Study at a glance

This study compared Theia3D markerless motion capture to a marker-based system for assessing lower limb kinematics, muscle activation, and joint reaction forces in participants walking and running under four levels of military body-borne load, finding strong correlations between systems alongside specific, load-dependent differences in joint angles and joint loading estimates.

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 not specified in the paper.

Study Overview

Design

Sixteen healthy adults performed walking and running trials under four progressively heavier body-borne loading conditions, recorded simultaneously by a marker-based and a markerless (Theia3D) motion capture system alongside force plates and EMG sensors. Both systems’ data were processed through the same OpenSim musculoskeletal modeling pipeline to estimate joint kinematics, muscle activation, and joint reaction forces, which were then compared between systems.

Participants

Sixteen healthy adults over age 18 (8 male, 8 female; mass 72.8 ± 10.0 kg; height 1.7 ± 0.1 m; age 30 ± 12 years).

Data Collection

Participants completed a static calibration pose and three repetitions each of walking and running over a 6-meter distance under four loading conditions: LC1 (5 kg, boots and helmet), LC2 (21 kg, adding tactical and fragmentation vests), LC3 (35 kg, adding a 14 kg small pack), and LC4 (41 kg, adding a 20 kg rucksack instead). Marker-based data were captured using 11 infrared cameras at 120 Hz with 37 individual markers and 10 marker clusters; markerless data were captured simultaneously using 7 RGB cameras at 60 Hz. Ground reaction forces were recorded from two in-ground force plates, and EMG data were recorded from eight lower-limb muscles, including the gastrocnemius lateralis and medialis.

Analysis

Both motion capture datasets were processed through an identical OpenSim pipeline using a full-body musculoskeletal model, including inverse kinematics, residual reduction, static optimization for muscle forces, and joint reaction analysis for ankle, knee, and hip joint loading. Joint angles, muscle activation timing, and joint reaction forces were compared between systems using root mean square error (RMSE), Pearson’s correlation coefficients, and Bland-Altman limits of agreement, with statistical models testing whether differences between systems varied by joint and by loading condition.

Key Findings

  • Joint angle curves showed strong to very strong agreement between systems across all loading conditions. Correlation coefficients ranged from 0.86 to 0.996 across the hip, knee, and ankle for both walking and running, with the hip and knee showing the strongest agreement.
  • Differences between systems grew with load for the hip and knee, but shrank for the ankle. RMSEs for hip and knee flexion increased as equipment load increased, while ankle flexion RMSEs actually decreased with heavier loads, a pattern the authors could not fully explain and flag for further investigation.
  • Muscle activation timing matched EMG closely for both motion capture systems. Gastrocnemius activation estimated from both marker-based and markerless data peaked at the same point in the gait cycle as EMG-measured activity, around 40-50% of the cycle for walking and 30-40% for running.
  • Joint reaction forces were consistently higher when estimated from markerless data. Despite very strong correlation between systems (r > 0.9), the markerless-derived model estimated greater hip and knee joint reaction forces than the marker-based model in most conditions, which the authors attribute to small kinematic differences driving different estimated muscle recruitment strategies between the two models, including greater hip extensor and knee flexor muscle forces in the markerless-derived data, rather than a simple accuracy failure on either side.
  • The marker-based “gold standard” system itself struggled under heavy equipment, a point worth weighing against any instinct to treat markerless differences as markerless error. The authors explicitly note that the marker-based system frequently lost line of sight to multiple pelvis markers at once under the bulkiest loading conditions, due to obstruction from the rucksack and rifle, meaning at least some of the divergence between systems likely reflects marker-based tracking difficulty under heavy gear, not purely a markerless shortfall.

What This Means for Military Biomechanists, Ergonomists, and Human Performance Researchers

For researchers and practitioners studying load carriage biomechanics, equipment design, or soldier injury risk, this study suggests Theia3D can track overall movement patterns and estimate muscle activation timing reliably even under heavy, bulky military loads, conditions that are specifically difficult for traditional marker-based systems due to marker occlusion. The authors are candid that absolute joint reaction force magnitudes differ between systems, particularly as load increases, so researchers using markerless-derived kinetic estimates for load carriage research should be aware of this systematic difference rather than treating markerless and marker-based joint loading values as interchangeable. The practical time savings from skipping marker placement and gap-filling could be especially valuable for field-based assessment of soldiers in different equipment configurations, a use case marker-based systems are poorly suited for.

The Role of Theia3D in Movement Analysis

Theia3D allowed the research team to capture lower limb kinematics in participants wearing heavy, bulky military equipment without needing to place individual markers underneath or around tactical gear, and without the marker occlusion problems that affected the marker-based comparison system when rucksacks and rifles blocked camera lines of sight to pelvis markers. The authors highlight that markerless capture proved to be a meaningful time-saving method in the laboratory, eliminating the need for marker application, labeling, and gap-filling, and suggest it could facilitate in-field assessment of soldiers’ biomechanics across different equipment configurations in ways marker-based systems cannot practically support.

Study Limitations and Future Directions

The authors identify several important limitations. Little is publicly disclosed about how Theia3D’s underlying model is constructed in Visual3D, which made it difficult to precisely match joint centers and segment centers of mass between the Theia3D-derived data and the Rajagopal musculoskeletal model used in OpenSim, introducing uncertainty into the kinematic and kinetic results that the authors recommend future research address directly. Ankle flexion RMSEs exceeded previously published minimal detectable change thresholds in some conditions, unlike hip and knee flexion, which the authors flag as needing further work to understand how markerless ankle data translates into musculoskeletal modeling software. Participants wore ankle-high military boots, requiring markers to be placed on top of the boots for the marker-based system, under the assumption that boot motion fully reflected foot motion, a modeling assumption that applies to both systems’ foot-tracking approach and could itself be a source of error independent of either camera system. Backpack and rifle loads were modeled as simple added mass to the torso and hand rather than accounting for inertial effects or relative movement between the body and equipment, an approach the authors note was applied equally to both systems so it should have limited effect on the between-system comparison itself.

Citation: Coll, I., Mavor, M.P., Karakolis, T., Graham, R.B., & Clouthier, A.L. (2024). Validation of markerless motion capture for soldier movement patterns assessment under varying body-borne loads. Annals of Biomedical Engineering. Advance online publication.

Link: https://doi.org/10.1007/s10439-024-03622-w

Could Markerless Capture Support Your Load Carriage Research or Soldier Readiness Program?

Assessing how equipment load affects movement and injury risk is hard enough in a lab, and nearly impossible with marker-based systems once bulky gear starts blocking camera views of markers. If your program needs to evaluate soldier or first-responder biomechanics under real equipment loads, in the lab or closer to the field, talk to our team about how markerless motion capture can support that work.

Recent Posts