Model-based angel-only tracking of a known-type aircraft using a single imaging sensor

Abstract

In the tracking of a target using a radar or an imaging sensor, the acceleration of the target is usually modeled as a random vector (or the process noise in the state-space model) with known statistical properties. For a highly maneuvering target, the process noise has a large covariance matrix, and consequently, the estimated state has a large amount of error. In this paper, we propose an approach that estimates the acceleration from the attitude of the target, and uses the estimated acceleration for the accurate tracking of the target. An imaging sensor is used for the attitude estimation of the target (an aircraft, in this paper) as well as for its tracking, not relying on a radar. To detect the attitude of a target, we use cross-correlation matching cost-function because of the background noise. A main difficulty of this approach is in the presence of multiple local minima. To overcome these problem, we propose to use the global optimization method, simulated annealing during the acquisition mode and to use a faster and simpler descending-type optimization method, downhill simplex during tracking mode. Using the attitude information, we formulate 12-state equations with attitude terms. Then, we compose 12-state EKF with 5-measurement(two angles and three Euler angles). Finally, our simulation study shows that the proposed method can track a maneuvering target much more accurately than the traditional Kalman filters.