VRML model of a tubular structure and the image mosaic of its interior
MSc dissertation proposal 2010/2011
Mosaicing
the interior of tubular structures
Introduction:
Mosaicing the interior of tubular shapes
consists in combining multiple images into a single one (mosaic) representing all the interior. Mosaicing finds
applications in e.g. simplifying inspection works on pipelines: mosaics are
fast reports of the structures as compared to (eventually long) videos.
In the framework of the present work, mosaicing
the interior of a tubular shape involves acquiring images, finding and
reconstructing corresponding feature points, fitting a simple 3D model to the
reconstructed points, estimating the camera path and dewarping
to one single mosaic image. This work focus on fitting a
simple 3D model to the reconstructed points. We propose a 3D model that
is based on cylindrical sections, and is useful both for generating simulated
data and implementing the fitting procedure.
Objectives:
The objectives of this work are twofold: (i)
proposing a robust reconstruction method of the structure of the tubular shape;
(ii) building image mosaics from the reconstructed structure. At the end, this
work is expected to demonstrate a fast and practical methodology for the
visualization of the interior of tubular shapes.
Detailed description:
In the case of tubular shapes perfectly cylindrical (no curves), watched
by cameras perfectly positioned and aligned with the cylinder axis, mosaicing the interior would be just a polar to cartesian dewarping followed by
the registration of corresponding image features. Registration would than be
just computing an homography between each pair of
images. In many situations, the camera motion and the structure of the tubular
shape are not so perfect (see e.g. the application of the capsule endoscope [Sayaka]).
In our case, we want to consider tubular shapes including straight and
curved sections, and to allow free movement of the camera. Hence we start from
the traditional idea of reconstructing points of the scene and then focus on
fitting a simple 3D model, to the various tube sections, which makes again
simple the dewarping step [Ruivo07].
The work is therefore organized in the following main steps:
1) reconstructing the tubular shape by
selecting and registering distinctive visual features along the tubular shape
(the main motion parameters may be locally optimised in order to best interpret
the input data) [SIFT features, OpenCV]
2) dewarping the tubular shape to obtain the
image mosaic of its interior.
References:
[Sayaka] Sayaka:
Next-generation capsule endoscope, http://www.rfamerica.com/sayaka/index.html,
http://www.youtube.com/watch?v=UHYPfcESvR0.
[Ruivo07] Mosaicing the Interior of Tubular Shapes: 3D-model Fitting, Luís Ruivo, José Gaspar, José Santos
Victor, Proc. of
RecPad 2007 - 13ª Conferência Portuguesa de
Reconhecimento de Padrões, Lisbon, Portugal , 2007
[SIFT features] http://www.cs.ubc.ca/~lowe/keypoints/
[OpenCV] Open Computer Vision Library, http://sourceforge.net/projects/opencv/
Expected results:
At the end of the work the students will have enriched their knowledge
in:
* Computer vision
* Image mosaicing, vision based reconstruction
Examples of expected demonstrations in simulated and/or real
environments:
* reconstruction of a tubular shape given a set
of points tracked along the tubular shape
* obtaining the image mosaic of a simple (constant section) tubular
shape
Observations:
--
More MSc dissertation
proposals on Computer and Robot Vision in:
http://omni.isr.ist.utl.pt/~jag