A Confidence Measure for Feature-based Localization using Random Finite Sets

A reliable confidence measure for the localization of road vehicles is a crucial requirement to enable highly automated driving with an inherent self-awareness of its robustness and capabilities. Therefore, the present contribution introduces a novel approach to estimate not only the spatial uncertainty of a feature-based localization algorithm but also an associated confidence measure. The methodology makes use of Random Finite Sets to model and quantitatively describe the (expected) difference between a given map of landmarks and a set of corresponding online measurements. Based on a probabilistic representation of the two sets, the above-mentioned confidence measure is derived. Simulation results based on a Monte-Carlo localization framework are provided.