Seamless Indoor Outdoor Positioning Using Bayesian Sensor Data Fusion on Mobile and Embedded Devices

G. Hejc, J. Seitz, J. Gutierrez Boronat, T. Vaupel

Abstract:	Indoor positioning requires a combination of sensor data and other information in order to work accurately and reliably. The fusion approach proposed in this paper uses Wi-Fi fingerprinting, Global Navigation Satellite System (GNSS) and dead reckoning based on inertial sensors as inputs to a particle filter. While GNSS receivers are not sensitive enough to determine the position in indoor environments, Wi-Fi fingerprinting closes the gap by measuring the received signal strength of existing Wi-Fi access point beacons and correlating these measurements with a database of recorded fingerprints. The availability of inertial sensors in mobile phones and embedded platforms provide additional information on the dynamical state, which can be used to predict the trajectory from a known starting position. This process is called dead reckoning and can be used to improve the quality of position estimates, because especially Wi-Fi fingerprinting position estimates can have ambiguities, which can be resolved with this method. The bias instabilities of these sensors must be compensated, which is automatically done within the particle filter as soon as absolute position updates from GNSS or Wi-Fi are available. The advantage of the particle filter is its ability to work with non-linear dynamical models and non-Gaussian probability distributions. The computational power of mobile and embedded devices today allows integrating more sophisticated filtering approaches than in the past. Particle filter use Monte Carlo simulation and propagate a set of particles through the state space based on a state evolution model and regularly corrected by measurements. It is essential for the design of meaningful and realistic measurement models that the individual sensor inputs are carefully characterized in terms of their performance. This is in particular difficult on mobile phones where a variety of different sensors, receivers and components are used and the application has no a priori knowledge and only can learn about the quality of sensor data at run time. Furthermore the algorithms must be robust against unevenly sampled data, where it is typically difficult to infer if the measured values are taken at non-equal times or only delivered as such. While mobile phones have no access to pseudorange measurements for each satellite, there are embedded platforms available with an interface to GNSS raw data e.g. miniLOK. With this information a better model of the measurement process in terms of a likelihood function can be established. This allows a tighter coupling of GNSS to Wi-Fi fingerprinting. The algorithms were implemented in Java to allow an easy deployment to various hardware platforms: a commercial Android smartphone and an ARM-processor based development platform with Linux as operating system, the miniLOK board. A typical everyday life scenario was used to evaluate the system: a user drives his car in a parking garage, leaves the car and enters an office building through the staircase, walks to the targeted room, goes back to the car and leaves the parking garage again. The scenario requires that the software distinguishes between car and pedestrian movement and applies the proper process model for each situation. The pedestrian motion model uses a step counter together with a step length estimator based on acceleration sensors. The car motion is predicted from the estimated car velocity and the steering angle of the front wheels. These parameters are also estimated using inertial sensor data if no direct information from the vehicle is available. It also includes transitions from outdoor to indoor and vice versa, where GNSS or Wi-Fi becomes unavailable. The movements follow previously planned tracks with exactly known positions based on indoor and outdoor maps, which have been defined during the setup the Wi-Fi fingerprinting database. These reference positions will be used to evaluate the accuracy of the proposed approach. The paper starts with an overview of Bayesian filtering for positioning with emphasis on particle filters, describes the sensor and process models and explains the used algorithms. The experimental results are presented together with a description of the setup and equipment. The achieved positioning uncertainty in indoor and outdoor environments and in transition regions is discussed. This work has been done under the support of the project “Gemeinschaftliche-e-Mobilität” founded by the Fraunhofer’s initiative “Märkte von Übermorgen”.
Published in:	Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013) September 16 - 20, 2013 Nashville Convention Center, Nashville, Tennessee Nashville, TN
Pages:	1252 - 1259
Cite this article:	Hejc, G., Seitz, J., Boronat, J. Gutierrez, Vaupel, T., "Seamless Indoor Outdoor Positioning Using Bayesian Sensor Data Fusion on Mobile and Embedded Devices," Proceedings of the 26th International Technical Meeting of the Satellite Division of The Institute of Navigation (ION GNSS+ 2013), Nashville, TN, September 2013, pp. 1252-1259.
Full Paper:	ION Members/Non-Members: 1 Download Credit Sign In