A Machine Learning based Indoor Localization

Abstract

Since many years, researchers and engineers are looking for a precise Indoor Positioning System (IPS) accurate in various indoor scenarios. Several techniques were used in order to reach this goal. Deterministic methods in IPS, based on the received signal strength (RSS), usually use the average value of RSS from different sensors. We convert the RSS value into a distance estimation, and then calculate by trilateration the coordinate values. RSS is generally unstable because of the huge amount of variation parameters in indoor environments. In this paper, we propose a machine learning based approach for IPS. We compare our results to the one using some filtering method of three beacons among four. The results reached are substantially better than those of the classic method.