Multilayer perceptron for gait type classification based on inertial sensors data

Abstract

In this study, the gait type classification process is considered. Input data are obtained by using the inertial sensors tracking system. Three types of gait are recorded: normal walk, tiptoeing and walk retaining long stance phase. Two data set types, describing the registered motion, are prepared. The most significant input features are selected by means of the sensitivity analysis (SA) procedure. The classification process is conducted using multilayer perceptron (MLP) with various structures. The classification accuracy of the network is computed with the use of a cross validation procedure. The obtained results show that the successful classification of presented gait types can be achieved using relatively simple MLP architecture.