VISTA research work is concerned with image sequence analysis and active vision.
More precisely, we address two broad issues:
analysis of physical phenomena for providing image-based or scene-based
motion-related measurements, and for investigating recognition and
interpretation of temporal events;
perception and control of automated or robot systems, for handling
manipulation, tracking, navigation, surveillance or exploration tasks.
In that context, we are interested in several types of spatio-temporal
images, from optical imagery (video, infra-red) as well as from acoustics
(sonar, echography).
We mainly follow a statistical approach (involving Markovian and Bayesian
models and methods) to properly handle motion and deformation analysis.
The aspects of long-term spatio-temporal analysis (e.g., tracking)
are particularly studied. We consider situations of active observer,
i.e. we investigate active perception, when the sensor motion or attitude
can be controlled. This involves the design of visual servoing techniques
and of higher level exploration strategies.
Research themes
Three sectors of applications motivate and validate our work:
motion and deformation metrology (in meteorological satellite imagery,
medical imagery, experimental visualization in fluid mechanics),
robot vision and surveillance systems (power industry,
underwater robotics, sonar, transportation),
content-based video indexing.
International and industrial relations
Esprit Projects
We are participating in the DiVAN ("Distributed audio-Visual Archives
Network") R&D Esprit project (Multimedia Systems program) together with
Greek, German, Italian and French partners. Our contribution to the project
concerns video indexing by contents.
Contracts
We have ongoing research or transfer agreements in the three above
mentioned application sectors. These agreements are with Météo-France
(Spatiotemporal analysis of satellite images for meteorological
observation), GIS Sciences de la Cognition (Brain imaging), EDF (2D
follow-up of complex shapes with exploitation of CAD models), Ifremer
(Motion estimation and image stabilization, transfer of the Motion-2D
software), DCN (Information processing for detection and tracking systems),
Thomson-RCM (Data merging and tracking, Cifre agreement),
Thomson-CSF-Optronique and DGA (Image sequence enhancement), Alcatel-CRC
(Video indexing by contents, jointly with project Movi), Arts Video
Interactive (Commercial using license for the MD-Shots software) and Ina
(Video indexing, Cifre agreement). In addition, we are participating in the
Agir RNRT project (Video indexing by contents).
Academic Relations
Naval Academy, Brest (Sonar imaging);
LEA, Poitiers (Fluid mechanics);
GIS GSTB "Brittany Remote Detection Group" (Costel
UHB Rennes, Enssat Lannion);
Inria cooperative research initiatives with project Movi (Video indexing),
projects Epidaure and Isa (3D ultrasound scan), and projects Bip, Icare and
Movi (Visual servo-control);
France-Israel project on video indexing in collaboration with S. Peleg’s
laboratory (Jerusalem Hebrew University) and project Movi.
