Hierarchical Indoor Localization From Crowdsourced Samples

Although fingerprinting-based indoor localization techniques have been researched for decades, their practical deployments are still facing the great challenge of time-consuming and labor-intensive site surveys. Recently, sample crowdsourcing and radio surfacing have been proposed for indoor localization for their advantages of not involving burdensome site surveys. As crowdsourced samples are often with inaccurately annotated location labels, the fitted radio propagation surfaces are normally of low quality, which in turn adversely impacts localization accuracies. In this letter, we propose a hierarchical indoor location (HiLoc) scheme to improve localization accuracies of radio surfacing from crowdsourced samples. In HiLoc, we construct subarea radio surfaces to first estimate the initial location of a test sample. We next construct hierarchical tiers of radio surfaces each for one of concentric disks centered at the initial location and obtain a series of estimated hierarchical locations. We then remove possible outliers from the hierarchical locations and propose a weighted location fusion to output the final localization result. Experiments on two field datasets validate great improvements over the peer schemes in terms of much lower localization errors.