Hybrid map-building and localization for unstructured and moderately dynamic environments

In this paper, a hybrid map-building and localization approach suiting unstructured and moderately dynamic environments is proposed. The map-building phase reduces the domain of extracted local point features through an information-theoretic analysis, which simultaneously selects distinctive features only. Reduced features are compressed into a dictionary yielding data for fast robot localization on a topological level. The uncompressed features, additionally tagged with their metric position, are used together to resolve the robot's position on a second metric level. The complexity of geometric localization is reduced because of the hierarchical processing based on a previously identified topological location. The dynamics of the environment are detected through the spatial layout of features and are isolated at the metric level. The proposed map-building and localization approach enables fast hybrid localization without degenerating the accuracy of localization.