Covariance-Based Measurement Selection Criterion for Gaussian-Based Algorithms

Abstract

Abstract:Process modeling by means of Gaussian-based algorithms often suffers from redundant information which usually increases the estimation computational complexity without significantly improving the estimation performance. In this article, a non-arbitrary measurement selection criterion for Gaussian-based algorithms is proposed. The measurement selection criterion is based on the determination of the most significant measurement from both an estimation convergence perspective and the covariance matrix associated with the measurement. The selection criterion is independent from the nature of the measured variable. This criterion is used in conjunction with three Gaussian-based algorithms: the EIF (Extended Information Filter), the EKF (Extended Kalman Filter) and the UKF (Unscented Kalman Filter). Nevertheless, the measurement selection criterion shown herein can also be applied to other Gaussian-based algorithms. Although this work is focused on environment modeling, the results shown herein can be applied to other Gaussian-based algorithm implementations. Mathematical descriptions and implementation results that validate the proposal are also included in this work.

An Improved Strong Tracking UKF Based on Fading Factor

“…UKF [1][2] (Unscented Kalman Filter) algorithm can solve the nonlinear filtering problem of GPS/DR integrated navigation system model [3][4]. However the precision and stability of the filter will be affected badly when the exception of measurement vector happens in integrated navigation system, and it leads to excessive location deviation [5][6]. This paper presents an improved strong tracking UKF algorithm (ISTUKF) based on STUKF algorithm.…”