||Loop Closing in SLAM
||Knudsen, Kinna Majvig
||Aanæs, Henrik (Image Analysis and Computer Graphics, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
||Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
||A general problem in the navigation system of mobile robots is the integration
error, which will cause the robot's location to be very inaccurate at some point.
The loop closing procedure detects previously visited places which is used to
adjust the positions, and thus reduce the effect of the error. This thesis is
examining a visual approach of detecting the presence of a loop. Here, images are
used to represent places and the SIFT algorithm is used to extract information
from these images. A database of previously visited places is created from this
information, where each place is represented as in the Bag-of-Words model in
a visual version. The structure of the database is created as a vocabulary tree.
The partition of known and unknown places is based on a statistical approach.
Altogether the system received a 44% final recognition rate at its maximum
performance whereas the recognition rate for the image detection alone was
This shows that the visual approach is useful, but the system would still need
some improvements before it can be used as a robust tool for loop detection.
||Technical University of Denmark (DTU) : Kgs. Lyngby, Denmark
Creation date: 2011-01-03
Update date: 2011-01-03