WiFi Fingerprinting based Floor Detection with Hierarchical Extreme Learning Machine

Abstract

The indoor location-based services are high demand in the market, and precise location estimation in multi-floor buildings has received significant attention in recent years. In these environments, the absolute floor recognition is a precondition for accurate positioning. In this article, to floor determination based on the WiFi-fingerprinting technique, the hierarchical structure of extreme learning machine (H-ELM) is exploited. This deep architecture of ELM comprises of two sections: the multilayer feature encoding with unsupervised learning (ELM-sparse-autoencoder) and the supervised multiclass classification (original ELM). Floor identification using H-ELM can be more accurate than traditional ELM. For evaluating the proposed method, we utilize TI building data available in the public UJIIndoorLoc dataset. As indicated by our simulation results, using the proposed WiFi-fingerprint based floor detection system can achieve a more accurate hit rate than other state-of-the-art techniques.