Adaptive indoor positioning system based on locating globally deployed WiFi signal sources
Recent trends in data driven applications have encouraged expanding location awareness to indoors. Various attributes driven by location data indoors require large scale deployment that could expand beyond specific venue to a city, country or even global coverage. Social media, assets or personnel tracking, marketing or advertising are examples of applications that heavily utilise location attributes. Various solutions suggest triangulation between WiFi access points to obtain location attribution indoors imitating the GPS accurate estimation through satellites constellations. However, locating signal sources deep indoors introduces various challenges that cannot be addressed via the traditional war-driving or war-walking methods. This research sets out to address the problem of locating WiFi signal sources deep indoors in unsupervised deployment, without previous training or calibration. To achieve this, we developed a grid approach to mitigate for none line of site (NLoS) conditions by clustering signal readings into multi-hypothesis Gaussians distributions. We have also employed hypothesis testing classification to estimate signal attenuation through unknown layouts to remove dependencies on indoor maps availability. Furthermore, we introduced novel methods for locating signal sources deep indoors and presented the concept of WiFi access point (WAP) temporal profiles as an adaptive radio-map with global coverage. Nevertheless, the primary contribution of this research appears in utilisation of data streaming, creation and maintenance of self-organising networks of WAPs through an adaptive deployment of mass-spring relaxation algorithm. In addition, complementary database utilisation components such as error estimation, position estimation and expanding to 3D have been discussed. To justify the outcome of this research, we present results for testing the proposed system on large scale dataset covering various indoor environments in different parts of the world. Finally, we propose scalable indoor positioning system based on received signal strength (RSSI) measurements of WiFi access points to resolve the indoor positioning challenge. To enable the adoption of the proposed solution to global scale, we deployed a piece of software on multitude of smartphone devices to collect data occasionally without the context of venue, environment or custom hardware. To conclude, this thesis provides learning for novel adaptive crowd-sourcing system that automatically deals with tolerance of imprecise data when locating signal sources.