Particle Filtering with Geospatial Analysis for Indoor Positioning

Abstract

Modern smartphones enable a wide range of people the use of location based services, such as personal navigation, in everyday life. The localisation in outdoor scenarios with smartphones is usually based on Global Navigation Satellite System (GNSS). However, the positioning with GNSS fails to provide sufficient and accurate measurements in an indoor environment. To enable the usage of location based services, for example the navigation at exhibition or airport areas or large university buildings, other indoor positioning approaches have been researched. Most of those can’t provide the needed accuracy or are expensive to install and maintain. A possible alternative is the usage of the Inertial Measurement Unit (IMU) of the smartphone using a pedestrian dead reckoning system Pedestrian Dead Reckoning (PDR). To correct the drift of the position estimate that results from noisy measurements, corrections through additional information is necessary. For this reason, two Particle Filter (PF) methods, a simple bootstrap PF and a backtracking PF, have been developed in this thesis. They both use building information from floor plans and routing edges, to improve the position estimate. The PF uses weighted particles to represent the probability density of the position estimate. The backtracking PF further stores the propagation history that can be used for further improvements of the position estimate. One novelty of the developed PF methods is the usage of geospatial analysis tools to derive information about the spatial relation between the particles and the geometries from the building information. The implementation of geospatial analysis tools further enables the use of geodata, which can be derived from building plans in Computer Aided Design software (CAD) format. The developed PF methods have been tested and compared with different methods for the weighting of particles regarding their positioning accuracy on two different test paths through the HafenCity University (HCU)-building in Hamburg. It was possible to achieve a position error of less than 3 m for a path through narrow corridors and less than 5.5 m for a path including wider spaces, such as hallways, 90 % of the time for some of the tested methods.