Assessment of spatiotemporal gait parameters using a deep learning algorithm-based markerless motion capture system

Abstract

Spatiotemporal parameters can characterize the gait patterns of individuals, allowing assessment of their health status and detection of clinically meaningful changes in their gait. Video-based markerless motion capture is a user-friendly, inexpensive, and widely applicable technology that could reduce the barriers to measuring spatiotemporal gait parameters in clinical and more diverse settings. The aim of this work was to determine whether spatiotemporal gait parameters measured using Theia3D markerless motion capture demonstrate concurrent validity with those measured using marker-based motion capture. Thirty healthy adult participants performed treadmill walking at self-selected speeds while 2D video and marker-based motion capture data were collected simultaneously. Kinematic-based gait events were used to measure nine spatiotemporal gait parameters from both systems independently. The parameters were compared using their group means, Bland-Altman methods, Pearson correlation coefficients, paired-samples t-tests, and intraclass correlation coefficients (ICC(A-1) and ICC(C-1)). Group means between systems were indistinguishable across all nine gait parameters, and the Bland-Altman plots showed no systematic biases or clinically meaningful differences between the systems. Pearson coefficients indicated near-perfect correlations (r ≥0.96) between systems for all but two parameters, which had strong correlations (double-limb support time, r=0.87; swing time, r=0.88). T-tests indicated differences in stance, swing, and double-limb support time parameters, but strong correlations were found for all ICCs, the lowest being 0.84 for both double-limb support time and swing time. The measurements made by the Theia3D markerless motion capture system demonstrated concurrent validity with those from the marker-based system, indicating sufficient accuracy for research, clinical, and other uses.

Concurrent assessment of gait kinematics using marker-based and markerless motion capture

“…Theia3D (Theia Markerless Inc., Kingston, ON, Canada) is a deep learning algorithm-based approach to markerless motion capture which uses deep convolutional neural networks for feature recognition (humans and human features) within 2D camera views (Kanko et al, 2020a(Kanko et al, , 2020bMathis and Mathis, 2020). The neural networks were trained on digital images of over 500,000 humans in the wild.…”

“…By default, the lower body kinematic chain has six degrees-of-freedom (DOF) at the pelvis, and three DOF at the hip, knee, and ankle. This markerless system has been shown to accurately measure spatiotemporal gait parameters, and measure gait kinematics over multiple sessions with greater reliability than marker-based motion capture (Kanko et al, 2020a(Kanko et al, , 2020b).…”

A review of 3D human pose estimation algorithms for markerless motion capture

“…This explains that it took time for the scientific community to create large benchmarks : today a lot of variations exist between monocular vs stereo multi-view, laboratory controlled vs in the wild environments (see Table 1). With new commercial solutions (ie: Theia, The Captury) starting to produce results that are similar to traditional motion capture Kanko et al (2020), some benchmarks are also starting to use markerless labeled ground truths. This as the advantage of easily providing in-the-wild images.…”