2011 International Conference on Indoor Positioning and Indoor Navigation (IPIN), 21-23 September 2011, Guimarães, Portugal Using Photographed Evacuation Plans to Support MEMS IMU Navigation Michael Peter, Norbert Haala, Dieter Fritsch Institute for Photogrammetry, University of Stuttgart, Germany Email: [firstname.lastname]@ifp.uni-stuttgart.de Abstract — In this paper, we present various procedures to support a ZUPT-based MEMS IMU navigation application by available external building models and photographs of evacuation plans. In detail, these are approaches for horizontal alignment of the track using the external building shell, for height correction by stair and elevator detection, for the extraction of the initial position and direction using the photographed plan and for the derivation of coarse indoor models from the plan. Keywords—IMU; Navigation; Modeling; Evacuation Plan I. INTRODUCTION In the past years, an increasing scientific interest in indoor navigation has been observed. While in outdoor scenarios GPS exists as prevalent and sufficiently accurate source for position information, MEMS IMUs just started to emerge as corresponding positioning devices for building interiors. These IMUs are often used as foot- mounted systems and are combined with algorithms like ZUPT (zero velocity updates) [1]. However, they deliver only coordinates relative to a starting point and an initial direction. In most scenarios, this limitation is overcome by a GPS-guided user navigating to an entrance, followed by a handover to inertial navigation. Another disadvantage of this positioning approach, despite the need to use ZUPTs, is the existence of drift errors. In most scenarios these are resolved by map matching algorithms using given high Figure 1. Evacuation plan photographed using a mobile phone camera (a: floor, b: north direction, c: legend, d: address) precision indoor models (e.g. [2]). However, while 3D city models provide the external contours of many buildings to end users, this is not available for high precision models of building interiors. Nevertheless, information on indoor environments is frequently accessible from evacuation plans, which are compulsory for public buildings in a number of countries (see figure 1). As it will be demonstrated within the paper, indoor navigation can be improved considerably, if map-like information from the interpretation of such an evacuation plan is integrated in the processing pipeline. In this paper, we present various procedures to support foot mounted MEMS IMU navigation by a given external building shell and photographed evacuation plans. In section II, we describe a basic approach to align IMU tracks horizontally using both the principal directions of the outer building shell and the assumption that pedestrian movement in building interiors usually will be parallel to one of these directions. Secondly, height correction using stair and elevator detection is also demonstrated in this section. Section III is used to explain how a photographed evacuation plan may be employed to derive the initial position and direction for IMU navigation. In section IV, we show that these plans may furthermore be used to reconstruct coarse models of building interiors. These models then may serve as a basis for map matching or for the acquisition of more detailed models, e.g. using a modeling strategy similar to the one employed by OpenStreetMap. II. INDOOR NAVIGATION USING MEMS IMU A. Zero Velocity Updates In our indoor navigation scenario, we use a foot mounted XSens MTi-G MEMS IMU as positioning sensor. To compensate the system immanent drift errors, we use the well-known zero velocity updates [1]. In short, the accelerometer measurements are integrated once and the resulting velocity values are supposed to be zero during a stance phase detected using the gyro measurements. However, due to measurement errors, these values differ from zero. The offset may be used to correct the errors which occurred since the last stance phase. The resulting coordinates of every second step (as only one foot is equipped with a sensor) may then be computed by a second integration. B. Alignment Using Building Model While ZUPTs significantly reduce the drift errors in comparison to naïve double integration, especially long tracks that are not supported by GPS measurements or other fixed points still suffer from drift effects.