Robust information fusion using variable-bandwidth density estimation

We present a new approach to information fusion based on kernel density estimation. Employing a density estimation strategy with ddaptive bandwidths, we develop the variablebandwidth mean shift as an efficient technique for mode detection. Interestingly enough, the mean shifi procedure leads to the definition of a new fusion estimate as the location of the highest mode of a density function which takes into account the uncertainty of the estimates to be fused. We show that the new VariableBandwidth Density-based Fusion (VBDF) is consistent and conservative in the sense defined by Covariance Intersection, while being robust to outliers. At the same time, the new framework can reveal the presence of multiple source models in the input data. Experimental comparisons with the BLUE fision and Covariance Intersection show some of the advantages of our technique.