Target tracking based on a multi-sensor covariance intersection fusion Kalman filter

Abstract

In a multi-sensor target tracking system, the correlation of the sensors is unknown, and the cross-covariance between the local sensors can not be calculated. To solve the problem, the multi-sensor covariance intersection fusion steady-state Kalman filter is proposed. The advantage of the proposed method is that the identification and computation of cross-covariance is avoided, thus the computational burden is significantly reduced. The new algorithm gives an upper bound of the covariance intersection fused variance matrix based on the convex combination of local estimations, therefore, ensures the convergence of the fusion filter. The accuracy of the covariance intersection (CI) fusion filter is lower than and close to that of the optimal distributed fusion steady-state Kalman filter, and is far higher than that of each local estimator. A numerical example shows that the covariance intersection fusion Kalman filter has enough fused accuracy without computing the cross-covariance.