Revisiting the Bearings-only Filtering Problem

Abstract

We analyze the filter initialization problem in the bearings-only filtering (BOF) problem of a non-maneuvering target for the scenario where real data is processed. We assume that the target moves with the nearly constant velocity motion in the XY−plane. The ownship, also moving in the XY−plane, performs a maneuver to observe the state of the target. The state of the target is a four-dimensional Cartesian state vector with 2D position and velocity components. Although the dynamic model for the relative Cartesian state vector is widely used for state estimation in the BOF problem, we argue that it is simpler and computationally efficient to use the absolute Cartesian state vector in place of the relative Cartesian state vector. The Cartesian unscented Kalman filter (CUKF), Cartesian cubature Kalman filter (CCKF), and Cartesian PF (CPF) are used in this study. Sensor technology and real-time signal processing algorithms are expected to improve significantly in future. We analyze the performance of these filters in the high measurement accuracy and high data-rate regime. Our results show that the state estimation accuracy of all the filters in this regime are nearly the same. Therefore, a simple filter such as the CCKF is suitable with a low computational cost.