A nonparametric pattern recognition approach using the k/sub n/ nearest neighbors estimation to combine measurements

The goal of this paper is to present a sequential combination rule of possibly inconsistent measurements based on a pattern recognition approach. As a pattern recognition problem, the multisensor fusion one is divided into two parts: first, the recognition of the observed parameter value (or mode) corresponding to the classical consistency test: second, the updating procedure of the modes estimations (and their associated error covariance matrix) with the information observed in the new measurement. The procedures presented below have been developed with a probabilistic measurement model and deal with the nonparametric case: the k/sub n/ nearest neighbors method is used in order to estimate the probability density functions.