graph made using 2000+ images; blue circles indicate
    dominating nodes

connencted dominating sets for robot mapping

For image based topological mapping, a graph is build in which each node represents an image and a link between two nodes denote that the corresponding images were taken from the same part of the environment. During mapping new images have to be compared with images in the topological map. One can imagine that it is inefficient to compare new images with all images in the map. It is better to compare with a small subset of images that gives a good representation of all images in the map. In this work, I use a graph theoretic method called connected dominating sets (CDS), to determine such a subset of images.

The use of the CDS method for the topological localization and mapping problem was suggested to me by Zoran Zivkovic, one of my PhD-supervisors. It resulted in a couple of conference papers. In an ASCI paper (2005) I presented it as a solution for topological robot localization. An extended version of this paper was later presented at IROS 2006. Later, in an RSS workshop paper, I presented it as a solution for robot mapping, adding a technique to use the CDS for incremental mapping. This was published as a RAS journal paper. This paper formed the basis for Chapter 5 in my PhD thesis.

Slides I used for a presentation for the ISLA research group can be found here: pdf (2.2 Mb).