Sensor fusion using particle filtering in the scenario of indoor-localization

M.Sc. Lukas Köping, University of Applied Sciences Würzburg

Date: 12/02/2013
Time: 13:00-14:00
Room: H-F 016

Localization in outdoor areas with accuracy of a few meters can mostly be achieved using satellite systems like GPS. On contrary, due to technical limitations this technique is not applicable in the inner of buildings. Instead different approaches in the field of indoor-localization have to be taken which are still in the focus of recent research. One widely used approach is to redefine the position finding problem as state estimation problem. Generally speaking, given some observations at different time steps the goal is to estimate a probability density for the non-directly observable state of interest. A well-known technique to approximate such densities is particle filtering. Thereby observations are typically represented by sensor data of all forms. Since only in rare cases a single sensor supplies enough information to allow an accurate estimation the combination of different sensor outputs in form of a sensor data fusion process often helps to overcome this problem. The aim of this approach can be two-fold: On the one hand a single sensor can only deliver sensor type-specific data, e.g. an accelerometer delivers only data in form of acceleration for the three-dimensions. Almost never this data on its own is sufficient to make any useful assumptions. Using additional sensors of different types expands the quantity of specific information which can be used to fill the lack of information. On the other hand sensor fusion can lead to more accurate data if sensors of the same type are used redundantly and their outputs are combined. In my talk I will give a brief overview over the thematic field of state estimation and sensor fusion as well as the research issues that occur when using such techniques and which are part of my upcoming work.