Recursive Gaussian Processes and Fingerprinting for Indoor Navigation

摘要

This paper investigates the use of Gaussian Processes (GP) for RSS fingerprinting-based navigation. In particular, a recursive GP scheme to adapt the fingerprints as they evolve over time is discussed which accounts for the uncertainty of position labels associated to new RSS measurements. The marginalization over the uncertainty of position labels is here implemented numerically through cubature rules, which is seen from computer simulations to enhance field estimation performance and, ultimately, positioning results.