Indoor Radio Map Construction Based on Crowdsourced Fingerprint Splitting and Fitting

Abstract

Recently, crowdsourcing has been promoted as a promising solution for fingerprint-based indoor localization systems. However, when using crowdsourced fingerprints to construct an indoor radio map, the following challenges should be first addressed: inaccurate location information, nonuniform spatial density, fingerprint dimension diversity and collection device diversity. In this paper, we propose a radio map construction scheme based on crowdsourced fingerprint splitting and fitting to deal with the first three challenges. Our objective is to construct a grid radio map from crowdsourced fingerprints. We propose to split a crowdsourced fingerprint into several neighboring grid cells with some probability. For each grid cell, we propose to first filter out some unimportant features when composing a grid cell fingerprint. Furthermore, for a grid cell with very few or non supporting crowdsourced fingerprints, we propose to construct its fingerprint by using the surface fitting technique. We conduct field measurements and experiments to examine the localization performance. Results show that the proposed scheme can achieve 1.48m mean localization error and is more robustness to location annotation errors.