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 v2022.1.0.2309, as stated in the paper.
Study Overview
Design
A secondary analysis directly comparing knee joint kinematics and kinetics derived simultaneously from a marker-based and a markerless (Theia3D) motion capture system, during self-selected and fast-paced walking and sit-to-stand tasks, in adults with knee osteoarthritis.
Participants
Twenty-two adults with knee osteoarthritis (mean age 63.2 years; mean BMI 28.2; 64% women) with mild-to-moderate osteoarthritis-related disability, enrolled from a single study visit within a broader registered trial. Participants were included based on an osteoarthritis diagnosis meeting American College of Rheumatology criteria, a knee pain score of 3 or greater on a standardized weight-bearing pain scale, a BMI of 40 or less, and the ability to walk without an assistive device for at least 20 minutes; exclusion criteria included health conditions limiting walking ability other than knee osteoarthritis, recent knee surgery, and pain in a limb or back region greater than their knee pain.
Data Collection
Participants, wearing study-provided shorts, tank-top, and neutral shoes, walked at a self-selected pace and a fast pace along a 12-meter walkway, and completed five sit-to-stand repetitions from an armless chair without using their arms. Marker-based data were collected using 55 retroreflective markers and 12 synchronized infrared cameras at 250 Hz; markerless data were collected simultaneously using 8 synchronized color video cameras (Qualisys Miqus) at 50 Hz and processed through Theia3D, with both systems' cameras mounted on an overhead rail and centered over floor-embedded force platforms recording ground reaction force data at 2000 Hz synchronously with both motion capture systems.
Analysis
Knee joint angle and moment waveforms, along with discrete measures including peak knee flexion and adduction angles, sagittal and frontal knee range of motion, and peak and impulse knee flexion and adduction moments, were compared between the markerless and marker-based systems using intraclass correlation coefficients, Pearson's correlation coefficients, average root-mean-square differences across the stance or ascent phase, and Bland-Altman analysis. Joint center position agreement between systems was additionally assessed using the 3D Euclidean distance between estimated hip, knee, and ankle center positions for each task.
Key Findings
- Across all three tasks, root-mean-square waveform differences between the two systems ranged from 3.8 to 5.7 degrees for knee kinematics and 0.2 to 1.2 percent body-weight-times-height for knee kinetics, with joint center position differences under 5 centimeters for every joint analyzed. These values were broadly consistent with figures previously reported for the same markerless system in healthy and other populations.
- Agreement was excellent for knee adduction moment impulse during stance across both walking speeds (ICC = 0.94) and for the second peak knee adduction moment during fast walking (ICC = 0.91). Knee adduction moment is a key predictor of medial knee joint loading and is frequently targeted in biomechanical interventions for knee osteoarthritis, making this level of agreement particularly relevant for osteoarthritis-focused applications.
- Agreement was good for several additional measures, including peak knee adduction angle (ICC = 0.78-0.83), sagittal knee range of motion (ICC = 0.78-0.83), first peak knee flexion moment (ICC = 0.76-0.81), and knee adduction moment (ICC = 0.87-0.90), as well as second peak knee adduction moment during self-selected walking (ICC = 0.84) and both knee flexion moment impulse during fast walking (ICC = 0.84) and peak knee flexion moment during sit-to-stand (ICC = 0.78). The remaining measures showed moderate or poor agreement.
- Frontal-plane knee range of motion consistently showed weaker agreement between systems than sagittal-plane measures across all three tasks, with markerless estimates tending to lean toward greater abduction, particularly later in the stance or ascent phase, a pattern the authors attribute in part to known challenges in reliably measuring smaller, more subtle frontal-plane movements with either motion capture approach.
- Joint center position differences increased moving from the ankle to the knee to the hip across all tasks, with the hip showing the largest differences (average root-mean-square distances of roughly 2.9 to 3.7 centimeters), which the authors note may reflect the inherent challenge of reliably estimating hip joint center position, a known source of error for both markerless and marker-based systems, particularly in people with higher BMI.
What This Means for Movement Science Researchers and Osteoarthritis Research Teams
For researchers studying knee osteoarthritis biomechanics, this study offers encouraging, practically specific guidance: markerless motion capture showed good-to-excellent agreement with marker-based capture for several of the knee measures most closely tied to osteoarthritis progression and joint loading, including knee adduction moment and its impulse, sagittal knee range of motion during walking, and sagittal knee moments during sit-to-stand, supporting its potential as an efficient, more accessible alternative for these specific measures and tasks. At the same time, the comparatively weaker agreement observed for frontal-plane range of motion, and the wider individual variability seen across measures, are useful cautions: these findings point toward markerless capture being most immediately useful at the group level and for specific, well-supported outcomes, with the authors noting that further test-retest reliability and sensitivity-to-change research is needed before individual-level application can be recommended.
The Role of Theia3D in Movement Analysis
Theia3D was the markerless motion capture system evaluated throughout this study, compared directly against a marker-based reference system for knee kinematics and kinetics during walking and sit-to-stand in adults with knee osteoarthritis. This head-to-head design extends prior markerless validation work, much of which has focused on healthy populations during level-ground walking, into a population and a broader set of everyday functional tasks, walking at two speeds and a weight-bearing sit-to-stand transfer, that carry direct relevance for tracking osteoarthritis-related mobility and joint loading over time.
Study Limitations and Future Directions
The authors note that differences in the underlying biomechanical models used by the two systems present an inherent challenge to direct comparability: the markerless system automatically generates models from deep-learning-based pose estimation, while the marker-based system relies on manually defined anatomical landmarks, and although both systems were evaluated with knee models matched to three degrees of freedom for this analysis, model customization choices may still influence results as the underlying technology continues to evolve. As a secondary analysis, the available sample was limited to participants with a BMI of 40 or less, which may not reflect markerless modeling accuracy in individuals with greater adiposity, and osteoarthritis radiographic severity grade was not collected, leaving the structural severity of participants' disease unknown, though their reported symptom scores indicated mild-to-moderate disability; results may differ in people with more severe disease or joint deformity. The relatively limited sample size also widened the confidence intervals around the reported agreement estimates, meaning these findings should be considered preliminary rather than definitive. The study's markerless data were also collected at a lower camera frame rate (50 Hz) than is sometimes used, which may have limited accuracy during the more rapid limb movements that occur during the swing-to-stance transition, and only the index knee was analyzed, meaning these findings may not fully capture within-participant variability or generalize to the opposite limb. The authors identify future test-retest reliability studies and research establishing sensitivity to practically meaningful change as necessary steps before markerless-derived knee measures in this population can be endorsed for individual-level application.
Citation: Torres, R. T. G., Senderling, B., Kim, E., Rose, M., Gheller, M., Neogi, T., & Kumar, D. (2026). Agreement between markerless and marker-based motion capture for knee kinematics and kinetics during functional activities in knee osteoarthritis. Osteoarthritis and Cartilage Open, 8, 100853. https://doi.org/10.1016/j.ocarto.2026.100853
Link: Read the full study
Does Markerless Capture Hold Up for Knee Osteoarthritis Research Beyond Walking?
This study finds that Theia3D shows good-to-excellent agreement with marker-based motion capture for several of the knee kinematic and kinetic measures most relevant to osteoarthritis research, across both walking and sit-to-stand tasks in adults with knee osteoarthritis. If your work involves studying knee osteoarthritis biomechanics or tracking joint loading across everyday functional tasks, talk to our team about applying Theia3D to your own osteoarthritis research.