Wearable motion sensors and artificial neural network for the estimation of vertical ground reaction forces in running

Abstract

Biomechanical load assessments are becoming increasingly important in the sporting community; however, there are still numerous difficulties in monitoring them in a field environment outside of specialized biomechanical monitoring laboratories. Inertial Measurements Units (IMUs) have been showing promising results in the modeling of biomechanical variables. This study explores the application of an artificial neural network (ANN) in the estimation of runners' vertical ground reaction forces (GRFs) based on the accelerometry collected from two wearable motion sensors developed in-house and attached on the shanks. Data collected from fourteen runners running at three different speeds (8, 10, 12 km/h) were used to train and validate the ANN. Predictions were compared against gold-standard measurements from a pair of pressure in-soles. Root mean square error (RMSE) was used to evaluate the performance of the models. Further investigations, e.g., the use of principal components analysis (PCA) and the impact on the estimation of several GRF-related variables, were carried out to provide useful insights regarding the portability of the model to low-power resource-constrained devices. Findings indicate that ANNs in conjunction with accelerometry may be used to compute vertical ground reaction forces (RMSE: 0.148 BW) and related loading metrics in running accurately.