On the crowdsourcing-based radio map construction with noisy location labels

Abstract

In order to reduce the overheads of constructing a dense radio map as well as to prevent the accuracy degradation, various crowdsourcing-based methods have been developed to automatically collect WiFi RSS measurements to build the radio map. Unlike existing studies focusing on crowdsourcing techniques, this paper deals with how to efficiently and accurately produce the radio map based on crowdsourcing RSS measurements which suffer from noisy location labels. In the literature, gaussian process regression (GPR) is commonly adopted to construct radio maps by sufficiently making use of the spatial correlation among received signal strength (RSS) measurements at nearby locations. However, the standard GPR does not take into account the uncertainties in the location labels attached to the crowdsourcing RSS measurements, which consequently deteriorates the performance of localization systems relying on the corresponding radio maps. Hence, the standard GPR is extended to mitigate the influences of noisy location labels. Experiments are carried out based on practical RSS measurements, and confirm the feasibility and superiority of the proposed method.