|
|
 |
HOME |  |
Research | |
Publications | |
CVisions |
|
The aim of this thesis proposal is to create a new set of algorithms for solving the problem to register a given 3D shape to an image showing the same shape in an arbitrary position. This work will be supported by the longstanding experience of Instituto de Sistemas e Robotica (ISR) on image registration/matching and 3D reconstruction from images and closely with REVERSE, a company manufacturing state of the art 3D acquisition systems.
|
|
The grow of sensors capable to extract an accurate 3D description of the world will increase dramatically in the near future. Contrary to the cumbersome equipment of the past (i.e. laser scanners), they will be characterised by their reduced cost and compactness. Having readily available a massive library of 3D objects, the problem is now to relate each 3D shape to the new and old 2D image information in the world. This autonomous task is still unsolved in a general case. This thesis work will be dedicated to tackling this forthcoming problem in image processing.
|
|
The project is divided in two stages. First, the candidate will have the chance to use a real system for 3D shape models acquisition currently used in many industrial applications. ISR laboratories are equipped with such machine and they are actively in contact with the company manufacturing them. As an initial and necessary step, we will make the acquisition of different 3D shape models and corresponding shapes. The second stage will face the theoretical problems on 2D-3D image matching building on the current experience of ISR in such scenario. Our expected results sum to have an application which can solve and anticipate the problems arising from the new sensors technology of the next few years. The algorithm developed will be able, given a 3D shape and its 2D image, to correctly assign the 3D object to the image data.
|
|
Our machine from REVERSE now offers us new perspectives. We plan to augment deformable and articulated models to a higher level of realism, where a model is defined with a dense mapping of the shape in 3D. In this way, subtle but important details such as skin and muscle deformations may be finally grasped by observing the real 3D motion of shapes and not from a priori synthetic models. In order to do so, we need to overcome a present limitation of the REVERSE machine since it is not able to extract 3D information in real-time. This feature is a focal point since, in order to fulfil motion analysis, we need to observe a set of closely spaced frames describing the shape deformations.
|
Project ObjectivesA new system for the 3D reconstruction of non-rigid shapes which may have a relevant impact in science engineering. The candidate will learn solid basis in Computer Vision and new sensor technologies for visual reconstruction. These tasks will let the candidate work on a real and innovative system for 3D acquisition with the possibility to make considerable impact in the 3D modelling field. Required skills are an entry-level knowledge of MATLAB scientific language. Knowledge on Image processing and Computer Vision are a plus for this project. For further questions please contact:Prof. João Paulo Costeira. Torre Norte, Piso 7, Room 7.18 Dr. Alessio Del Bue Torre Norte, Piso 7, Room 7.25 Phone Number: 2299 |
| Updated Apr 2009 by Alessio Del Bue |