Study at a glance
Summary
A motion capture system can be accurate on a single visit and still be unreliable for tracking someone over time if its numbers drift every time they come back for another session. This early validation study from the Human Mobility Research Lab at Queen’s University tested that directly, bringing eight healthy adults back for three separate walking sessions, roughly a week apart, and measuring how consistent Theia3D’s gait kinematics were across those repeat visits compared to similar previously published repeatability values for marker-based systems.
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: The specific Theia3D software version used was not stated in the published paper, but was no later than Theia3D v2021.
Why This Matters
Marker-based motion capture has a well-documented repeatability literature: multiple studies have measured how much a participant’s gait numbers vary from one session to the next, separate from the natural stride-to-stride variation within a single visit. For markerless capture to be taken seriously as a rigorous research tool, it needed the same kind of scrutiny; agreeing with marker-based systems on a single day isn’t the same question as whether it gives you a stable baseline to track someone’s movement against over repeated visits. This study, one of the earliest formal repeatability analyses of Theia3D, set out to answer that using the same established method researchers have used to benchmark marker-based systems for years.
Study Overview
Design
A repeatability study using an established test-retest protocol, comparing inter-trial variability (stride-to-stride noise within a session), inter-session variability (variation across separate visits), and the ratio between them.
Participants
8 healthy, recreationally active adults (2 female, 6 male; mean age 30.3), who returned for three sessions averaging 8.5 days apart. Participants wore whatever clothing they arrived in, with no instructions on attire.
Procedure
At each session, participants completed five overground walking trials at a comfortable pace, captured by eight synchronized cameras and processed through Theia3D. Each full session, including setup, typically took 5 to 10 minutes.
Key Findings
- Session-to-session consistency was the best of any system compared. Average inter-session variability was 2.8 degrees, the smallest of any study in the comparison, against a range of 3.0 to 3.6 degrees previously reported for marker-based systems.
- The multi-session protocol itself added very little extra noise. The variability ratio, which measures how much additional variability comes specifically from testing on separate days rather than from normal stride-to-stride differences, averaged 1.1. That means returning for a separate session only added about 10% more variability on top of what already exists between strides in a single visit, smaller than any of the marker-based studies used for comparison.
- Single-session, stride-to-stride variability was slightly higher than some marker-based benchmarks. Inter-trial variability averaged 2.5 degrees, modestly higher than the 1.0 to 2.4 degree range reported in prior marker-based work. The authors attribute this to the frame-by-frame nature of the markerless tracking algorithm rather than to genuine differences in how the healthy participants walked.
- Results held up regardless of which internal model configuration was used. Comparing Theia3D’s default model to an alternative ankle configuration produced nearly identical repeatability results, with differences no larger than 0.1 degrees.
- Participants’ walking speed stayed consistent across sessions on its own, with an average standard deviation of just 0.054 m/s, suggesting the repeatability findings reflect genuine measurement consistency rather than participants simply walking the same way because they were told to.
What This Means for Research Teams
For anyone considering markerless capture for longitudinal tracking, whether that’s following how a research participant’s movement changes over multiple visits or running a multi-visit research protocol, this study’s core finding is reassuring: the day-to-day drift that matters most for tracking change over time was smaller here than in established marker-based systems, even though this was markerless capture’s first time being tested against this specific benchmark. Combined with a 5 to 10 minute session time and no attire restrictions, it points to a system that’s not just accurate once, but stable enough to build a longitudinal protocol around.
The Role of Theia3D in Movement Analysis
This study is notable for what it didn’t do: no reflective markers, no requirement for participants to change into tight-fitting clothing, and no dedicated marker-placement time before each session. Eight participants in their own street clothes walked in front of eight cameras, and Theia3D produced gait kinematics that repeated more consistently across separate sessions than published marker-based benchmarks. This is also one of the foundational studies in Theia3D’s reliability record, work that later studies examining markerless capture in community and real-world research settings have built directly on.
Study Limitations and Future Directions
This study intentionally did not compare markerless and marker-based kinematics within the same trials, since the authors wanted participants to wear unrestricted clothing; that direct comparison was addressed separately in a companion paper. Most participants happened to wear predominantly dark clothing, which the authors note could plausibly make it harder for the system to distinguish anatomical features, though this wasn’t tested directly. The approach also requires both legs to be visible, which rules out full-length coats or skirts. Despite markerless capture’s flexibility for real-world environments, this particular study was still conducted in a lab, and the authors call for further work testing sensitivity to different collection environments, subject appearance, and capture volume size.
Full Study and Further Reading
Citation: Kanko, R.M., Laende, E., Selbie, W.S., & Deluzio, K.J. (2021). Inter-session repeatability of markerless motion capture gait kinematics. Journal of Biomechanics, 121, 110422.
Link: Read the full study
Building a Longitudinal Gait Protocol?
This study shows markerless capture holding steady across separate sessions days apart, without the setup burden a repeat marker-based visit requires. If your work depends on tracking the same person’s movement over multiple visits, talk to our team about what that looks like for your protocol.