Advanced Inertial/GNSS Sensor Fusion for Smart Devices Using Factor Graph Optimization

Vista sencilla de ítem
dc.contributor.advisorDíez Blanco, Luis Enrique
dc.contributor.authorHussain, Amjad
dc.date.accessioned2026-07-23T07:46:12Z
dc.date.available2026-07-23T07:46:12Z
dc.date.issued2026-02-27
dc.description.abstractPedestrian navigation on smartphones mainly depends on Global Navigation Satellite Systems (GNSS) due to their accessibility and widespread adoption. However, GNSS performance deteriorates in challenging environments such as urban canyons, tunnels, under trees, and covered or partially obstructed environments, where multipath, non-line-of-sight (NLOS) effects, and atmospheric disturbances, combined with limitations of low-cost smartphone GNSS receivers, lead to significant accuracy degradation. To overcome these limitations, this thesis first presents a systematic review of inertial and GNSS fusion methods implemented on smartphones, providing contextual background for pedestrian positioning and identifying the strengths and shortcomings of existing approaches. Building on these insights, the thesis investigates the fusion of GNSS with Pedestrian Dead Reckoning (PDR) using Factor Graph Optimization (FGO) as a robust alternative to classical Kalman Filter (KF)-based fusion. Unlike strapdown inertial navigation with low-cost MEMS, PDR offers slower drift growth and better resilience to smartphone carrying-mode variations, making it wellsuited for infrastructure-free pedestrian navigation. Moreover, while PDR+GNSS integration has been studied using traditional methods, its fusion through FGO remains relatively underexplored in the literature, highlighting a research gap this thesis aims to address. The first case study evaluates FGO against KF in real-world walking and running trials, demonstrating that FGO consistently improves horizontal accuracy by 30.6% in walking and 10% in running while offering greater robustness in GNSS-degraded conditions. In addition, this case study investigates FGO’s resilience to two typical PDR error types, Short and High Errors (SHE) and Continuous and Low Errors (CLE) caused by gait variability and carrying-mode changes. Tests with ten diverse participants show that FGO outperforms both KF and a Smoothed KF (SKF), reducing mean error by about 25% compared to KF (24% compared to SKF) and enabling faster recovery after transient disturbances. The second case study addresses FGO’s computational cost by introducing two sliding-window strategies: Naïve Windowing FGO (NW-FGO) and Marginalized Windowing FGO (MW-FGO). Evaluations under GNSS multipath and PDR error conditions identify optimal window lengths (50 s for GNSS and 30 s for PDR) that preserve Batch FGO accuracy while reducing processing time from 2500 s to 38 s using MW-FGO. Marginalization consistently outperforms naïve windowing, improving mean positional accuracy by 2.18 % across tested window sizes. The third case study examines the impact of GNSS quality on fusion accuracy by combining PDR with GNSS Single-Point Positioning (SPP) and Post-Processed iv Kinematic (PPK) solutions through both FGO and KF. Experiments with a Google Pixel 4 along a 1.8 km semi-obstructed path show that integrating PPK within FGO reduces the 2D mean error by 22.5% compared to FGO with SPP (3.29 m vs. 4.25 m). FGO consistently outperforms KF under both SPP and PPK fusion strategies, although smartphone-based PPK remains at the meter level due to low ambiguity resolution. Finally, the results highlight the high computational cost of batch FGO, suggesting the need for more efficient windowed formulations for real-time applications. In summary, this thesis advances smartphone-based pedestrian localization by (1) systematically reviewing inertial+GNSS fusion methods, (2) introducing and validating FGO as a high-accuracy alternative to KF-based fusion, and (3) developing computationally efficient windowing strategies for real-time deployment and (4) assessing the impact of higher-quality GNSS processing (PPK) when fused with PDR through FGO. The results demonstrate that FGO-based fusion delivers robust, accurate, and efficient pedestrian positioning in GNSS-challenged environments, offering practical pathways to reliable, infrastructure-free navigation on mobile and wearable platforms.en
dc.identifier.urihttps://hdl.handle.net/20.500.14454/6422
dc.language.isoeng
dc.publisherUniversidad de Deusto
dc.subjectMatemáticas
dc.subjectCiencia de los ordenadores
dc.subjectSistemas de navegación y telemetría del espacio
dc.subjectCiencias de la tierra y del espacio
dc.subjectGeodesia
dc.subjectGeodesia de satélites
dc.subjectCiencias tecnológicas
dc.subjectTecnología electrónica
dc.titleAdvanced Inertial/GNSS Sensor Fusion for Smart Devices Using Factor Graph Optimizationen
dc.typedoctoral thesis
Archivos
Bloque original
Mostrando 1 - 1 de 1
Miniatura
Nombre:
Hussain_advanced_2026.pdf
Tamaño:
23.4 MB
Formato:
Adobe Portable Document Format
Descargar
Colecciones
Tesis doctorales