Gait Identification using Smartphone Handheld Placement with Linear Interpolation Factor, Single Magnitude and One-vs-One Classifier Mapping

Abstract

Gait signal of a person can be easily obtained using a smartphone sensor. To get the source of the signal, the smartphone need to be placed in the pocket, pouch or attached to other parts of the body. In the real world application, it is hard to place the device on the mentioned position. The easiest way is to put it on hand. In another issue, the single magnitude is known in the use of multiple orientations. However, this method may discard useful features for machine learning classification. Another problem is that the signal captured using a smartphone is not in a fix sampling rate and in the small distance, hence interpolation needs to be applied so that the sampling can be in a fix sequence with more fix point data. However, too much application of interpolation may result in low prediction rate. Finally, a multiclass dataset may contain overlapped class boundary which produces low accuracy on a single classifier mapping. In this paper, hand based smartphone placement position is implemented and evaluated. Single magnitude application is also evaluated in representing multiple positions of a person into one signal. Besides that, the linear interpolation factor is introduced in sampling the signal. Lastly, OvO classification model is implemented in binarizing the multiclass gait dataset. From the experiment, it shows that using the mentioned method do produce satisfactory result hence opening a new gateway in a better gait identification/recognition system.