An Adaptive Bayesian Approach to Fusion of Imaging and Kinematic Data

The primary goal of this paper is to develop an approach to fusion of two streams of data – imaging and kinematic – for optimization of target identification. Specifically, we focus on the fusion of rangeprofile data (obtained from a high range resolution sensor) and kinematic information for observation-to-track assignment and target recognition. A more reliable target identification is possible due to the strong correlation between kinematic characteristics and range profile. Indeed, the range profile signal depends on the range to the target and its aspect angle, while the latter is related to the target velocity (via Euler’s equation), thus yielding a strong correlation of both types of data with the aspect angle. Effective estimation of the aspect angle is therefore the key to successful target identification. The dynamics of the aspect angle is modeled by a Markov process with a switching parameter. The latter parameter describes transitions from one target maneuver to another. In this model the state process and the observation are nonlinear. This rules out application of standard methods of estimation based on Kalman filter and necessitates the use of a nonlinear filtering algorithm. The crucial part of the fusion and identification algorithm is the fully coupled optimal nonlinear filter for the aspect angle. This filter allows us to compute recursively joint unnormalized posterior distributions of the target class and aspect angle. Then specially designed adaptive sequential multihypothesis classification procedures, which exploit the optimal nonlinear estimates of the aspect angle for all classes, are used to identify targets of interest.