Validation, Methods & System Performance
Qualisys Miqus

Concurrent Validity of Theia3D Markerless Motion Capture for Detecting Sagittal Kinematic Differences Between Gait Speeds

Journal of Sports Sciences
•
2025
•
Original research
Theia3D version:
v2023.1.0.3161
More Articles

Study at a glance

Across four walking and running speeds on an instrumented treadmill, Theia3D detected changes in hip, knee, and ankle sagittal-plane kinematics with accuracy comparable to marker-based motion capture, though pelvic tilt agreement between the two systems was consistently weaker.

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: 2023.1.0.3161, patch 4 (explicitly stated in the paper, which also notes this software continues to be updated and that results may not fully transfer to newer versions).

Study Overview

Design

This was a concurrent validity study using a within-subject, repeated-measures design. Each participant completed four gait conditions on a single motorized, instrumented treadmill, two walking speeds (3 and 5 km/h) and two running speeds (10 and 12 km/h), presented in randomized order. Sagittal-plane kinematic data for the pelvis, hip, knee, and ankle were collected simultaneously by a marker-based optoelectronic system (treated as the criterion method) and by Theia3D markerless motion capture, allowing direct, synchronized comparison of both single-speed kinematics and the change in kinematics between speeds.

Participants

Fourteen recreationally active adults volunteered (10 male, 4 female; mean age 35 ± 7 years; stature 1.77 ± 0.09 m; body mass 77.8 ± 11.6 kg), recruited via convenience sampling. All participants completed medical screening and were confirmed free of any musculoskeletal or neurological injury or condition that might affect gait. The study received ethical approval from the university’s Local Research Ethics Committee (approval number 116353), and all participants provided written informed consent in accordance with the Declaration of Helsinki.

Data Collection

Participants walked or ran on a motorized treadmill (Gaitway3D 170/65 Stellar) for up to 3 minutes per speed, with 10 seconds of data collected from the final 30 seconds, and 2 minutes of active recovery between conditions. A 5-minute warm-up of intermittent walking and jogging preceded testing. Marker-based data were collected using 14 infrared cameras (Oqus 7+, Qualisys AB) with retroreflective markers placed per the Calibrated Anatomical Systems Technique (CAST) by a single experienced operator. Markerless data were collected simultaneously using 12 video cameras (Miqus Video, Qualisys AB) at 1,920×1,080 resolution, processed through Theia3D (version 2023.1.0.3161, patch 4), which tracked the pelvis, hips, and ankles as 6-degree-of-freedom joints; the knee was modeled with 3 degrees of freedom, as Theia3D does not yet support 6DoF knee modeling. Both camera systems were integrated into the same data acquisition software and calibrated together, and both kinematic models were brought into a shared Visual3D workspace for direct comparison.

Analysis

Gait cycles were time-normalized to 101 data points and averaged per condition. The similarity of joint angle waveforms between systems was assessed using the coefficient of multiple determination (CMD), interpreted as poor (<0.50), moderate (0.50–0.74), good (0.75–0.89), or excellent (>0.90). Agreement between discrete outcome measures (peak flexion, peak extension, range of motion) was assessed using the intraclass correlation coefficient (ICC, single-rating, absolute agreement, two-way mixed-effects model), interpreted per established guidelines, alongside standard error of measurement (SEM) between systems. The change in kinematics between speeds was calculated for each variable by subtracting the slower-speed waveform or value from the faster-speed waveform or value, then comparing that computed change between the two systems using the same CMD, ICC, and SEM approach.

Key Findings

  • Hip, knee, and ankle waveforms showed excellent agreement at every individual speed. CMD values were ≥0.95 across all four speeds for these three joints, despite some consistent systematic offsets between systems attributable to differences in model definitions rather than differences in how the underlying movement itself was tracked.
  • Pelvic tilt agreement was notably weaker and more variable, especially during walking. Waveform similarity for pelvic tilt was poor at both walking speeds and moderate at both running speeds, with markedly higher inter-individual variability than seen at the hip, knee, or ankle.
  • Theia3D detected speed-related changes in hip, knee, and ankle kinematics with accuracy comparable to marker-based motion capture. Waveform similarity for the change in kinematics between speeds was excellent for hip and knee during both walking and running, and excellent for ankle during walking, though only “good” for ankle during running.
  • Discrete outcome measures for lower-limb joints showed mostly good-to-excellent agreement, with sub-1° measurement error. Agreement (ICC) for mean differences in peak flexion, peak extension, and range of motion ranged from poor to excellent across measures, but SEM between systems stayed below 1° for all lower-limb joint comparisons, with the exception of ankle range of motion change during running, which showed poor ICC despite a similarly low SEM.
  • Pelvic kinematic outcome measures showed consistently lower agreement than lower-limb joints. ICC values for pelvis-related measures ranged from 0.032 to 0.776 (poor to moderate), and mean differences between systems for change in peak anterior pelvic tilt and range of motion during running were notably larger, around 2.5°, than the sub-1° differences seen at the hip, knee, and ankle.

What This Means for Researchers and Practitioners Using Markerless Gait Analysis

For anyone using or considering Theia3D for gait analysis across a range of speeds, whether in sports performance testing, graded treadmill protocols, or exercise physiology research, this study offers reassurance specifically for lower-limb sagittal-plane kinematics: hip, knee, and ankle angle changes with speed appear to be captured with accuracy similar to marker-based systems, up to the running speeds tested here (12 km/h). Pelvic tilt is a different matter. The authors’ data suggest practitioners should treat pelvis kinematics from Theia3D with more caution, particularly in applications where small pelvic tilt changes carry clinical or performance significance, since both single-speed agreement and sensitivity to speed-related change were weaker and more variable for this segment than for the lower-limb joints.

The Role of Theia3D in Movement Analysis

In this study, Theia3D served as the markerless motion capture system being directly validated against a marker-based optoelectronic reference system. Using synchronized video from 12 cameras, Theia3D generated 3D pose estimates and joint kinematic data for the pelvis, hip, knee, and ankle without requiring markers to be placed on participants, with its output integrated into the same Visual3D workspace as the marker-based data to allow like-for-like comparison. The study’s central contribution is evaluating not just whether Theia3D’s joint angles match marker-based values at a given speed, but whether Theia3D accurately captures how those angles change as a person speeds up, a distinct and practically important form of validity for any application involving variable-speed testing.

Study Limitations and Future Directions

The authors note that their findings are specific to the Theia3D software and hardware configuration used in this study and may not generalize to other markerless systems, such as OpenPose, AlphaPose, or OpenCap, since Theia3D tracks a higher number of key points than many open-source pose estimation algorithms, meaning these results could represent a “best case scenario” for markerless motion capture rather than a universal benchmark. The study was also limited to treadmill walking and running and to sagittal-plane kinematics only; the authors explicitly caution that practitioners should be careful about extending these findings to overground gait or to multi-planar movements such as change-of-direction tasks, where frontal- and transverse-plane kinematics become more important. Finally, the authors flag that the specific Theia3D software version used (2023.1.0.3161, patch 4) continues to undergo updates and modifications, and while they believe the transferability of their results to updated versions likely remains valid, they recommend users stay mindful of potential changes to modeling and tracking in future releases. The authors propose that future research explore individual differences in kinematic responses in more detail, accounting for cycle-to-cycle variability.

Citation: Walker, J., Thomas, A., Lunn, D. E., Nicholson, G., & Tucker, C. B. (2025). Concurrent validity of Theia3D markerless motion capture for detecting sagittal kinematic differences between gait speeds. Journal of Sports Sciences, 43(16), 1560–1571.

Link: https://doi.org/10.1080/02640414.2025.2513151

How Sensitive Is Your Motion Capture to Changes in Speed?

Validating a system’s accuracy at one fixed speed only tells part of the story, what matters in many real-world testing protocols is whether that system reliably captures how movement changes as effort increases. If your team is designing speed-based testing, training, or rehabilitation protocols and wants to understand what markerless motion capture can and can’t tell you across that range, talk to our team about building a measurement approach suited to your specific application.

Recent Posts