Signal modelling based scalable hybrid Wi-Fi indoor positioning system
Syed, Usman Basha
Location based services (LBS) such as advertising, navigation and social media require a mobile device to be aware of its location anywhere. Global Positioning System (GPS) is accurate outdoors. However, in case of indoor environments, GPS fails to provide a location due to non-line of sight. Even in cases where GPS does manage to get a position fix indoors, it is largely inaccurate due to interference of indoor environment. Wi-Fi based indoor positioning offers best solution indoors, due to wide usage of Wi-Fi for internet access. Wi-Fi based indoor positioning systems are widely based on two techniques, first Lateration which uses distances estimated based on signal properties such as RSS (Received Signal Strength) and second, Fingerprint matching of data collected in offline phase. The accuracy of estimated position using Lateration techniques is lower compared to fingerprinting techniques. However, Fingerprinting techniques require storing a large amount of data and are also computationally intensive. Another drawback of systems based on fingerprinting techniques is that they are not scalable. As the system is scaled up, the database required to be maintained for fingerprinting techniques increases significantly. Lateration techniques also have challenges with coordinate system used in a scaled-up system. This thesis proposes a new scalable positioning system which combines the two techniques and reduces the amount of data to be stored, but also provides accuracy close to fingerprinting techniques. Data collected during the offline/calibration phase is processed by dividing the test area into blocks and then stored for use during online/positioning phase. During positioning phase, processed data is used to identify the block first and then lateration techniques are used to refine the estimated location. The current system reduces the data to be stored by a factor of 20. And the 50th percentile accuracy with this novel system is 4.8m, while fingerprint system accuracy was 2.8m using same data. The significant reduction in database size and lower computational intensity benefits some of the applications like location-based search engines even with slightly lower performance in terms of accuracy.