Map-Aided Indoor Navigation

Kurzfassung

Maps representing aspects of an environment that affect pedestrian motion can be very informative sources of data in indoor localization. Their proper representation and usage is mandatory to fully leverage their potential. In this chapter, we show how probabilistic representations facilitate accuracy and availability of position estimates even in the absence of usable satellite navigation signals or similar forms of localization signals. We will show that maps may effectively substitute infrastructure, such as active or passive (RFID-type) radio beacons when their information is properly used in combination with dynamic models of movement and some form of motion estimate such as pedestrian dead reckoning. This chapter aims at illuminating the details of how to generate, represent and use probabilistic maps for indoor localization. While this discussion applies to a wide range of sensors, we will focus on showing how maps are essential in achieving long-term stability in combination with inertial sensors. We begin by motivating why the use of a probabilistic map of human motion is a natural way of incorporating building information into a sequential Bayesian filtering framework. This stands in contrast to the often used ad-hoc solutions which is to use a floor plan as a “kill or live” weighting function in a particle filter (PF), driven by some form of pedestrian dead reckoning (PDR) such as foot mounted inertial sensors based PDR. We show how the latter method can fail catastrophically and how a probabilistic map formulation addresses these problems. We present a number of ways of how to obtain such maps for real world applications. The first is based on knowledge of the building layout and applies a diffusion algorithm to compute an estimate of the probability distribution of the motion direction of a pedestrian at each point in the building. Secondly, we compare these maps with those obtained using Simultaneous Localization and Mapping (SLAM) by applying FootSLAM that requires no sensors other than a source of dead reckoning. The map concept can be further extended in order to include features that are relevant to radio-based localization techniques, like transmitter positions and a model for radio propagation or, eventually, a database of fingerprints. The influence of the different kind of maps on positioning accuracy is discussed in detail and the maps are compared to each other by means of metrics derived from information theory.