Adaptation of a sequential discrimination rule with reject options to multisensor fusion

The goal of this paper is to present a sequential combination rule of inconsistent measurements. Identified with a statistical pattern recognition problem, the multisensor fusion problem is divided into three parts. The first one is the recognition of the new measurement mode. We adapt a discrimination rule applied in diagnosis based on a statistical pattern recognition approach. This rule allows the creation of new classes and the partial classification of ambiguous measurements. In the second part, the measurements associated with the same class are combined in order to produce a mode estimation. A modified weighted least squares algorithm taking into account the partial classification is proposed. In the third part, the discrimination rule is updated. Two cases are considered: the parametric Gaussian case and the nonparametric case