Human Movement & Mobility Research
FLIR (Vicon)

Comparison of Joint Kinematics and Spatiotemporal Gait Parameters Between Markerless and Marker-Based Motion Capture in a Large Knee Osteoarthritis Cohort

Annals of Biomedical Engineering
•
2026
•
Original research
Theia3D version:
v2023.1.0.3161
More Articles

Study at a glance

In 162 individuals with advanced knee osteoarthritis and 50 healthy controls, Theia3D showed excellent agreement with marker-based motion capture for spatiotemporal gait parameters and for sagittal-plane hip and knee angles, but most other joint angle measures showed weaker agreement, with errors frequently exceeding the commonly used 5-degree acceptability threshold, and measurement consistency was significantly lower in the osteoarthritis group than in healthy controls for several specific angles.

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 v2023.1.0.3161, as stated in the paper.

Study Overview

Design

A cross-sectional study directly comparing joint kinematics and spatiotemporal gait parameters derived simultaneously from a marker-based and a markerless (Theia3D) motion capture system during overground walking and sit-to-stand tasks, in a large cohort of individuals with knee osteoarthritis and a group of healthy controls, with a secondary aim of comparing measurement error between the two groups.

Participants

A convenience sample of 162 individuals with knee osteoarthritis (mean age 64.8 years; mean height 156.0 cm; mean weight 62.7 kg; mean BMI 25.7; 24% men), recruited from a single hospital, and 50 healthy controls (mean age 56.3 years; mean height 160.3 cm; mean weight 58.9 kg; mean BMI 22.9; 28% men). Inclusion for the osteoarthritis group required a definitive radiographic diagnosis of knee osteoarthritis with a Kellgren-Lawrence grade of III or IV in at least one knee, a scheduled knee replacement procedure, and the ability to walk independently without an assistive device; participants were excluded if they had undergone lower-extremity surgery within the previous six months. Controls were a convenience sample without radiographic or symptomatic evidence of knee osteoarthritis. The two groups were not matched at baseline and differed meaningfully in age, BMI, and height.

Data Collection

All participants underwent simultaneous marker-based and markerless motion capture during the same trials, with both systems temporally synchronized and recording at 150 Hz. The marker-based system used a 10-camera setup with 34 reflective markers following a standard lower-limb marker model; the markerless system used 7 synchronized Blackfly S cameras processed through Theia3D. For the walking task, participants walked barefoot along a 7-meter walkway at a self-selected speed; for the sit-to-stand task, performed at the middle of the walkway, participants crossed their arms over their chest and rose from seated to standing and back without using their hands. Individuals with knee osteoarthritis completed assessments within one week prior to their scheduled surgery; participants unable to complete the sit-to-stand task due to limited functional capacity completed only the walking task.

Analysis

Video data were processed through Theia3D's inverse kinematics pipeline, with hip, knee, and ankle joints constrained to three degrees of freedom, and the resulting biomechanical model was automatically generated in Visual3D. Marker-based data were separately processed in Visual3D using a subject-specific model built from a static calibration trial. Agreement between systems was assessed using Bland-Altman analysis, mean difference, Pearson correlation, and intraclass correlation coefficient (ICC) for both spatiotemporal parameters and joint angle waveforms, with ICC values interpreted as poor (below 0.50), moderate (0.50-0.75), good (0.75-0.90), or excellent (above 0.90). Root-mean-square error (RMSE) quantified joint angle differences between systems, and group differences in ICC and RMSE between the osteoarthritis and control groups were evaluated using independent-samples t tests or Wilcoxon rank-sum tests.

Key Findings

  • Spatiotemporal gait parameters showed excellent agreement between the two systems in both groups, with mean absolute differences under 1 centimeter for spatial parameters and under 0.01 seconds for temporal parameters, and very strong correlations and excellent agreement across all nine parameters measured, including gait speed, stride length, and stance and swing time.
  • Sagittal-plane hip and knee angles during walking showed good to excellent agreement between systems (ICC = 0.85-0.91 in the osteoarthritis group; 0.93-0.98 in controls), consistent with prior Theia3D validation findings in other populations, while sagittal-plane ankle angle agreement was notably weaker, particularly in the osteoarthritis group (ICC = 0.54, considered only moderate, compared with 0.83 in controls, a statistically significant difference).
  • Agreement was poor for the large majority of frontal- and transverse-plane joint angles in both groups, with most of these comparisons falling below the threshold considered even moderate agreement. The authors attribute this broad pattern to compounding factors affecting both systems: marker-based motion capture's own established susceptibility to soft tissue artifact in non-sagittal planes, and markerless motion capture's inherent difficulty resolving axial rotation from the relatively cylindrical, feature-limited silhouette of the lower limb.
  • The majority of lower-limb joint angle errors in the osteoarthritis group during walking exceeded the 5-degree threshold commonly used to judge acceptable motion capture accuracy, with root-mean-square errors as high as 9.4 degrees for sagittal ankle angle, a significantly larger error than the 4.6 degrees observed in healthy controls for the same measure.
  • Measurement consistency was significantly lower in the osteoarthritis group than in healthy controls for several specific joint angles, including sagittal-plane ankle angle during both walking and sit-to-stand and frontal-plane hip angle during walking, a pattern the authors attribute in part to markerless pose-estimation algorithms being trained predominantly on data from individuals without the joint deformities, swelling, and altered movement patterns common in advanced knee osteoarthritis.
  • The authors highlight a notable mismatch between high ICC values and RMSE values exceeding the acceptability threshold, particularly for sagittal hip angle: a high ICC shows the system reliably captures the overall shape of an individual's movement pattern, useful for tracking change over time, but a large RMSE means the absolute joint angle values themselves may be meaningfully inaccurate, a distinction the authors argue both need to be evaluated together before a markerless system is used for applications depending on precise joint angle values.

What This Means for Movement Science Researchers and Osteoarthritis Research Teams

For researchers and teams working with knee osteoarthritis populations, this study offers both encouraging and important cautionary findings. Spatiotemporal gait parameters and sagittal-plane knee angle showed strong, consistent agreement with marker-based motion capture in this large cohort, supporting Theia3D's practical use for these specific measures even in a population with substantial joint deformity. At the same time, the majority of other joint angle measures, and frontal- and transverse-plane angles in particular, showed meaningfully weaker agreement, with the osteoarthritis group showing significantly larger errors than healthy controls for several specific angles. This is a useful, concrete reminder that a markerless system's validated performance in healthy populations should not be assumed to transfer automatically to populations defined by the very joint changes that make objective movement assessment valuable, and that applications depending on precise absolute joint angle values, rather than relative change over time, warrant particular caution in this population until algorithms and training data better represent diverse structural presentations.

The Role of Theia3D in Movement Analysis

Theia3D was the markerless motion capture system evaluated throughout this study, assessed directly against a 34-marker Vicon-based reference system in both individuals with advanced knee osteoarthritis and healthy controls. This head-to-head, large-sample design allowed the authors to establish not only how well Theia3D agreed with marker-based motion capture overall, but specifically whether that agreement held up in a population whose joint deformities and altered gait patterns represent a meaningfully different, more practically relevant test of the software's accuracy than the healthy populations most markerless validation research has focused on to date.

Study Limitations and Future Directions

The authors note that spatiotemporal gait parameters from both systems were derived from kinematically detected gait events rather than compared against an independent reference tool such as a pressure-sensitive gait mat. The evaluation was limited to individuals with end-stage knee osteoarthritis, typically characterized by pronounced lower-limb deformity, so those with milder, early- or middle-stage disease might show better markerless agreement than observed here. The osteoarthritis and control groups were not matched at baseline and differed meaningfully in age, BMI, and height, individual characteristics that could independently affect markerless tracking performance, making it difficult to determine how much of the observed error difference is attributable specifically to osteoarthritis itself rather than these other factors; the authors identify future studies using matched cohorts as a needed step to isolate that effect. Twenty-nine participants with osteoarthritis, 17.9% of the group, were unable to complete the sit-to-stand task and were excluded from that portion of the analysis, likely representing those with more severe deformity and movement alteration, meaning the sit-to-stand findings should be interpreted cautiously and may not fully generalize to individuals with more severe functional limitation. Finally, because Theia3D is built on computer vision algorithms under ongoing development, the authors note that future software updates may improve pose-estimation performance, meaning these findings are specific to the version of Theia3D evaluated here and may not generalize to future releases; they identify comparing markerless system outputs against a gold-standard biplane radiographic reference, and expanding training datasets to include more diverse structural presentations, as important next steps.

Citation: Wang, J., Zhang, Q., Wang, B., Deng, T., Niazi, I. K., Nie, Y., & Li, K. (2026). Comparison of joint kinematics and spatiotemporal gait parameters between markerless and marker-based motion capture in a large knee osteoarthritis cohort. Annals of Biomedical Engineering. https://doi.org/10.1007/s10439-026-04339-8

Link: Read the full study

Does Markerless Accuracy Hold Up in a Large Knee Osteoarthritis Cohort?

This large comparative study finds that Theia3D shows excellent agreement with marker-based motion capture for spatiotemporal gait parameters and sagittal-plane hip and knee angles in people with knee osteoarthritis, while most other joint angle measures showed weaker agreement, with errors often exceeding standard accuracy thresholds and measurement consistency significantly lower than in healthy controls for several specific angles. If your work involves studying knee osteoarthritis biomechanics and you want to understand where markerless capture currently stands for this population, talk to our team about the latest Theia3D performance for your specific research application.

Recent Posts