Efficient approximation of the optimal bayesian filter for tracking a target in a cluttered environment

Abstract

The task of target tracking can be classified into three subtasks: track formation (initiation), track maintenance, and track deletion. In this dissertation work, we propose new filters for track maintenance and evaluate the performance of the two-stage cascaded logic which is a logicbased technique for track formation. First, for tracking a maneuvering target in a clutter-free environment, we propose modified versions of the Moose``s filter and the adaptive tracking filter by Ricker and Williams. The original Mooes``s filter and the adaptive tracking filter estimate the target maneuver in order to update the state of a target. In the process of state update, they use the same Kalman gain as the standard Kalman filter, which is based on an assumption that the target maneuver must be estimated exactly at every time. However, since the assumption does not always hold for those filter, they are bound to suffer from performance degradation. To alleviate this performance degradation, we propose modified versions of them of which the associated Kalman gains are made to reflect the estimation error of the target maneuver. Computer simulation results show that the modified filters perform better than the original ones in terms of position error. Second, for tracking a target in a cluttered environment, we propose a new data association filter called the interacting multiple data association filter (IMDAF). The N-scan back filter which is a suboptimal Bayesian filter for target tracking in a cluttered environment performs almost optimally for N $\geq$ 1 and yields relatively lower performance for N = 0 which corresponds to the probabilistic data association filter (PDAF). This implies that the one-scan back filter is an efficient one in view of performance and computational load. However, the computational load of the onescan back filter is still considerable. So it is necessary to reduce the computational load of the ond-scan back filter without performance degradation...