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 versions used: both v2021.2 and v2023.1.0.3160 were tested; results presented in the published paper are based on v2023.1.0.3160, with the authors noting minimal differences between the two versions.
Study Overview
Design
Sixteen healthy young adults performed four motor tasks, walking, a forward-lean balance-loss task, narrow beam walking, and countermovement jumps, while being recorded simultaneously by paired marker-based and markerless (Theia3D) camera systems. Center of mass position, extrapolated center of mass, and whole-body angular momentum were calculated from both datasets and compared using Bland-Altman analysis, RMSD, and correlation coefficients.
Participants
Sixteen healthy young adults (9 men, 7 women), mean age 25.1 ± 3.0 years, mean BMI 22.4 ± 3.2 kg/m², mean height 1.7 ± 0.1 m, mean weight 68.4 ± 13.5 kg, with no history of musculoskeletal or balance problems.
Data Collection
Marker-based data were captured using 10 Qualisys M3 cameras at 300 Hz, with participants wearing a full-body set of 46 reflective markers. Markerless data were captured simultaneously using 10 Qualisys Miqus Video cameras at 60 Hz (1920 × 1088 pixels), synchronized and spatially calibrated with the marker-based system. Tasks included treadmill walking at a self-selected comfortable speed, a forward-lean task inducing loss of balance and recovery steps, walking across a narrow raised beam, and countermovement jumps, each performed with multiple repetitions.
Analysis
Markerless video was processed using Theia3D to generate a 17-segment, 56-degree-of-freedom kinematic model, with body segment inertial parameters added in Visual3D. Marker-based data were processed through Qualisys Track Manager and Visual3D using a custom, participant-specific 17-segment, 54-degree-of-freedom model built via multibody kinematics optimization. CoM position, linear and angular segment velocities, and segment rotation matrices were exported for both systems, filtered, and used to calculate XCoM (following Hof et al.’s 2005 method) and unitless WBAM. Agreement between systems was assessed using Bland-Altman bias and limits of agreement, RMSD, and coefficient of determination (R²), computed across all tasks combined and for each task individually.
Key Findings
- Center of mass and extrapolated center of mass showed close agreement, with the largest discrepancy in the vertical direction. Bias was only a few tenths of a millimeter in the anteroposterior and mediolateral directions, but reached about 1 cm (specifically -1.17 cm for CoM and -1.14 cm for XCoM) in the superior-inferior direction across all tasks, with the markerless system consistently positioning CoM and XCoM slightly lower than the marker-based system.
- Whole-body angular momentum RMSD stayed under 10% of total amplitude in every direction. The largest relative RMSD was in the superior-inferior direction (7.58% of amplitude), compared to under 4% in the anteroposterior (3.16%) and mediolateral (3.87%) directions.
- Countermovement jumps produced the largest differences between systems across all three measures. XCoM RMSD reached 2.09 cm during jumps, compared to roughly 1 cm or less for the other three tasks, and WBAM correlation coefficients dropped to their lowest values during jumps (R² as low as 0.63 in the superior-inferior direction and 0.65 in the anteroposterior direction).
- Mediolateral direction consistently showed the smallest differences between systems for CoM and XCoM. The authors attribute this to the sagittal-plane-dominant nature of the tasks studied, which involve little displacement or velocity in the mediolateral direction.
- Markerless-derived WBAM differences were comparable to or smaller than clinically meaningful differences reported elsewhere in the literature. Comparing their RMSD values to a prior study of WBAM differences between young and older adults, the authors found the smallest clinically significant difference in that study was of the same order of magnitude as the RMSD found here between markerless and marker-based systems, suggesting markerless-derived WBAM could meaningfully distinguish between study groups or conditions, though researchers should remain aware of this margin when interpreting smaller effects.
What This Means for Balance Researchers and Biomechanists
For researchers studying dynamic balance and postural control, this study suggests that Theia3D can reliably reproduce whole-body center of mass, extrapolated center of mass, and angular momentum measures for tasks like walking, balance recovery, and beam walking, with differences on the order of a centimeter or less, well within the range reported between other markerless and marker-based comparisons in the literature. For more explosive or dynamic movements like jumping, the authors recommend some caution and encourage verifying beforehand whether the accuracy achievable with markerless capture is compatible with the magnitude of difference a given research application requires to detect.
The Role of Theia3D in Movement Analysis
Theia3D enabled the research team to compute advanced balance-related quantities, center of mass position, extrapolated center of mass, and whole-body angular momentum, directly from markerless video across four distinct motor tasks, without requiring marker placement or the full-body suit of reflective markers used by the marker-based comparison system. This supports more naturalistic data collection for balance research, potentially reducing the risk that marker placement or marker-related awareness alters how a participant actually balances or moves during testing.
Study Limitations and Future Directions
The authors identify several limitations. Participants were healthy young adults, raising the question of whether similar agreement would hold for participants with different ages, anthropometric characteristics, or body deformities, since the neural network underlying the markerless system may have been trained primarily on similar body types. The study was conducted in a controlled laboratory environment with minimal, tightly fitting clothing, controlled lighting, and a substantial number of synchronized cameras, conditions that may not reflect more typical or field-based data collection setups. Markers were also clearly visible in the markerless video, and no study to date has evaluated whether the presence of markers affects deep-learning landmark identification, though prior clothing-condition research found no meaningful changes to estimated parameters, suggesting marker presence likely has limited impact. Finally, because markerless data were compared against marker-based data rather than a true ground truth, and marker-based capture itself is subject to marker placement and soft tissue artefact errors, the authors note it is not possible to attribute the differences found in this study definitively to one system or the other.
Citation: Chaumeil, A., Lahkar, B.K., Dumas, R., Muller, A., & Robert, T. (2024). Agreement between a markerless and a marker-based motion capture systems for balance related quantities. Journal of Biomechanics, 165, 112018.
Link: https://doi.org/10.1016/j.jbiomech.2024.112018
Ready to Bring More Dynamic Balance Measures Into Your Research?
Whole-body balance measures like center of mass and angular momentum are powerful tools for understanding stability and movement strategy, but collecting them with traditional marker-based capture takes time and setup most labs would rather spend on data collection itself. If your balance or postural control research could benefit from a markerless approach to these measures, talk to our team about how Theia3D fits into your specific study design.