Indoor Localization Based on Magnetic Anomalies and Pedestrian Dead Reckoning

Abstract

localization is regularly related to beacons or war-driving. However, indoor geo-magnetism based indoor localization is one of the few which is beacon-free. Evidence[1] shows that indoor geo-magnetism which has been distorted by both the steel and concrete skeletons of modern building as well as man-made sources such as electric power systems and electric and electronic devices remains static and extremely low-frequency for several months. These kind of stable indoor magnetic anomalies in other words magnetic landmarks have been adopted to establish indoor localization systems [1] [2]. However, those systems depend on the prior knowledge of the magnetic map data which is collected previously to work. The magnetic map which frequently involves a large amount of raw magnetic data requires great effort to establish. In this paper, pedestrian dead reckoning (PDR) is integrated with indoor magnetic anomalies to provide indoor localization and mapping using foot-mount IMUs and magnetometers in an unsupervised way. We apply indoor magnetic measurements from magnetometers to discover the landmark rather than to store every magnetic measurement collected. Magnetic landmarks’ pattern is discovered through the travel of the floor plan. At the meantime, landmark together with the trajectory provided by PDR determine the location of landmarks as well as the location of the pedestrian. Our system discovers magnetic landmark adopting north-east-earth measurement of magnetic flied through the travel of the floor plan, so neither the knowledge of the magnetic landmark’s location nor the floor plan is needed. Database which consists of large amounts of raw magnetic measurements is also avoided. Magnetometers have been applied in robot localization. In those applications, the movement of a robot is mainly smooth and restrict to 2D which enables the implement of north-east-earth measurement of geomagnetic field. Due to the fact that the motion of a human is much more complicated, modulus of 3-axis magnetic measurement is commonly applied. But in our system, we still have the access to the north-east-earth measurement of magnetic flied since we have acquired the attitude of the sensor through AHRS algorithm and rotate the sensor-frame measurements into the north-east-earth frame coordinate. North-east-earth measurement of the magnetic field can distinguish magnetic landmark more accurate than using only the modulus. Though there are magnetic landmarks scattered in the indoor environment, we have no idea where they are or what they look likes. In order to discover magnetic landmarks in an unsupervised way, we first implement a magnetic landmark discover algorithm holding the assumption that the floor plan is available (The assumption will be relaxed later). The magnetic landmark search algorithm is closely related to the idea of motif search which is well-known in the data-mining and knowledge-discover community which refers to a previously unknown pattern that appears frequently in a time-series. Moreover, we expect the motifs to be densely clustered in spatial rather than scattered all over the floor plan. As the speed of the feet is varying for the interval of the pace, each time a pedestrian walk through a magnetic landmark, the magnetic anomaly pattern could be slightly different. To correctly identify a magnetic landmark, we apply the dynamic time warping and hidden Markov Models which are appreciated in the voice recognition applications where the same word spoken could be different in length, tone and volume. A naive motif search algorithm could be quadratic time which is infeasible even for medium size database. However several methods have been proposed to accelerate the search. In our system, trajectory of the pedestrian is provided by pedestrian dead reckoning. PDR based on IMU measurements benefits from zero velocity updates (ZUPT) when IMUs are installed on the pedestrian’s foot to acquire reasonable accuracy using cheap but noisy