Data fusion and tracking of complex target maneuvers with a simplex-trained neural network-based architecture

The capabilities of trained neural networks to perform fusion of data collected from dissimilar sensors employed in target surveillance and tracking environments facilitate an attractive framework for developing advanced target tracking architectures. We present a scheme that employs an integration of a multilayer neural network trained with features extracted from multisensor data and a Kalman filter that yields a reliable tracking algorithm capable of following even noncooperative targets executing complex evasive maneuvers. A learning strategy based on a simplex optimization algorithm that seeks the global minimum of the training error and a progressive network growing procedure are employed to develop the required capabilities underlying the desirable tracking performance delivered by the neural network tracking algorithm. Some representative performance validation studies are given in the form of tracking experiments involving targets executing not only straight line acceleration maneuvers but also complex turns in environments characterized by severe noise and clutter. A fundamental characteristic that deserves emphasis in the proposed target tracking architecture is the role of the neural network in performing data fusion and in providing assistance to a simple linear Kalman filter for tracking the maneuvering target, which provides an intelligent way of implementing an overall nonlinear tracking filter without any attendant increases in computational complexity.