Study at a glance

In 23 typically developing children and 34 healthy adults, Theia3D showed differences from the widely used Conventional Gait Model across nearly every joint level and plane of motion, with the largest differences in the transverse plane, while segment lengths derived by the two systems differed by only 4 to 14 millimeters on average.

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 within-subject study directly comparing lower-body gait kinematics, joint kinetics, and body segment lengths derived simultaneously from Theia3D and the Conventional Gait Model (CGM1.1), one of the most commonly used marker-based gait models, in typically developing children and healthy adults.

Participants

Thirty-four healthy adults (22 women, 12 men; mean age 36.5 years; mean height 170.8 cm; mean mass 72.5 kg) and 23 typically developing children (15 girls, 8 boys; mean age 9.9 years; mean height 142.5 cm; mean mass 36.9 kg), recruited from hospital staff and their family members. All participants were free from any condition that would affect their walking.

Data Collection

Ten Qualisys Miqus video cameras and twelve Qualisys Arqus marker-based cameras captured synchronized trials at 100 Hz, along with four AMTI force plates recording ground reaction forces at 1000 Hz. Both camera systems were calibrated simultaneously within a shared global reference frame. Each participant wore minimal, tight-fitting clothing and was fitted with a reflective marker set following the CGM1.1 model, and completed at least five valid walking trials at a self-selected pace. One static trial and one additional walking trial were collected to evaluate and, if needed, correct the knee varus-valgus marker profile for crosstalk, consistent with standard marker-based gait assessment procedures.

Analysis

Video data were processed through Theia3D to generate joint centers, segment orientations, and a full-body kinematic model, while marker-based data were processed using the CGM1.1 model; both outputs were imported into Visual3D to ensure kinetic calculations and post-processing were performed consistently across both systems. Thorax, pelvis, hip, knee, and ankle kinematics were calculated across all three planes, with knee rotation and hip joint moments and powers additionally assessed. One-dimensional statistical parametric mapping was used to identify phases of the gait cycle where differences between systems exceeded chance, and root-mean-square deviation (RMSD) quantified the magnitude of significant differences. Segment lengths, calculated as the distance between proximal and distal segment endpoints, were compared between systems using mean differences, standard error, and minimal detectable change.

Key Findings

  • Significant kinematic differences between Theia3D and the Conventional Gait Model were observed across the gait cycle in all but one joint level and plane of motion, with differences reaching 8.5 degrees in the sagittal plane, 5.3 degrees in the frontal plane, and up to 10.2 degrees in the transverse plane. The transverse plane showed the largest and most consistent differences of the three planes assessed, in both children and adults.
  • Theia3D showed a systematic offset toward a more neutral pelvis position compared with the Conventional Gait Model's pelvis definition, which was moderately offset toward anterior tilt, a pattern the authors attribute in large part to differing underlying model definitions between the two systems rather than to measurement inconsistency. Knee flexion showed one of the larger sagittal-plane differences observed (6.7° in children; 8.5° in adults), consistent throughout the entire gait cycle in both groups.
  • Hip and ankle rotation showed the largest differences of any kinematic measures: the markerless system showed significantly greater external hip rotation during stance and pre-swing (10.2° in children; 8.8° in adults) and significantly greater external ankle rotation between mid- and terminal swing (9.2° in children; 10.2° in adults). These rotation-plane differences were broadly consistent with findings from prior markerless-versus-marker-based comparisons in both children and adults.
  • Theia3D produced a significantly greater peak knee extensor moment than the Conventional Gait Model in both children and adults, while joint kinetic waveforms otherwise showed broadly similar shapes between systems, with differences concentrated at specific phases of the gait cycle rather than throughout. Hip and ankle moment and power measures showed smaller, more localized differences between systems.
  • Segment lengths measured by the two systems were systematically different but numerically modest: the Conventional Gait Model produced longer segment lengths than Theia3D for nearly every segment in both children and adults, except the thorax in children, with the largest difference observed at inter-hip width (18 mm in children; 17 mm in adults) and most other segment differences falling within 4 to 14 millimeters.

What This Means for Movement Science Researchers and Gait Assessment Teams

For researchers and teams using, or considering adopting, Theia3D alongside or in place of the Conventional Gait Model, this study offers a detailed, directly applicable reference point across kinematics, kinetics, and segment lengths together. Sagittal- and frontal-plane kinematic differences frequently fell within commonly cited acceptability thresholds, and segment length differences were modest and systematic, supporting markerless capture's practical comparability for many everyday gait assessment purposes. The consistently larger transverse-plane and rotation-related differences are a useful, specific caution: teams whose work depends on precise hip or ankle rotation values should interpret those measures carefully until further research, and clearer evidence-based thresholds for what constitutes a practically meaningful difference, are established. The authors' broader point, that observed differences likely reflect expected offsets between fundamentally different model definitions rather than simple inaccuracy in either system, is a useful frame for interpreting any Theia3D-versus-marker-based comparison going forward.

The Role of Theia3D in Movement Analysis

Theia3D was the markerless motion capture system evaluated throughout this study, compared directly against the Conventional Gait Model, the marker-based gait model most widely used in real-world gait assessment settings, across kinematics, kinetics, and segment lengths in both typically developing children and healthy adults. This head-to-head comparison against this specific, most commonly used marker-based model fills a gap the authors identify in the existing evidence base, and the inclusion of joint kinetics and segment lengths alongside kinematics offers a more complete picture of markerless-versus-marker-based agreement than kinematic comparisons alone.

Study Limitations and Future Directions

The authors note that Theia3D's underlying neural network training dataset is proprietary, so the specific age, ethnicity, and BMI diversity represented in that training data cannot be independently assessed, even though broader and more diverse training data is understood to be important for reducing bias and improving how well markerless systems generalize across different populations. They also note that reflective markers used for the marker-based system were visible in the video recordings processed by Theia3D, and that because Theia3D's underlying network was likely not trained specifically on images showing markers attached to the body, marker presence may have influenced markerless segment and kinematic estimates in ways that remain unclear, representing a potential, if speculative, confounding effect specific to this kind of synchronized comparison study. Nine children and three adults were excluded from the kinetic analysis due to incomplete force plate contact during their walking trials, which may have reduced the kinetic analysis's representativeness. The authors emphasize that because neither marker-based nor markerless motion capture can be considered fully free of error, given well-documented sources of inaccuracy in marker-based systems themselves, these results should be interpreted as differences between two imperfect systems rather than as evidence of overestimation or underestimation by either one specifically; they identify comparison against imaging-based methods such as DEXA or EOS imaging as a more accurate benchmark for segment lengths in future work, and extending this kind of comparison into populations with altered gait patterns as an important next step before markerless gait assessment is more broadly adopted in care settings.

Citation: Kearney, J., Greaves, H., Robinson, M. A., Barton, G. J., O'Brien, T. D., Pinzone, O., Wright, D. M., Gibbon, K., & Foster, R. J. (2025). Comparison of Theia3D and the conventional gait model in typically developing children and adults in a clinical gait laboratory. Journal of Biomechanics, 193, 112995. https://doi.org/10.1016/j.jbiomech.2025.112995

Link: Read the full study

How Does Theia3D Compare Against the Most Widely Used Marker-Based Gait Model?

This study directly compares Theia3D against the Conventional Gait Model, the marker-based model most widely used in gait assessment settings, across kinematics, kinetics, and segment lengths in both children and adults, finding broadly comparable sagittal and frontal plane agreement alongside larger, more consistent transverse-plane and rotation differences. If your work involves gait assessment and you want to understand how Theia3D compares against your own established marker-based model, talk to our team about applying Theia3D to your own research.

Recent Posts