RMapCS: Radio Map Construction From Crowdsourced Samples for Indoor Localization

Fingerprint crowdsourcing has recently been promoted as a promising solution for fingerprinting-based indoor localization systems to relieve the burden of site survey. When constructing an indoor localization map from crowd-sourced samples, the following challenges should be addressed: inaccurate sample annotation, unequal sample dimensionality, measurement device diversity, and nonuniform spatial distribution. In this paper, we propose the radio map construction from crowd-sourced sample (RMapCS) scheme to handle these challenges. The RMapCS consists of four main modules: outlier detection, source selection, fingerprint interpolation, and device calibration. For each device in each grid, we first propose an improved clustering algorithm to remove outliers and use a threshold-based approach to select only those important signal sources. For a grid without enough samples, we propose a fingerprint interpolation algorithm to construct its device-specific fingerprints. Then, we propose a device calibration algorithm to fuse samples from different devices to obtain grid fingerprints. We also propose a two-step online positioning algorithm consisting of both set comparison and similarity computation. We conduct field measurements and experiments to examine the localization performance. Results show that the proposed RMapCS can achieve significant improvements over the peer schemes and the average localization error can achieve around 1.5 m by using only the received signal strength-based fingerprints.