Coordinate-Transformation-Based Filter for Improved Target Tracking

A maximum likelihood estimation method is developed for applications to the target tracking problem based on bearings-only observations from a single observer. The method involves propagation of states in rectangular coordinates in which the linear dynamics permit a closed form solution. At the measurement times the states are converted to a special polar coordinate system in which the measurement is modeled as linear in the transformed state and updated using the Kalman methodology. The coordinate transformation is chosen so that the direct transformation of the maximum likelihood estimate is approximately preserved. The numerical experiments for a target-intercept problem are presented, which show that the performance of this coordinate transformation based filter is superior to that of the Cartesian system based extended kalman filter. Approximate analytical results also corroborate the numerical results.