The extensive deployment of wireless infrastructure provides a low-cost way to track mobile users in indoor environments. Crowdsourcing has been promoted as an efficient way to reduce the labour-intensive site survey process in conventional fingerprint-based localization systems. Alongside its promising advantages, crowdsourcing produces a number of new challenges, including the heterogeneity of devices resulting in signal diversity and varying sensitivities to different access points. These are caused by different sensor specifications and antenna attenuation. These challenges are exacerbated by differences in the device populations between the survey and client phases. This paper presents a prototype model of a multiple-surveyor-multiple-client system to localize mobile users based on a crowdsourced fingerprint. A linear regression model is applied to calibrate across participating training devices. The conditional likelihood of a client observing an access point not visible in the training phase is obtained using a geometric distribution. The proposed system is able to achieve comparable localization performance. Field test results demonstrate the efficiency and benefit that the previously built radio map can be adapted to localize a new device that has not participated in the training phase with an average matching accuracy of 94% in a real world wireless network.
Read full abstract