Wi-Fi fingerprinting is a promising approach for indoor localization due to its ease of deployment and high accuracy. As the signal from access points (APs) may change (due to, for examples, AP movement or power adjustment), the offline site survey often needs to be regularly conducted to maintain localization accuracy. This is costly and time-consuming.

In this thesis, we propose LAAFU (Localization with Altered APs and Fingerprint Updating), which achieves both accurate indoor localization and automatic fingerprint update in the presence of altered APs without the need of site survey. LAAFU is based on implicit crowdsourcing. Using novel subset sampling, it is able to efficiently identify the altered APs and filter them out before localization, hence achieving high accuracy. W...[ Read more ]

Wi-Fi fingerprinting is a promising approach for indoor localization due to its ease of deployment and high accuracy. As the signal from access points (APs) may change (due to, for examples, AP movement or power adjustment), the offline site survey often needs to be regularly conducted to maintain localization accuracy. This is costly and time-consuming.

In this thesis, we propose LAAFU (Localization with Altered APs and Fingerprint Updating), which achieves both accurate indoor localization and automatic fingerprint update in the presence of altered APs without the need of site survey. LAAFU is based on implicit crowdsourcing. Using novel subset sampling, it is able to efficiently identify the altered APs and filter them out before localization, hence achieving high accuracy. With the client locations, the fingerprint signal due to the altered APs can be adaptively and transparently updated using the non-parametric Gaussian process regression method. We have implemented LAAFU and conducted extensive experiments in our campus. Our results show that LAAFU is robust to altered AP signal changes to achieve high localization accuracy, and its fingerprint database is able to adapt to the current signal environment.