For the multi-sensor systems with uncertain-variance linearly correlated white noises

based on the mini-max robust estimation principle

by the Lyapunov equation approach

the two classes of guaranteed cost robust covariance intersection(CI) fusion Kalman estimators (predictor

filter

smoother) are presented based on the parameterization representation of the uncertain noise variance perturbations.Both the minimal upper bound and the maximal lower bound of the accuracy deviations are given.It is proved the robust accuracy of the guaranteed cost CI fuser is higher than that of the original CI fuser

and is higher than that of each local estimator

and the geometric interpretation of accuracy relation is given by the covariance ellipses.A simulation example applied to tracking system verifies the correctness and effectiveness of the proposed method.