Semi-supervised Fingerprint Construction and Localization System For Large Indoor Area

Abstract

Many applications of location-based indoor navigation services require precise location information of a user. While Global Positioning System (GPS) loses reliability indoors, fingerprints-based localization technology (FBLT) embodies superiority regarding accuracy and robustness. In a Bluetooth-based fingerprint localization system, a radio map is constructed offline and used as a reference for subsequent real-time localization tasks. However, the quality of the fingerprint radio map could be problematic when it comes to a large, broad space with low beacon density. Data collection in such a space could be exhausting as well. Another main issue is that different mobile devices receive heterogeneous signal strength at the same location. In this article, we propose a highly practical localization system with a semi-supervised learning fingerprints construction method that provides an efficient solution for a large-scale localization system in a complex indoor environment. We also conducted a series of experiments to evaluate the performance of this system.