A Probabilistic Radio Map Construction Scheme for Crowdsourcing-Based Fingerprinting Localization

Fingerprinting localization using received signal strength (RSS) has been considered as a promising solution for indoor localization systems, due to its ease of use in deployed wireless networks. A major bottleneck is the difficult trade-off between localization accuracy and site survey costs. Crowdsourcing-based fingerprint collection is cost efficient because of user participation, but its accuracy is questioned due to the noisy sample locations. In this paper, we present a probabilistic radio map construction scheme with crowdsourcing collection, considering both accuracy and survey costs. In particular: 1) based on further studies of RSS properties, we introduce a concept of unfixed data collection-in which sample locations are estimated by an additional localization mechanism-and verify its effectiveness, although sample locations are noisy; 2) we present a modified parametric fitting method to better describe location signatures by transforming RSS into signal envelope; and 3) we propose a clustering-based space partitioning algorithm to improve the fitting effect by reducing the number of multimodal distributions. Extensive simulations and experiments show the proposed method can construct an accurate radio map at a low cost, and achieve a localization accuracy of about 2.5 m, 20% improvement compared to using the classical radio map in our experimental scenario.