Gait phase detection optimization based on variational bayesian inference of feedback sensor signal

Stroke patients often suffer from gait disorders which can remain chronic. Mechanical or electrical aids designed to deal with this problem often rely on accurate estimation of current gait phase as this information is used for active ankle joint control. In this paper we present the method for optimization of the gait phase detection algorithm. The method is based on Variational Bayesian inference which is employed on signals from feedback sensors positioned on both paretic and healthy foot of patient. Main aim of Variational Bayesian inference application was to remove noise and provide smooth sensor signal which is suitable for robust gait phase detection algorithm. We modeled foot trajectory with linear model. Results presented in this paper show significant reduction of high frequency noise in gyroscope signal. The reduction was dominant during transitions between gait phases making our method applicable in any algorithm based on signal features in time domain.