Toward improving indoor magnetic field-based positioning system using pedestrian motion models

Indoor magnetic field has attracted considerable attention in indoor location-based services, because of its pervasive and stable attributes. Generally, in order to harness the location features of the magnetic field, particle filters are introduced to simulate the possibilities of user locations. Real-time magnetic field fingerprints are matched with model fingerprints to adjust the location possibilities. However, the computation overheads of the magnetic matching are rather high, thus limiting their applications to mobile computing platforms and indoor location-based service providers that serve massive users. In order to reduce the computation overhead, the article presents a low-cost magnetic field fingerprint matching scheme. Based on the low-frequency features of the magnetic field, the scheme updates particle weights according to the mass center of the magnetic field deltas of pedestrian steps. The proposed low-cost scheme decreases the complexity of real-time fingerprints without harming the positioning performance. In order to further improve the positioning accuracy, not asking users to hold the smartphone in fixed attitudes, we also present a smartphone attitude detection method that enables the proposed scheme to automatically select proper fingerprints. Experiments convincingly reveal that the proposed scheme achieves about 1 m accuracy at 80% with low computation overheads.