Adaptive Information Matrix Filtering fusion with nonlinear classifier

An adaptive filtering approach is present for fusing the tracks of multi-sensor surveillance systems. The approach is an algorithm of hierarchical estimation fusion which consists of several local nodes and a global node. A linear Kalman filter is employed by each local node to produce the track estimate of the same target. The outputs of all local nodes are transmitted to the global node. In this node, an adaptive filter, which consists of the dual-band Information Matrix Filter (IMF) and a nonlinear classifier, is utilized to combine the local estimates to generate a global estimate appropriately. The feature vectors are yielded by dual-band IMF. The classifier, which is designed by using the radial basis function network in Gaussian form, is used to divide the feature space into regions that correspond to decide the output of either high-level-band IMF or low-level-band IMF against the uncertainties of target dynamics. The proposed filter has better tracking performance than each individual IMF. Simulation results are included to demonstrate the effectiveness of proposed filter.