Target tracking by neural network maneuver detection and input estimation

Abstract

Although the Kalman filter is a powerful linear estimator for a continuous random process, it may fail to converge in the presence of sharp measurement discontinuities which may be caused by clutter or sudden target maneuvers. On the other hand, conventional models for the detection and compensation of target maneuvers are primarily based on a linear mapping of the innovation process onto an artificial noise process which is used to further adjust the covariance matrices of the Kalman filter. The nonlinear mapping capabilities of trained neural networks are employed to generate an estimate of the input noise through parallel processing of the Doppler information, the innovation process, and heading change estimate of a maneuvering target in clutter. It is shown that a neural network in conjunction with the Kalman filter can better resolve the bias caused by target maneuvers.