Précédent : Références Remonter : Projet ROBOTVIS

Abstract

The scientific and technological objectives of the project fall in two categories:

• to develop theories of machine visual perception and in particular of three-dimensional perception,
• to demonstrate the validity of these theories by applying them to practical problems, typically in robotics, and implementing for solving them.

Formalisation and in particular mathematical formalisation is a central motivation for us and is always checked by experiments which we believe to be the only way to validate a theory. Our theoretical work has been concentrated in two main directions. The first one is the investigation of the use of algebraic and geometric methods for studying systems with many cameras (see the section on geometry and the reports from previous years). We will continue to invest in that direction in the years to come in particular through collaborations with the Safir group (B. Mourrain) and through the Cumuli ESPRIT LTR project. The second direction is the investigation of the use of analysis methods, mainly from the theory of partial differential equations, to study the problems of image processing (enhancement, restoration) and computer vision (edges, stereo, motion), see section on EDP. This is also a lasting investment which we would like to reinforce in particular through collaborations with the Nice-Sophia Antipolis University (Gilles Aubert, Michel Merle), the Paris IX Dauphine University (Jean-Michel Morel), and the Cermics (Renaud Keriven).

We also have developed in the past a fairly large number of industrial collaborations in the framework of european projects. We have transferred to industrials a lot of our know-how and algorithms (see activity reports for the years 90 to 93). We continue to seek these collaborations but try to broaden them to a wider class of applications than Robotics, in particular to the area of 3-D modelling from image sequences and to the area of the processing of video sequences for such applications as the virtual studio.

Within this framework we have particularly pursued the following directions:

• geometry of systems of cameras, calibration, self-calibration;
• stereo and applications;
• three-dimensional shape and motion perception by analysis of monocular sequences of images;
• partial differential equations, image processing and computer vision.

Précédent : Références Remonter : Projet ROBOTVIS