[an error occurred while processing this directive]
In june 2003, I graduated with a DEA in computer science and with a bachelor of engineer in computer science department from INSA de Rennes (Institut National de Sciences Appliquées).
In december 2006, I defensed my PhD whose topic was "Hybrid approaches for real-time tracking of complex objects in video sequences". This work has been undergone first in the VISTA project and then the LAGADIC project at IRISA/INRIA Rennes supervised by Éric Marchand.
Since october 2006, I am a teaching assistant in University of Rennes 1.
My work is linked with tracking and visual servoing technics. They consist in controling a robotic system using visual informations acquired by a vision system and integrated in a closed-loop control law.
Object tracking is an important open issue in computer vision. The aim of my PhD work was to propose and implement tracking methods simultaneously based on gradient (contour-based features) and grey-levels (texture-based features). Indeed, these two kind of features provide complementary information, each with differents drawbacks or advantages. The main principle of the hybrid tracking is then to take advantage of both visual features in the same objective function. The minisation is performed by a non-linear approach that gives efficient results in model-free tracking or using a 3D model of the object.
Attention was particularly paid to robustness of the algorithm with respect to partial occlusions and illumination changes.
The methods have been tested and validated on outdoor video sequences as well as indoor ones or using the team robot. The algorithms are implemented in C++ on a Linux station. cf. publi
|Homography estimation||Pose computation|
My current works deal with 3D hybrid tracking that does not require any apriori information about texture and that give still efficient results.
This work is based on the following statement: pose computation which is a classical problem in computer vision, requires too much 3D scene knowledge. Another approach consists in using at a maximum the visual informations extracted from images to estimate motion during a video sequence. This work relies on non linear minimization using virtual visual servoing technics to develope this approach. Geometrical constraints are used to elaborate the control law. They are based on the epipolar constraint in the case of a motion with translation or on the homography estimation between two successive images in case of pure rotation or planar structures. This work has been tested on real video sequences and validated on efficient augmented reality applications with very few 3D knowledge. cf. publi
During my first work in VISTA at IRISA, in the robotic part that became LAGADIC, I developed, implemented and tested on the team robot the robust control law used by Andrew Comport for virtual visual servoing, in order to achieve robust camera positioning by visual servoing. Indeed, one major issue in visual servoing is sensitivity to outliers. This control law simultaneously accomplishes the visual servoing task and takes these outliers into account. These outliers include a large class of errors as noise from image feature extraction, small scale errors in the tracking and even large scale errors in the matching between current and desired features. This approach is based on robust statistical technics, and in particular the M-estimators cf. publi
Complete List (and postcript files if available)
M. Pressigout, E. Marchand. Real-time 3D Model-Based Tracking: Combining Edge and Texture Information. In IEEE Int. Conf on Robotics and Automation, ICRA'06, Orlando, Florida, Mai 2006.
M. Pressigout, E. Marchand. A model free hybrid algorithm for real time tracking. In IEEE Int. Conf. on Image Processing, ICIP'05, Volume 3, Pages 97-100, Ge`ne, Italie, Septembre 2005.
M. Pressigout, E. Marchand. Model-free augmented reality by virtual visual servoing. In IAPR Int. Conf. on Pattern Recognition, ICPR'04, Cambridge, Great Britain, August 2004.
A. Comport, M. Pressigout, E. Marchand, F. Chaumette. A Visual Servoing Control Law that is Robust to Image Outliers. In IEEE Int. Conf. on Intelligent Robots and Systems, IROS'03, Volume 1, Pages 492-497, Las Vegas, Nevada, Octobre 2003.[an error occurred while processing this directive]