Estimation of Three-Dimensional Ground Reaction Force and Center of Pressure During Walking Using a Machine-Learning-Based Markerless Motion Capture System.

OBJECTIVE

We developed two neural network models to estimate the three-dimensional ground reaction force (GRF) and center of pressure (COP) based on marker trajectories obtained from a markerless motion capture system.

METHODS

Gait data were collected using two cameras and three force plates. Each gait dataset contained kinematic data and kinetic data from the stance phase. A multi-layer perceptron (MLP) and convolutional neural network (CNN) were constructed to estimate each component of GRF and COP based on the three-dimensional trajectories of the markers. A total of 100 samples were randomly selected as the test set, and the estimation performance was evaluated using the correlation coefficient (r) and relative root mean square error (rRMSE).

RESULTS

The r-values for MLP in each GRF component ranged from 0.918 to 0.989, with rRMSEs between 5.06% and 12.08%. The r-values for CNN in each GRF component ranged from 0.956 to 0.988, with rRMSEs between 6.03-9.44%. For the COP estimation, the r-values for MLP ranged from 0.727 to 0.982, with rRMSEs between 6.43% and 27.64%, while the r-values for CNN ranged from 0.896 to 0.977, with rRMSEs between 6.41% and 7.90%.

CONCLUSIONS

It is possible to estimate GRF and COP from markerless motion capture data. This approach provides an alternative method for measuring kinetic parameters without force plates during gait analysis.