Predicting ground reaction forces from 2D video: Bridging the lab to field nexus

The accurate measurement of ground reaction forces (GRFs) is confined to laboratory- based settings, posing an ongoing problem for sports biomechanists working in the field given the necessity of this variable for a variety of biomechanical modelling approaches. The ability to remote estimate on-field GRFs could facilitate the development of a tool that could positively impact the side-line management of athletes during match play. Using laboratory collected GRF side stepping and running data, alongside concurrently collected two-dimensional (2D) video, the aim of this study was to use a least squares estimator (LSE) matrix to estimate GRFs from 2D video. Results of r>0.8 were found for the vertical and horizontal GRF components which was slightly lower than the r>0.9 observed for the higher complexity convolutional neural network (CNN) approach which was used as a comparator model. These results provide early support for the efficacy of remote on-field estimation of GRFs determined from 2D video footage in isolation.