Indoor Location Services through Multi-Source Learning-based Radio Fingerprinting Techniques

Abstract

Proximity advertising, smart parking and tourism are just examples of use-cases of location-based services that have become extremely popular in the last few years, also thanks to the pervasive diffusion of GNSS-enabled mobile devices. These devices, however, are not able guarantee adequate accuracy in indoor scenarios, that represent the actual frontier of next-generation location-based services. To this aim, we present in this paper Wireless Locator (WI-LO), a novel framework for the indoor localization of smartphone devices and the automation of location-based tasks. Through the WI-LO Web portal, users can import an indoor planimetry, set the Reference Points (RPs), and define the actions to execute at each RP or region or RPs. The WI-LO localization engine implements hybrid Radio Finger-Printing (RF) techniques, and it leverages on a variety of sensors embedded in Commercial Off The Shelf (COTS) smartphones (Wi-Fi, BLE, LTE, magnetometer). We investigate the utilization of Machine Learning (ML) techniques for the processing of the radio fingerprints of each source, and the application of fusion policies in order to aggregate the hard-decisions of each source. The evaluation analysis, conducted at the DISI@UNIBO department, confirms the ability of the WI-LO platform to deliver geo-fencing messages with over 90% accuracy, and it investigates the impact of different ML techniques, application parameters and scenario settings on the overall localization performance.