DorFin: WiFi Fingerprint-based Localization Revisited

Abstract

Although WiFi fingerprint-based indoor localization is attractive, its accuracy remains a primary challenge especially in mobile environments. Existing approaches either appeal to physical layer information or rely on extra wireless signals for high accuracy. In this paper, we revisit the RSS fingerprint-based localization scheme and reveal crucial observations that act as the root causes of localization errors, yet are surprisingly overlooked or even unseen in previous works. Specifically, we recognize APs' diverse discrimination for fingerprinting a specific location, observe the RSS inconsistency caused by signal fluctuations and human body blockages, and uncover the RSS outdated problem on commodity smartphones. Inspired by these insights, we devise a discrimination factor to quantify different APs' discrimination, incorporate robust regression to tolerate outlier measurements, and reassemble different fingerprints to cope with outdated RSSs. Combining these techniques in a unified solution, we propose DorFin, a novel scheme of fingerprint generation, representation, and matching, which yields remarkable accuracy without incurring extra cost. Extensive experiments demonstrate that DorFin achieves mean error of 2 meters and more importantly, bounds the 95th percentile error under 5.5 meters; these are about 56% and 69% lower, respectively, compared with the state-of-the-art schemes such as Horus and RADAR.