Heading-Parametrized Multiple Model Method for Bearing-Only Filtering

Bearing-only filtering in two dimensions using a single sensor is a challenging nonlinear filtering problem. In filter initialization, it is commonly assumed that the target is moving towards the sensor and then a large interval for target heading is used to calculate the variance of initial heading. In reality, the heading of the target can be quite different from this assumed heading. We propose a heading-parametrized (HP) multiple model filtering method to handle an arbitrary heading of the target. We use the unscented Kalman filter with Cartesian coordinates (CUKF) and extended Kalman filter with modified polar coordinates (MPCEKF) in the HP multiple model framework. We demonstrate the validity of our approach using Monte Carlo simulations and compare the state estimation accuracy of the HP-CUKF and HP-MPCEKF with the posterior Cramer-Rao lower bound (PCRLB).