Optimal Non-Assignment Costs for the GNP Problem

Abstract

The global nearest pattern (GNP) approach to data association is closely related to the global nearest neighbor (GNN) problem, and both require that a cost of non-assignment of tracks be established. The existing theory for GNN can be reasonably applied to GNP problems, but adjustments are required to optimally account for bias estimation and uncertainty in GNP. These adjustments are presented along with Monte Carlo analysis showing the achieved performance is nearly optimal.