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: This paper does not state a specific Theia3D software version or a specific camera brand/model in its methods; it reports only that 8 high-resolution video cameras were used.
Study Overview
Design
A two-part study: a concurrent validation comparing markerless (Theia3D) and marker-based motion capture during late pregnancy, paired with a longitudinal evaluation of gait kinematics, kinetics, and spatiotemporal parameters using markerless motion capture alone, tracked from early pregnancy through late pregnancy and into the postpartum period, alongside a non-pregnant comparison cohort used to establish measurement repeatability.
Participants
Forty-six pregnant women recruited through obstetric care settings, assessed at approximately 12 weeks and 36 weeks of pregnancy and again at 12 weeks postpartum (with some missed visits due to early birth or scheduling conflicts). Ten of the third-trimester participants also wore a reflective marker set for the direct validation comparison. A separate cohort of 20 non-pregnant women with no history of lower-extremity trauma completed two repeat laboratory visits roughly 12 weeks apart to establish repeatability independent of pregnancy-related change.
Data Collection
At each visit, participants completed a standing static trial followed by six walking trials at a self-selected normal pace over a 6-meter distance, captured using 8 high-resolution video cameras at 100 Hz and, for the validation subset, synchronized with floor-embedded force plates. Video data were processed using Theia3D to generate three-dimensional kinematics, with joint moments calculated using an inverse dynamics approach in Visual3D for pelvis tilt and four lower-extremity joint angles and moments at the ankle, knee, and hip.
Analysis
For the validation component, right-leg joint angles and moments from the markerless and marker-based systems were compared in 10 third-trimester participants using waveform correlation, root-mean-square error, and normalized root-mean-square error across the gait cycle. For the longitudinal component, statistical parametric mapping was used to compare joint angle and moment waveforms, and paired tests were used to compare spatiotemporal parameters, between early pregnancy and late pregnancy, and separately between early pregnancy and postpartum. Repeatability was assessed using intraclass correlation coefficients across repeat sessions in both the non-pregnant and pregnant cohorts.
Key Findings
- Markerless and marker-based motion capture showed excellent waveform agreement at the hip, knee, and ankle during late pregnancy, with high correlations and low normalized error (correlations above 0.89, normalized root-mean-square error below 20%) across both kinematic and kinetic measurements.
- Pelvis anterior-posterior tilt was a clear exception: the two systems showed poor waveform agreement and high error (correlation of 0.14, normalized root-mean-square error of 110%), a limitation the authors attribute to the underlying markerless model not having been trained specifically on pregnant body shapes or the volumetric changes pregnancy produces, particularly around the pelvis.
- Markerless motion capture showed moderate to good repeatability across repeat sessions in non-pregnant participants and moderate to excellent repeatability within pregnant and non-pregnant participants during a single session, supporting its use for tracking change over time in this population.
- From early to late pregnancy, hip, knee, and ankle motion all shifted measurably: hip flexion and ankle dorsiflexion-then-plantarflexion increased, knee flexion decreased during terminal stance while increasing during mid-swing, and hip adduction patterns shifted during stance and swing, alongside increases in hip abduction moment, ankle plantarflexion moment, and a notable 86% decrease in knee extension moment during terminal stance and pre-swing.
- By 12 weeks postpartum, pelvis tilt and knee extension moment had returned to early-pregnancy levels, but hip flexion, ankle motion, and step width remained altered compared to early pregnancy, indicating that not every gait change pregnancy produces fully resolves within the first three months after birth.
What This Means for Movement Science Researchers and Perinatal Movement Assessment Teams
For researchers and teams studying movement across pregnancy and the postpartum period, this study offers two useful, complementary takeaways. First, Theia3D can be used with confidence for hip, knee, and ankle gait assessment in pregnant populations, showing agreement with marker-based motion capture comparable to what has been reported elsewhere, while pelvis tilt specifically should be treated as an open question in this population until further validation work, ideally including a true skeletal reference standard, is completed. Second, the longitudinal findings themselves are practically useful on their own terms: markerless motion capture was sensitive enough to detect clear, interpretable shifts in walking mechanics across pregnancy, and to show that some but not all of these changes persist three months after birth, information that could help inform expectations around recovery timelines and where continued movement-focused support may be most useful postpartum.
The Role of Theia3D in Movement Analysis
Theia3D served a dual role in this study: first as the system directly validated against marker-based motion capture in a subset of third-trimester participants, and then as the sole measurement tool used to track gait changes longitudinally across the full pregnant cohort and into the postpartum period. This combination, validating a markerless system specifically within a population it had not previously been tested in, then using it to generate new longitudinal findings that would be considerably harder to collect with a marker-based setup, is a useful illustration of how markerless capture can both earn and apply trust within an understudied population.
Study Limitations and Future Directions
The authors are direct that their validation sample was limited to only 10 participants at a single pregnancy time point, and that standard biomechanical modeling methods used to estimate body segment inertial parameters may introduce some error when applied to the unique mass changes pregnancy produces, though they note prior sensitivity analyses suggest this error is unlikely to be practically meaningful for the measures examined here. They are unable to confirm whether the markerless or marker-based system more accurately estimated pelvis tilt specifically, since neither was compared against a true skeletal reference standard such as biplanar fluoroscopy, and they note that Theia3D's underlying deep-learning model is proprietary, which limits their ability to understand exactly how it identifies joint centers or adjusts to the volumetric changes of pregnancy. The study cohort was also predominantly white and highly educated, limiting how well these specific findings generalize to a broader population. The authors identify training a pregnancy-specific markerless model as a valuable future direction, and suggest that a validated markerless approach could meaningfully reduce participant and research burden for larger-scale studies of gait-related questions during pregnancy going forward.
Citation: Meyers, M., Song, K., Nkansah-Andoh, A., Russo, D., Zawel, M., Kwon, M., Schaubel, D. E., Durnwald, C., Humbyrd, C. J., & Baxter, J. R. (2026). Gait biomechanics change during pregnancy and postpartum measured using markerless motion capture. Journal of Biomechanics, 208, 113613. https://doi.org/10.1016/j.jbiomech.2026.113613
Link: Read the full study
Does Markerless Capture Hold Up for Tracking Gait Changes During Pregnancy?
This study of 46 pregnant women finds that Theia3D shows strong agreement with marker-based motion capture at the hip, knee, and ankle during late pregnancy, with pelvis tilt as a notable exception, and goes on to use markerless tracking to document gait changes across pregnancy that only partially resolve by 12 weeks postpartum. If your work involves perinatal movement research, talk to our team about applying Theia3D to your own studies.