N.-V. Nguyen, G. Shevlyakov, and V. Shin
Multisensor data fusion, robust estimation, outliers
In distributed multisensor fusion, local estimates are communicated to a distant central processor via noisy communication channels. Under Gaussian noise assumption, there exist optimal linear methods solving the problem. In practise, channel noises are often heavy- tailed non-Gaussian, hence alternative methods should be used to make the fusion estimation robust to outliers. M-estimates are well- known robust tools, however with considerable correlation of local estimates, fusion accuracy may decrease. A two-stage robust method is proposed to fuse local estimates after a trimming procedure. In experiment, the proposed method outperforms conventional M- estimates, especially with correlation of local estimates.
Important Links:
Go Back
