K-means aided Kalman Filter noise estimation calibration for integrated GPS/INS Navigation

Abstract

GPS/INS integrated Kalman Filter is widely used in vehicle navigation. Conventional Kalman Filter is based on the assumption that noise covariances are fully estimated as Gaussian. However, GPS/INS integrated systems may encounter with inaccurate noise estimation, transient interference, hence noise estimation calibration is required. In this paper, a novel method is proposed, it uses K-Means clustering to automatically identify and discard transient interferences. This method does not require a priori knowledge of transient interferes, and both noise estimation in dynamic update process and measurement update process can be calibrated. Only steady measurement errors are used to calibrate noise estimation. Experiment results show the effectiveness of this method.