Oikonomou Filandras, PA; (2016) Probabilistic graphical models for mobile pedestrian localization in 3D environments. Doctoral thesis , UCL (University College London).

Preview

Text phd-thesis-final.pdf - Accepted Version Download (1MB) | Preview

Abstract

This PhD thesis considers the problem of locating wireless nodes in indoors GPS-denied environments using probabilistic graphical models. Time-of-arrival (ToA) distance observations are assumed with Non-Line-of-Sight (NLoS) communications and a lack of adequate anchors. As a solution cooperative localization is developed using Probabilistic Graphical Models (PGMs). The nodes infer their position in an iterative message-passing algorithm, in a distributed manner, given a set of noisy distance observations and a few anchors. The focus of this thesis is to develop algorithms that decrease computational complexity, while maintaining or improving accuracy. Firstly, we develop the Hybrid Ellipsoid Variational Algorithm (HEVA), which extends probabilistic inference in 3D localization, combining NLoS mitigation for ToA. Simulation results illustrate that HEVA significantly outperforms traditional Non-parametric Belief Propagation (NBP) methods in localization while requires only 50% of their complexity. In addition, we present a novel parametric for Belief Propagation (BP) algorithm. The proposed Grid Belief Propagation (Grid-BP) approach allows extremely fast calculations and works nicely with existing grid-based coordinate systems, e.g. NATO military grid reference system (MGRS). This allows localization using a Global Coordinate System (GCS). Simulation results demonstrate that Grid-BP achieves similar accuracy at much reduced complexity when compared to common techniques. We also present an algorithm that combines Inertial Navigation System (INS) and Pedestrian Dead Reckoning (PDR), namely Probabilistic Hybrid INS/PDR Mobility Tracking Algorithm (PHIMTA), which provides high accuracy tracking for mobile nodes. We combine it with Grid-BP and stop-and-go (SnG) algorithms, showcasing improved accuracy, at very low computational cost. Finally, we present Stochastic Residual Belief Propagation (SR-BP). SR-BP extends the use of Residual Belief Propagation (R-BP) to distributed networks, improving the accuracy, convergence rate, and communication cost. We prove SR-BP convergence to a unique fixed point under conditions similar to those ensuring convergence of asynchronous BP. Finally, numerical results showcase the improvements in convergence speed, message overhead and detection accuracy of SR-BP.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as XMLJSONCSV

Downloads by country - last 12 months

Export as XMLCSVJSON

Archive Staff Only

View Item