Real-time Indoor Localization using Magnetic, Time of Flight, and Signal Strength Inference Maps

This paper presents a localization system that fuses inertial, magnetic field strength (MAG), RF time of flight (ToF), and received signal strength (RSS) measurements to track a pedestrian moving indoors in real time. Our method uses a pose graph to model a pedestrian’s trajectory as a sequence of discrete positions. Rather than using range predictions directly in the pose graph, our approach draws values from a Gaussian Process (GP) inference map built in advance from training data. To improve scalability, our implementation is divided into two separate threads. The first thread uses inertial measurements and Kalman filtering to produce a series of displacement vectors. These displacements are forwarded to a second thread where they are inserted into a pose graph along with ToF and RSS measurements. Finally, a gradient descent method is used to obtain the sensor trajectory that minimizes the sum squared difference between the predicted and measured ToF, RSS and MAG values.