Thesis Topic:

Monocular Localization in Feature-Annotated 3D Polygon Maps


M.Sc. Computer Science
Oct, 2019


6DoF localization and mapping is becoming increasingly relevant in autonomous robotics. Especially in environments where GPS is not available (e.g. indoor), there are more and more localization methods based on camera images. The aim of this work is to develop a closed system that allows a robot to locate itself in a 3D map through a camera. The developed software is distinguished into offline mapping and online localization. The map consists of a polygon mesh, which is reconstructed from 3D laser scans of the environment. In addition, image features are extracted from high-resolution camera images. These features are added to the map with their corresponding 3D positions. The online localization is then performed on the basis of low resolution camera images. To do this, features are extracted again and the most similar descriptors are searched in the map. The 6DoF pose is then determined from the 2D-3D point correspondences found.