Applications of phase registration in robotic mapping

  • Registration of sensor data has been an active area of research in many fields from video data to the processing of range information. This thesis presents spectral based registration solutions for a wide variety of applications. The matching itself is based on phase registrations of shifted structures which allows a robust processing with respect to occlusion, interference and distortions inherent to specific sensor types. One application is the determination of homography parameters within video frames of down-looking monocular cameras. The goal is a generation of photo-maps for local areas. Here the parameter space is restricted to rotation, scaling and translation which correspond to sensor movements as x, y, z - translation and rotation. For this task the well known Fourier Mellin Transform (FMT) is applied with improvements and adaptions concerning the image material which is to be processed. The movement of the sensor platforms is assumed to be mainly in parallel to the observed surface, hence the registration of sensor tilts is not considered but an assessment in which way it affects the FMT processing is given. A second application in this thesis is the processing of 2D range information as provided from e.g. Laser Range Finders (LRF) and 2D sonar systems. The spectral processing as the FMT depends on decoupling translation from other transformations within the data. The corresponding generation and processing of so-called descriptors is a necessary pre-processing step of the registration process. For the registration of rotated and translated scan data, the Mellin transformation is excluded and the parameter space is therefore reduced from 2D to 1D which makes the algorithm even more robust to interference and occlusion. In a third application this principle is extended to 3D. Compared to the registration in 2D which are based on already existing approaches like the FMT and the polar resampling approach a novel concept for the determination of 3D rotation is introduced. Since algebraically correct approaches like spherical harmonics will not succeed with the available resolution of the spectral magnitude and approximation is suggested in this work. This approximation allows again a spectral phase matching based on a 3D Cartesian grid. The resulting algorithms show extreme robust registration results on real world acquired scan data. The proposed phase matching with its Dirac response allows furthermore a probabilistic post-processing of the registered parameters.

Download full text

Cite this publication

  • Export Bibtex
  • Export RIS

Citable URL (?):

Search for this publication

Search Google Scholar Search Catalog of German National Library Search OCLC WorldCat Search Catalog of GBV Common Library Network Search Catalog of Jacobs University Library Search Bielefeld Academic Search Engine
Meta data
Publishing Institution:IRC-Library, Information Resource Center der Jacobs University Bremen
Granting Institution:Jacobs Univ.
Author:Heiko Bülow
Referee:Andreas Birk, Andreas Büchter, Udo Freese
Advisor:Andreas Birk
Persistent Identifier (URN):urn:nbn:de:101:1-2013052411121
Document Type:PhD Thesis
Language:English
Date of Successful Oral Defense:2010/10/08
Year of Completion:2010
Date of First Publication:2011/01/31
PhD Degree:Computer Science
School:SES School of Engineering and Science
Library of Congress Classification:T Technology / TJ Mechanical engineering and machinery / TJ210.2-211.47 Mechanical devices and figures. Automata. Ingenious mechanisms. Robots (General) / TJ211.495 Autonomous robots

$Rev: 13581 $