Force-plate-free estimation of ground reaction forces and lower-limb joint moments during gait using a single RGB-D...

Three-dimensional (3D) gait analysis provides clinically relevant kinematics and kinetics but is limited by the high cost and extensive infrastructure of marker-based motion capture and force plates.

This study validates a hybrid markerless approach for gait analysis. The approach estimates 3D poses from a single RGB-D camera, recovers 3D ground reaction forces (GRFs) using physics-based optimization, and computes the lower limb joint moments using inverse dynamics without force plates.

The study involved 10 healthy adults who walked at a self-selected speed. A marker-based motion capture system and embedded force plates were employed to collect data simultaneously. Two successful trials were conducted with each participant. Signals were synchronized, resampled to 120 Hz, and time-normalized to 0-100% stance (101 points). Waveform agreements for right-limb GRFs and sagittal plane hip, knee, and ankle angles and moments were quantified using the root mean square error (RMSE), normalized RMSE (NRMSE), bias, Pearson's correlation (r), and Lin's concordance correlation coefficient (CCC).

Sagittal joint angles showed high correlations (hip: r = 0.94 ± 0.05, RMSE = 11.84 ± 4.53°; knee: r = 0.91 ± 0.08, RMSE = 5.42 ± 1.93°; ankle: r = 0.89 ± 0.08, RMSE = 5.42 ± 2.10°). Sagittal joint moments were also correlated (hip: r = 0.89 ± 0.06, RMSE = 0.50 ± 0.16 N·m/kg; knee: r = 0.73 ± 0.15, RMSE = 0.40 ± 0.14 N·m/kg; ankle: r = 0.95 ± 0.02, RMSE = 0.40 ± 0.09 N·m/kg). The GRF agreement was component-dependent: anterior-posterior (CCC = 0.91 ± 0.04, NRMSE = 12.31 ± 3.26%), medial-lateral (CCC = 0.37 ± 0.18, NRMSE = 36.43 ± 18.57%), and vertical (CCC = 0.80 ± 0.10, NRMSE = 14.05 ± 3.23%).

This single-camera pipeline showed feasibility for sagittal-plane gait analysis in healthy adults, but mediolateral GRF remains limited.