|
On appearance-based methods
-
[Lab06] (download: pdf)
-
Frédéric Labrosse. Appearance-based heading estimation:
the visual compass. Technical Report UWA-DCS-06-048,
Department of Computer Science, University of Wales, Aberystwyth,
January 2006.
In this report we present an algorithm to estimate the heading
of a robot relative to a heading specified at the beginning of
the process. This is done by computing the rotation of the
robot between successive panoramic images, grabbed on the robot
while it moves, using a sub-symbolic method to match the images.
The context of the work is Simultaneous Localisation And Mapping
(SLAM) in unstructured and unmodified environments. As such,
very little assumptions are made about the environment; the few
made are much more reasonable and less constraining than the
ones usually made in such work.
The algorithm's performance depends on the value of a number of
parameters, values being determined to provide overall good
performance of the system. The performance is evaluated in
different situations (trajectories and environments) with the
same parameters and the results show that the method performs
adequately for its intended use. In particular, the error is
shown to be drifting slowly, in fact much slower than
un-processed inertial sensors, thus only requiring un-frequent
re-alignment, for example when re-localising in a topological
map.
On image representation
-
[Lab00] (download: ps.gz, pdf)
-
Frédéric Labrosse. Towards Continuous Image
Representations. Technical Report, Department of
Mathematical Sciences, Computing Group, University of Bath, UK,
April 2000.
In the cinema industry, special effects performed during
post-production generally use pixel based versions of the movie
frames. Although this type of representation is easy to obtain,
it has problems like the amount of data and processing
difficulties. (How do you remove an object in an image when its
outline is not well defined because of the blur naturally present
in the image?)
We propose to use continuous, i.e. vectorial,
representations. They are indeed easy to manipulate and it has
been shown that they can be used to render very high resolution
images, which is necessary for cinema, in affordable times.
In this report, we address a first step towards such
representations: the extraction from images of smooth continuous
contours at sub-pixel accuracy and some ways of representing their
interior. The sub-pixel accuracy is necessary to obtain
representations that are resolution independent.
Images are decomposed into structural regions that correspond to
specified image characteristics. This is done using standard
relaxation labelling. Information taken at different stage during
the relaxation is used to extract structural contours.
Sub-pixel accuracy is obtained by using snakes as well as the blur
present in images (because of the acquisition process). We propose
solutions, adapted to our context, to often mentioned problems of
snakes, namely initialisation, parameter determination, and
instability.
The interior of the structural regions must be represented to allow
the rendering of images as close as possible to the original ones.
We propose here two schemes, one using a single colour for each
region, the second sampling the original image to allow smoothly
varying colour in each region.
|