Continuous classification of spatio-temporal data streams using liquid state machines

Abstract

This paper proposes to use a Liquid State Machine (LSM) to classify inertial sensor data collected from horse riders into activities of interest. LSM was shown to be an effective classifier for spatio-temporal data and efficient hardware implementations on custom chips have been presented in literature that would enable relative easy integration into wearable technologies. We explore here the general method of applying LSM technology to domain constrained activity recognition using a synthetic data set. The aim of this study is to provide a proof of concept illustrating the applicability of LSM for the chosen problem domain. © 2012 Springer-Verlag.