Self-tuning Weighted Measurement Fusion Kalman Estimator And Its Applications

In modern C~3I (command, control, communications and intelligence) systems,information provided by the individual sensor-based estimate can't be satisfied by theneed of modern war. Multisensor data fuion must be used to provide measure data,detect and identify object in real time. Therefore technology of multisensor data fusiondeveloped rapidly, and was applied widely to modern C~3I systems and military fields.How can we combine the multisensor measurements to obtain a joint state-vectorestimate which is better than the individual sensor-based estimate? There are variousmultisensor data fusion approaches to resolve this problem, of which Kalman filtering isone of the most significant. Methods for Kalman-filter-based multisensor data fusion,including state-vector fusion and measurement fusion, have been widely studied overthe last decade.Currently there exist two commonly used measurement fusion methods forKalman-filter-based multisensor data fusion. The first called centralized measurementfusion method which simply merges the multisensor data, which increases thedimension of the observation vector of the Kalman filter, whereas the second calledweighted measurement fusion method which combines the multisensor data based on aminimum-mean-square-error criterion. Compared with the centralized measurementfusion method, weighted measurement fusion method have the global optimality andcompletely functional equivalence which can reduce the computational burdenobviously, so adapted for real time applications.This paper, for the multisensor system, using the weighted least squares (WLS)criterion, a weighted measurement fusion equation is obtained which accompanies thestate equation to constitute a equivalent weighted measurement fusion system. Based onRiccati equation, two kinds of steady optimal weighted measurement fusion Kalman estimators (filter, predictor and smoother) are presented and proved to have completelyfunctional equivalence. For the multisensor systems with unknown noise statistics, anew measurement process which presented by two moving average processes areobtained by introducing the left leff-coprime factorization, based on the solution of thematrix equations for correlation function, the on-line estimators of the noise variancematrices are obtained, whose consistency is proved by using the ergodicity of sampledcorrelation function of the new measurement process. Further, two kinds ofself-tuning weighted measurement fusion Kalman estimator are presented based on theRiccati equation. The convergences of the estimator are proved by the dynamic errorsystem. That is to say, it is strictly proved that the self-tuning fuser converges to thesteady-state Kalman fuser for the weighted measurement fusion system in a realizationor with probability one, so that it has asymptotic global optimality. The simulatedexamples for the tracking system show their effectiveness.