Human motion recognition through an adaptive fuzzy estimation of inertial sensing

This paper explores the possibility of an application based on a fuzzy estimator characterized by an accelerometer signal, to recognize simple movements. The sensor used is a tri-axial accelerometer located at the hip of the subject, with one of its axis perpendicular to the ground, and the signal from each axis will be used to train the estimator, meaning that each axis that´s being sample will generate one ANFIS, though, in the majority of any simple movement (in this case it will be walking, running and jumping), only one axis is needed because it contains the most representative signal and thus the other two axes can be omitted. Once trained, the signal of the motion can be compared to the estimation to see how close of a match it is to the original sample movement, and evaluate this difference.