Data fusion by SVMs
Translated title
Cooperation Mobile Positioning
Author
LAURENT, Philippe
Term
10. term
Publication year
2007
Pages
61
Abstract
Location-based services require accurate device positioning indoors and outdoors. This thesis studies cooperative mobile positioning by fusing wireless signal-strength measurements (RSSI) from WiFi access points using Support Vector Machines (SVMs). Unlike prior RSSI-based approaches relying on probabilistic regression, the method frames localization as a classification task. The work considers two realistic WLAN scenarios: an outdoor line-of-sight setting with access points and mobile stations, and an indoor setting with both line-of-sight and non-line-of-sight propagation, optionally including cooperating computers. Algorithms implemented include Lagrangian SVM, Maximum Response, Cooperative SVM (CSVM), and scenario-specific Outdoor and Indoor Cooperative Algorithms (OCA, ICA). Simulations and SVM processing are performed in MATLAB and Borland C++. Reported results indicate that cooperation improves position estimates, yielding up to about 70% correct predictions indoors (60% outdoors) when computers and mobile stations share information. The thesis also discusses protocol, interference, and data management considerations, and outlines how further cooperative techniques could enhance performance.
Lokationsbaserede tjenester kræver nøjagtig positionering af enheder både inde og ude. Dette speciale undersøger kooperativ mobilpositionering ved at fusionere trådløse signalstyrkemålinger (RSSI) fra WiFi-adgangspunkter med Support Vector Machines (SVM). I stedet for tidligere RSSI-baserede metoder med probabilistisk regression behandles lokalisering som en klassifikationsopgave. Arbejdet omfatter to realistiske WLAN-scenarier: et udendørs line-of-sight-miljø med adgangspunkter og mobile enheder samt et indendørs miljø med både line-of-sight og non-line-of-sight udbredelse, med mulighed for samarbejde med computere. Implementerede algoritmer omfatter Lagrangian SVM, Maximum Response, Cooperative SVM (CSVM) og scenariespecifikke Outdoor og Indoor Cooperative Algorithms (OCA, ICA). Simulationer og SVM-behandling udføres i MATLAB og Borland C++. Resultaterne viser, at samarbejde forbedrer positionsestimeringerne, med op til cirka 70% korrekte forudsigelser indendørs (60% udendørs), når computere og mobile enheder deler information. Specialet diskuterer også protokol-, interferens- og datastyringsovervejelser og skitserer, hvordan yderligere kooperative teknikker kan forbedre ydeevnen.
[This apstract has been generated with the help of AI directly from the project full text]