Automatic target recognition using kinematic priors
N.J. Cutaia, Joseph A. O’Sullivan · 2002
Traditional automatic target recognition (ATR) systems discriminate based upon target size, target shape, or both. In this paper, an ATR algorithm is proposed that exploits aircraft-class specific kinematics to assess the tracked target's likelihood. Prior information on kinematics includes the physical parameters of the aircraft, allowable input forces to a pilot, and pilot behavior in the aircraft. It is shown that the computation of the likelihood of observed events is intractable. A suboptimal approximation to the likelihood can be computed using a hypothesis reduction method based on the generalized pseudo-Bayesian (GPB) class of algorithms. A bound on the L/sub 1/ distance of a suboptimal approximate density from the true density is derived.>