UWB sequential Monte Carlo positioning using virtual anchors

We present a novel UWB indoor localization concept that performs the position estimation with a set of virtual anchor nodes, generated from a single physical anchor and floor plan information. Using range estimates to the virtual anchors, we perform multilateration to estimate the position of an agent. Previous work has shown the general applicability of this concept. In this contribution, we use a moving agent to exploit the correlation in successive positions using state-space concepts. A motion model for the agent and the measurement likelihood function allow for the use of the powerful framework of Bayesian state estimation. With this concept, we can propagate prior information on the agent position from one time step to the next. The statistical model for the ranging to the virtual anchors accounts for several imperfections, which lead to multimodal and heavy-tailed measurement distributions. We show how modified versions of the Kalman filter as well as a particle filter can account for these imperfections and yield accurate and robust position estimates. In a typical indoor pedestrian motion scenario, we can achieve an accuracy of about 45 cm for 90% of the estimates.