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 v2023, as stated in the paper.
Study Overview
Design
A within-subject, concurrent comparison of knee joint kinematics from Theia3D markerless and marker-based motion capture during walking, assessed before and after applying a reference-frame realignment method (REFRAME) to the markerless data.
Participants
Ten healthy adults (5 men, 5 women; mean age 26.8) with no history of lower-limb surgery or musculoskeletal condition or disease.
Data Collection
Two camera systems recorded simultaneously: eight Qualisys Miqus Hybrid cameras for markerless capture and 24 Qualisys Oqus and Arqus cameras for marker-based capture, both at 120 Hz and sharing a common global reference frame. Participants walked at a comfortable self-selected speed, with five right-foot-strike and five left-foot-strike trials collected; only right-foot-strike trials were analyzed for this study.
Analysis
Knee joint kinematics were calculated in Visual3D for both systems and compared using root-mean-square error (RMSE) across all trials and subjects, both before and after applying the REFRAME realignment method to the markerless reference frames. Statistical parametric mapping with a paired t-test was used to identify the proportion of the gait cycle showing statistically significant differences between systems.
Key Findings
- Reference-frame realignment sharply reduced the measured disagreement in knee adduction and rotation. RMSE for knee ab/adduction dropped from 6.1 to 1.7 degrees after REFRAME, and knee internal/external rotation dropped from 10.2 to 2.5 degrees; flexion/extension, already the closest-agreeing angle, improved from 3.9 to 1.7 degrees.
- The portion of the gait cycle with statistically significant differences between systems fell substantially after realignment. Significant differences dropped from 40.8% to 22.2% of the cycle for flexion, from 78.3% to 31.5% for adduction, and from 80.6% to 20.6% for rotation.
- The realignment needed to achieve this improvement fell within a biomechanically plausible range, rather than requiring extreme or implausible rotations, which supports treating the original disagreement as a frame-orientation issue rather than an arbitrary data-fitting exercise.
- The markerless system showed higher trial-to-trial variability than the marker-based system in the raw data, which the authors attribute to the markerless system calculating a new reference frame for every trial, unlike the marker-based system's fixed marker placements; this variability also improved after realignment for several participants.
- The study's scope was limited to the knee joint alone. The authors note that hip and ankle kinematics also play a central role in gait and recommend extending this same realignment approach to all three major lower-limb joints in future work.
What This Means for Movement Science Researchers
For researchers trying to understand why markerless and marker-based systems sometimes disagree on specific joint angles, this study offers a methodological lens worth adopting: before concluding that a markerless system is measuring the "wrong" movement, it's worth asking whether a reference-frame mismatch, rather than a true kinematic disagreement, could explain the gap. The authors are careful to note an important caveat of their own approach: reconciling two systems using a frame-realignment method shows that they can be made to agree, not which system was more accurate to begin with, since both marker-based and markerless capture carry their own sources of error. For teams running their own markerless-versus-marker-based comparisons, this study is a reminder that segment and joint axis definitions deserve as much scrutiny as the raw kinematic signals themselves.
The Role of Theia3D in Movement Analysis
Theia3D was one of two systems directly compared in this study, with its markerless knee kinematics evaluated against simultaneously recorded marker-based data both before and after reference-frame realignment. Because Theia3D automatically generates its own segment coordinate systems from estimated body pose rather than from manually placed markers, this study's approach, testing whether realigning those automatically generated frames resolves apparent disagreement, offers a way to interpret Theia3D's kinematic output even without full visibility into how its underlying model defines each segment's axes.
Study Limitations and Future Directions
The authors note that their approach assesses agreement between the two systems after realignment, not the absolute accuracy of either one, since both marker-based and markerless motion capture carry their own sources of error. The analysis was limited to the tibiofemoral (knee) joint; the authors identify extending the same realignment approach to the hip and ankle as a clear next step now that appropriate objective criteria for that extension have been proposed. The study's ten-participant sample consisted of healthy adults walking under controlled laboratory conditions, which may not capture how reference-frame differences behave during more complex movements or in populations with altered gait patterns.
Citation: Antognini, C., Ortigas-Vásquez, A., Knowlton, C., Utz, M., Sauer, A., & Wimmer, M. A. (2025). Comparison of markerless and marker-based motion analysis accounting for differences in local reference frame orientation. Journal of Biomechanics, 185, 112683. https://doi.org/10.1016/j.jbiomech.2025.112683
Link: Read the full study
Is Disagreement Between Systems Always What It Looks Like?
This study suggests that at least some of the apparent disagreement between markerless and marker-based motion capture can come down to how each system orients its internal reference frames, not a fundamental difference in the movement each one detects. If your work involves comparing markerless and marker-based kinematics and you want to understand how reference-frame definitions could be affecting your own results, talk to our team about what this kind of analysis could mean for your data.