WI-FI-BASED INDOOR LOCALIZATION USING MODEL-BASED AND DATA-DRIVEN APPROACHES

Abstract

This thesis investigates model-based and data-driven approaches for indoor localization using the Received Signal Strength Indicator (RSSI) of Wi-Fi signals. We study multiple model-based indoor localization approaches, including the free space path loss model, the log-distance path loss model, the International Telecommunication Union (ITU) model, and a nonlinear regression model. We examine their indoor localization accuracy using raw RSSI values, and filter RSSI values passed through a Moving Average filter and a Kalman filter. For data driven approaches, we employ a family of Extreme Learning Machine (ELM) algorithms including Basic-ELM, Online Sequential-ELM (OS-ELM), Hierarchical-ELM (H-ELM), and Kernel-ELM (K-ELM), to find the indoor position. We provide simulation results comparing the performances of both the Machine-learning based approaches and model-based approaches in terms of localization error to identify the algorithms with the lowest localization error.