SENSOR FUSION BY BAYESIAN FILTERING FOR SEAMLESS PEDESTRIAN NAVIGATION

Seamless pedestrian navigation in both indoor and outdoor environments is an unsolved challenge today. Though various navigation systems and sensors exist which are suitable in terms of size, cost, and power consumption, today none of these systems is expected to serve as a sole means for personal navigation in the mid-term future. In particular the characteristic drawbacks of today’s systems in specific environments prevent their successful use. This work shows how to solve the problem by the rigorous application of a sound theoretically motivated approach: The combination of various sensors and the optimal joint processing of their provided data by a Bayesian filter algorithm, which optimally takes into account the uncertainty inherently included in each sensor’s data and which exploits optimally all available knowledge about the movement of the navigating individual, such that in the end no information is lost during the processing of the data. After an introduction to personal navigation systems and sensors, particularly focusing on satellite and inertial navigation, and a summary on the concept and the implementation of Bayesian filters, the thesis addresses the application of Bayesian filtering to enhance the performance of satellite navigation receivers in urban multipath environments. The results confirm the benefit of the Bayesian approach, which is shown to outperform a conventional navigation receiver significantly. Subsequently a novel integration scheme for inertial sensors is proposed based on the concept of foot-mounted inertial sensing. Thereby particular emphasis is put on the incorporation of an adequate map-based pedestrian mobility model in order to reduce the heavy drift of today’s small-scale and low-cost micro-electro-mechanical inertial sensor platforms. The results show that the combination of inertial navigation with a map-based pedestrian mobility model can achieve a fully autonomous drift-free navigation in indoor environments. Finally it is shown how seamless pedestrian navigation systems can be designed successfully by the use of Bayesian filtering algorithms. The design of the filter algorithms is addressed and depending on the employed and available sensors the suitable filter implementation is chosen, including an extended Kalman filter for the combination of fingerprinting via a wireless local area network and foot-mounted inertial sensors and a particle filter for the integration of a satellite navigation receiver, a radio-frequency identification unit, a compass, a baro-altimeter, a foot-mounted inertial platform, and a map-based pedestrian mobility model.