GPSLAM

Abstract

Survey-based indoor positioning systems are expensive to establish and maintain. We describe how a crowdsourced WiFi radio map building problem can be framed as a trajectory alignment problem and further solved by the existing graph-based Simultaneous Localisation and Mapping (SLAM) framework. Specifically, we show how to exploit crowd-sourced WiFi signals and construct opportunistic constraints that satisfy the WiFi signal consistency modelled by Gaussian Processes (GP). We repurpose GraphSLAM to optimise these WiFi constraints collectively to approximate an optimised global signal consistency. We implement an automatic pipeline we call GPSLAM to produce radio maps based only on crowdsourced mobile data. Evaluation based on realistically crowdsourced mobile data generated naturally and passively by common visitors of multiple shopping malls demonstrates that GPSLAM can build radio maps that achieve comparable results to laborious manual surveys while the human intervention involved is minimised.