Multiple hypotheses generation for vehicle localization

This paper introduces a matching method designed for initiating real-time but approximate iterative localization schemes. Inaccuracy in the measurements and in the modeling of reality is taken into account as well as spurious measurements caused by unpredicted obstacles. The method builds a set of matching hypotheses, each one associated with a set of transformations that maps a part of the measurements into the environment map. These feasible transformation sets are computed by a set-membership estimation algorithm. Each valid hypothesis generates a configuration confidence domain that enables the initialization of an iterative tracking.