On prior navigation knowledge in multi sensor indoor localisation

Navigating to a desired destination is a key aspect of indoor localisation. Up to this point many different systems using present or past information for estimating the pedestrian's position were presented. Our work proposes a novel approach that incorporates prior navigation knowledge by using realistic human walking paths. In order to create such paths, we present a method that assigns an importance-factor to every node of a regularly tessellated graph by avoiding walls and detecting doors. The human movement is then modelled by moving along adjacent nodes into the most proper walking-direction. To enable 3D localisation, realistically shaped stairs for stepwise floor changes are used. The position is estimated over multiple floors integrating different sensor modalities, namely Wi-Fi, iBeacons, barometer, step- and turn-detection. The system was tested by avoiding any time-consuming fingerprinting and calibration process and starts with a uniform distribution over the whole building instead of a well known pedestrian location. The evaluation shows that adding prior knowledge is able to improve the localisation, even under unpredictable behaviour, faulty measurements and poorly chosen system parameters.