Probabilistic Estimation of Posture Metrics Using Novel Loadsols

Abstract This paper presents an original technique for estimating human posture metrics using Novel Loadsols®. Under the proposed technique, center of pressure (COP) metrics are derived by combining physics- and data-driven estimates to achieve reasonably high accuracy at relatively low cost. To develop a training set upon which the probabilistic data model was constructed, 79 trials were conducted in which participants stood comfortably still for 30 seconds at a time simultaneously on a force plate and a pair of Loadsols, where the force plate is considered to be the gold-standard of COP measurement. These data were then used to generate Gaussian mixture models (GMMs) of pairwise combinations of force plate and Loadsol metrics. The GMMs can then be conditioned on Loadsol measurements and fused using Bayesian inference. When the training set was re-processed by converting 12 Loadsol metrics into estimated force plate metrics, it was found that the converted metrics matched ground-truth more accurately on average than raw Loadsol metrics. Furthermore, there was improvement in the r2 values of the regression lines after conversion for 75% of the metrics. Given some experiment and algorithm refinement, the proposed probabilistic approach has potential to offer the accuracy of force plate COP estimation at a fraction of the cost.