Design and Implementation of an ubiquitous positioning system based on GNSS pseudoranges and assisted with WLAN-RSSI
Philipp Richter · 2016
Already today numerous applications, such as tracking people or assets, navigation and geofencing for people or robots, intelligent transport systems and location sensitive marketing and billing, require location information. The Internet-of-Things will further increase the amount of \glspl{lbs} and also the demanded accuracy will increase. Outdoors, GNSSs provide reliable location information. Other technologies such as pervasive computing systems, wireless communication networks, miniaturised sensors, and so forth enable the localisation of an object or person in indoor environments. Many systems that enable indoor localisation require particular infrastructure or a multitude of sensors; especially if the aim is ubiquitous and seamless localisation. Taking into consideration the use of available infrastructure, the potential accuracy and precision and the error sources, GNSS and WLAN location fingerprinting with RSSI are the most promising technologies to achieve ubiquitous and seamless indoor/outdoor localisation. This thesis proposes a method to fuse GNSS pseudoranges and WLAN RSSI in order to yield a general, ubiquitous positioning system that is likewise accurate indoors, outdoors and in the transition zones. We use the recursive Bayesian estimation framework to tightly integrate GPS pseudoranges with WLAN RSSI. We present a state space model that relies on statistical models for the object's/person's motion and for the pseudorange and RSSI observations. This study addresses the fundamental issue of different state space of measurements: pseudoranges reside on a spatially continuous state space and RSSI on a spatially discrete space. To overcome this problem, Gaussian process regression for WLAN RSSI is revised and used to interpolate RSSI on space. Thus, we yield a continuous RSSI model facilitating the accurate data fusion with GPS pseudoranges. To find and propose the best suited Gaussian process regression model for RSSI, we revive the discussion about RSSI distributions and explore and assess several Gaussian process models in depth. It was also hypothesised whether different models for indoor and outdoor environments would improve the systems localisation performance. The model for the GPS observations is based on the well-known pseudorange model. Once developed the observation models, a particle filter is presented that integrates the Gaussian process based likelihood function and fuses the two measurements. The filter deals intuitively with issues as the weighting of the two sources of information, the availability of less than four pseudoranges and the spatial limitation of the fingerprinting radio map. To demonstrate the effectiveness of the proposed algorithm we developed software to record real world data, regarding synchronisation and the different coordinate systems and projections. In experiments, conducted in an environment challenging for GNSS and WLAN fingerprinting localisation using off-the-shelf sensors, we achieved, accurate and robust seamless localisation with an average accuracy of 5 metre.