Localization using OSM landmarks

This paper presents an algorithm that uses data from the OpenStreetMap (OSM) project to complement the localization based on a Global Navigation Satellite System (GNSS). The landmarks extracted from OSM serve as a set of reference points and are compared to objects currently detected by a detector. The algorithm uses several metrics to find a match and calculate an estimated location. The goal is to improve high-level localization since in urban environments the freedom of movement is often restricted to defined paths like streets or paved ways. To store and compare the expected and detected objects, a scene graph is used to deal with the high-level logical landmarks around the robot. By doing so, we can broadly localize our self and integrate the knowledge of our expected surrounding based on OSM data.