Robust Decentralized Data Fusion Based on Internal Ellipsoid Approximation

Based on M-estimate, the problem of robust estimation fusion in decentralized architecture when the sensor noises are contaminated by outliers is considered. A simple robust Kalman filtering (RKF) scheme with weighted matrices of innovation sequences is introduced for local state estimation. Then, to avoid both the inconsistency of the Kalman filter and the performance conservation of the covariance intersection method, an internal ellipsoid approximation method (IEA) is proposed to fuse the local estimation in the fusion center. Finally, to demonstrate robustness of the proposed RKF and the effectiveness of IEA strategy, a simple tracking example in the presence of outliers is introduced.