Change-Point Detection in Kinetic Signals

Abstract

A method to precisely determine the onset of voluntary discrete movements in kinetic signals (e.g. joint angle) is presented. The movement onset is identified as an abrupt change in the (time varying) parameters of a statistical process model. An adaptive Kalman whitening filter transforms the digitized kinetic signal into a sequence of innovations which is examined for possible change-points by a generalized log-likelihood-ratio test. The accuracy of the algorithm is assessed by statistical simulations and compared to the accuracy of a standard threshold criterion. Results show that the method provides accurate change time estimates even for weak and highly variable response profiles.