evaluated here in a distinctive dual role: as the markerless motion-capture tool being validated against marker-based motion capture, and simultaneously as the subject of a direct, head-to-head comparison between two of its own software versions processed from identical source video. This approach offers a useful, practical illustration of how markerless software evolves over time and how those changes can measurably affect agreement with marker-based reference data, information that is directly useful to researchers and performance staff deciding how to manage software versioning in their own ongoing work.
Study Limitations and Future Directions
The authors note an important methodological limitation they identify themselves: Theia3D's exact anatomical landmark and segment coordinate definitions are not publicly available, so the marker-based model built for this comparison could not be made perfectly identical to the markerless model, meaning some of the differences observed may reflect two distinct anatomical modeling approaches rather than purely reflecting tracking or algorithm accuracy. The throwing motion was simulated without an actual ball due to laboratory space constraints, which may not fully represent the kinematics of a genuine, ball-in-hand throw. The sample consisted of healthy adults with no prior experience performing the specific task, which may not generalize to trained throwing athletes with more refined, consistent mechanics. The authors identify further validating markerless motion capture across additional throwing-related and other fast overhead tasks, continuing to track how software updates affect measurement agreement over time, and encouraging consistent, transparent reporting of software version across published research as important next steps.
Citation: Thomas, E., et al. (2025). Comparison of marker-based and markerless motion capture systems for measuring throwing kinematics. Biomechanics, 5(4), 100. https://doi.org/10.3390/biomechanics5040100
Link: Read the full study
Does Markerless Software Version Change Throwing Kinematics Accuracy?
This study of 13 healthy adults compares marker-based motion capture against two different Theia3D software versions for a simulated basketball throwing task, finding that the newer version generally showed closer agreement with marker-based data at the elbow, hip, and wrist, while the authors caution that absolute kinematic values should not be directly compared across motion-capture systems. If your work involves throwing or overhead-sport performance research, talk to our team about applying Theia3D to your own studies.