Prediction of Ground Reaction Forces during Walking with Depth Camera Sensors via Recurrent Neural Network

Abstract

In gait analysis, ground reaction forces (GRFs) provide important information about gait function assessment. However, due to the financial and time-consuming costs, the use of force plates for gait assessment in the rehabilitation field is limited. To solve these problems, previous studies have conducted to estimate the GRFs without the force plate. The purpose of this study is to estimate the 3-axis GRFs during walking from kinematic data of markerless motion capture system. For this purpose, we proposed a GRF estimation model using LSTM. The accuracy of the proposed model was verified by comparing with the estimation model using kinematic data obtained from marker-based motion capture system.The proposed model was found to have an estimation error of 20% compared to the measurements obtained from the force plates.