Validity and reliability of monocular 3D markerless gait analysis in simulated pathological gait: A comparative study with OpenCap

Abstract

Recent advances in markerless 3D motion capture raise hopes of making gait analysis more accessible and affordable. While tools like OpenCap.ai require at least two smartphones, emerging monocular approaches allow full-body 3D pose estimation from a single camera. This study evaluated the concurrent validity and test–retest reliability of a low-cost monocular markerless system (CameraHMR) based on 3D skinned multi-person linear models (SMPL) in comparison to a marker-based reference system and a two-camera OpenCap setup. We used videos and marker-based motion capture data from two previous studies on OpenCap’s validity and reliability. The validity dataset included 19 healthy participants instructed to walk with four gait patterns (physiological, crouch, circumduction, equinus), while 3D kinematics were recorded simultaneously using OpenCap and a marker-based system. The reliability dataset included 19 participants who performed physiological walking on two separate days, recorded only with OpenCap. CameraHMR was applied to single-view OpenCap videos, and 3D joint kinematics were extracted using the SMPL model and OpenSim’s inverse kinematics tool. Validity was assessed via waveform root mean square deviation (RMSD) against the marker-based system; reliability was assessed using RMSD and standard error of measurement (SeM) across sessions. Overall, CameraHMR’s performance was comparable to OpenCap (p $>$ 0.05). Despite challenges in tracking ankle joint kinematics, monocular 3D gait analysis showed promising reliability (RMSD: 3.0 $\pm$ 1.0 degrees) and reasonable kinematic accuracy (RMSD: 5.5 $\pm$ 1.1 degrees). These findings suggest that single-camera systems could enable broader, low-cost access to gait assessment, including in remote or home-based settings. However, further refinement is needed to reach clinically acceptable accuracy thresholds.