Wi-Fi fingerprint-based techniques for indoor localization have attracted extensive research interest and various algorithms have been proposed to improve localization accuracy. There are several challenges need to be addressed before a localization system can be successfully deployed. Firstly, virtual access points (VAPs) stemming from the same physical AP have high signal correlation, which results in redundant data and enhances computational overhead. Therefore, VAPs need to be identified and then filtered. Secondly, as devices of different brands or models may read signal differently, we need to calibrate the readings of heterogeneous devices. Furthermore, some highly accurate localization algorithms may be too computationally-intensive to deploy and hence we need to strike t...[ Read more ]

Wi-Fi fingerprint-based techniques for indoor localization have attracted extensive research interest and various algorithms have been proposed to improve localization accuracy. There are several challenges need to be addressed before a localization system can be successfully deployed. Firstly, virtual access points (VAPs) stemming from the same physical AP have high signal correlation, which results in redundant data and enhances computational overhead. Therefore, VAPs need to be identified and then filtered. Secondly, as devices of different brands or models may read signal differently, we need to calibrate the readings of heterogeneous devices. Furthermore, some highly accurate localization algorithms may be too computationally-intensive to deploy and hence we need to strike the balance between localization speed and accuracy. Last but not least, most localization systems only provide users their estimated locations. It would be beneficial for users to know the estimation error.

In this thesis, we address the above challenges and present our approaches. To remove the redundancy from VAPs and speed up localization, we identify and merge VAPs by transforming the problem to the clique-finding problem. To make the system applicable to heterogeneous devices, we propose a crowd-sourced approach to calibrate different devices efficiently. To balance between localization speed and accuracy, we revise an existing localization algorithm to make it significantly more computationally-efficient. We also propose a heuristic approach to estimate the localization error by providing a confidence range within which the user is likely to be. We implement all these approaches as individual modules and integrate them as a system. Extensive experimental trials conducted in Hong Kong International Airport and Hong Kong Olympian City show that our solutions are affective, and make the system more deployable in real environment.