Enhanced Kalman Filter using Noisy Input Gaussian Process Regression for Bridging GPS Outages in a POS

A Position and Orientation System (POS) integrating an Inertial Navigation Systems (INS) and the Global Positioning System (GPS) is a key component of remote sensing motion compensation. It can provide reliable and high-frequency high-precision motion information using a Kalman Filter (KF) during GPS availability. However, the performance of a POS significantly degrades during GPS outages. To maintain reliable POS outputs, this paper proposes a new hybrid predictor based on modelling the nonlinear time-series data-driven INS-errors using Noisy Input Gaussian Process Regression (NIGPR), which takes the input noise into account. The proposed approach is used to learn the nonlinear INS-errors model when GPS signals are available. When GPS outages occur, it starts to predict the observation measurement, and then feeds it to a KF as a virtual update to estimate all the INS errors. The proposed approach is verified in a real airplane, which combines a POS and Synthetic Aperture Radar (SAR). Experimental results show that the proposed approach significantly improves the performance of the POS, with improvements more than 90% better than a KF and 10% better than a Gaussian Process Regression (GPR/KF) combination during various GPS outages.