Magnetic Field based Indoor Localization System: A Crowdsourcing Approach

Over the past decade, crowdsourcing has been actively studied for indoor localization since surveying sites (e.g. wardriving) is a costly process. However, the existing localization systems based on crowdsourcing usually achieve lower location accuracy than the site survey based systems. We note that the magnetic field is robust to environmental changes like pedestrian activities and door/window movements, particularly compared with radio signals such as WiFi. To overcome the low performance of the crowdsourcing based approaches, we design an indoor positioning system using the crowdsourced data of the magnetic field. We substantiate a novel HMM-based learning model to construct a database of magnetic field fingerprints from smartphone users. Experiments in an indoor space consisting of aisles show that the proposed system achieves the learning accuracy of 96.47% and median positioning accuracy of 0.25m.