Unscented Kalman Filter based finger tracking utilising magnetoresistive sensors

Abstract

In this paper we utilise magnetoresistive sensors and an Unscented Kalman Filter to develop a finger tracking system for physical therapy and human-machine interaction. The aim of this finger tracking system is to provide a light weight apparatus with a prolonged operational period. The finger tracking algorithm utilises two-dimensional measurements of each joint relative to a magnetic field reference point to obtain an accurate model of the finger position as opposed to other light weight systems that provide information in only one dimension. The nonlinear interaction between joint movements and sensor measurement requires a nonlinear filtering technique, as such, a Kalman filter that utilises an unscented transformation was utilised.