Practical Server-side Indoor Localization: Tackling Cardinality & Outlier Challenges

In spite of many advances in indoor localization techniques, practical implementation of robust device-independent, server-side Wi-Fi localization (i.e., without any active participation of client devices) remains a challenge. This work utilizes an operationally-deployed Wi-Fi based indoor location infrastructure, based on the classical RADAR algorithm, to tackle two such practical challenges: (a) Low cardinality, whereby only the associated AP generates sufficient RSSI reports and (b) outlier identification, which requires explicit identification of mobile clients that are attached to the Wi-Fi network but outside the fingerprinted region. To tackle the low-cardinality problem, we present a technique that uses cardinality changes to demarcate periods of stationary behaviour, and then augment the RSSI reports with useful but apparently “stale” RSSI readings from neighbouring APs. To tackle the filtering of clients with outlier locations, we propose a model that combines a weighted path-loss propagation model with a Voronoi tessellation of the fingerprint map to define suitable boundary values for RSSI readings. We experimentally show how these two approaches improve the stability and robustness of location tracking, and consequently, the accuracy of overall occupancy estimation.