Data analysis and memory methods for RSS Bluetooth low energy indoor positioning

Abstract

The thesis aims at finding a feasible solution to Bluetooth low energy indoor positioning (BLE-IP) including comprehensive data analysis of the received signal strength indication (RSSI) values. The data analysis of RSSI values was done to understand different factors influencing the RSSI values so as to gain better understanding of data generating process and to improve the data model. The positioning task is accomplished using a methodology called \textit{fingerprinting}. The fingerprinting based positioning involves two phases namely \textit{calibration phase} and \textit{localization phase}. The localization phase utilises the memory methods for positioning. In this thesis, we have used \textit{Gaussian process} for generation of radio maps and for localization we focus on memory methods: \textit{particle filters} and \textit{unscented Kalman filters}. The Gaussian process radio map is used as the measurement model in the Bayesian filtering context. The optimal fingerprinting phase parameters were determined and the filtering methods were evaluated in terms root mean square error.