Seamless indoor pedestrian tracking by fusing INS and UWB measurements via LS-SVM assisted UFIR filter

Abstract

A seamless indoor pedestrian tracking scheme using a least square-support vector machine (LS-SVM) assisted unbiased finite impulse response (UFIR) filter is designed to achieve seamless reliable human position monitoring in indoor environments in this work. This novel scheme is based on the loosely-coupled integrated localization model, which can fuse the inertial navigation system (INS)-derived and ultra-wide-band (UWB)-derived positions and compensate for the INS position error. Based on the loosely-coupled model, the hybrid scheme includes a training stage and a predict stage. In the training stage, the UWB position is available, and the scheme employs a UFIR filter to compensate for the INS position error and provide training the data robustly. Meanwhile, the LS-SVM is used for training the mapping between the INS position and its error utilizing the INS position and UFIR filter outputs. When the UFIR filter can not work due to a UWB outage, the hybrid scheme is in the prediction stage; the LS-SVM replaces the UFIR filter to compensate for the INS position error with the mapping built in the training stage. An experimental study shows that the proposed scheme is capable of seamless reliable indoor pedestrian tracking.