Current PhD students
Benoît Pinier, "Similarity models and uncertainty models for ocean atmosphere coupled models".
In this thesis, granted by the IRMAR laboratory and co-supervised with R. Lewandovsky (IRMAR), we
aim at studying new formulations of ocean atmosphere coupled models. The idea is here to assess
the use of turbulence models based on similarity models and to derive new representations under
uncertainty models for such coupled models.
Pranav Chandramouli , "Coarse eddies simulation and data assimilation for recontruction of wake flows".
The thesis of Pranav takes place within a collaboration with Sylvain Laizet (Imperial College). In his thesis, we
study coarse eddies simulations formulated though the location uncertainty principle. Such stochastic representation
leads to large-scale flow representations amenable to efficient simulations. Data assimilation schemes to
couple 2D plans of 3 components velocity fields are also investigated in this PhD thesis. This thesis is co-supervised with Dominique Heitz
and Sylvain Laizet
Elise Arnaud “Filtering methods from image sequences””
In Elise’s thesis we have defined stochastic
filters for the tracking of points from image sequences. We have in particular explored a family of systems called partial linear Gaussian models. These models rely on informative dynamics estimated from the image sequences and likelihood defined as a Gaussian mixture. The thesis has been defended in November 2004. Elise is now assistant professor (Maitre de Conférences) in Grenoble at the University Joseph Fourier. She is a member of the Inria group headed by Emmanuel Prados”
Christophe Avenel "Stochastic level-sets for the tracking of closed curve".
In this thesis we have proposed
a stochastic level-set dynamics for the tracking of closed curves. Associated to a particle filter
this approach has been applied for the tracking of convective cells, the detection and tracking of precipitation
areas in weather watch radar, and the tracking of ice iso-density curves. This thesis has been
defended in december 2011. Christophe at the moment on a post-doc position at University of Uppsala
Ioana Barbu "3D motion estimation of turbulent fluids".
This thesis funded through a DGA/INRIA grant
focused on the estimation of 3D motion of particles transported by a turbulent fluid flow. In this work we
considered a unified framework, in which a sparse constraint and a non negativity condition related to
the particle density function have been considered. This study, associated to approached minimization
procedures allows us to achieve a good balance between the affordable computation load and efficiency.
The defense of this PhD, co-supervised with Cédric Herzet, has been defended the 15 of december 2014.
Sébastien Beyou "Estimation of sea surface currents from image sequences".
In this thesis we studied
the setting of efficient stochastic filtering techniques for image assimilation. The proposed particle filter
relies on a proposal density built from Kalman filter mechanisms. This thesis has been defended in july
Thomas Corpetti “ Fluid motion analysis from image sequences”
In this thesis we have studied motion estimators dedicated to image sequences of fluid flows. The motion estimator relies on a data model built from the integration along time of the continuity equation and on a second order div-curl regularization term. A method for characterizing motion fields in term of sink/ sources and vortices is also described. The thesis has been defended on July 2002. The thesis have been published as a book in hermes collection (in French).
Thomas is senior researcher (DR) at the CNRS
Anne Cuzol “Tracking and estimation of fluid motion velocity fields”
During Anne’s thesis we have
worked on the definition of low dimensional motion estimators for fluid motions. The motion is represented as a linear combination of adapted basis functions derived from so called vortex particles. The tracking of such representation is formulated through a stochastic filter. The thesis has been defended in December 2006. Anne holds now an assistant professor position (Maitre de Conférences) at the University of Bretagne Sud.
Pierre Dérian "Wavelets and Fluid motion estimation".
In the PhD thesis of Pierre Derian, we have
proposed several computationally efficient motion estimator based on wavelet bases. High order and
power law regularization were implemented on such bases. This thesis, has been defended in November 2012. Pierre is now on a Post-doctoral position at Chico
During Alejandro’s Post-doc (2009-2010), we proposed together with Dominique
Heitz an optimal control technique for inlet boundary conditions in experimental fluid mechanics. The
objective was to reproduce numerically a real world experimental flow behind an unknown obstacle.
Alejandro Gronskis has been hired as a CONICET researcher after his post-doc at the University of
Patrick was on a research engineer position within the FET FLUID project from 2005 to 2007 and on a post-doc position at Cemagref Rennes from 2007b to 2008. We worked on the estimation of stratified atmospheric motions and on power law regularization functionals for motion estimation. Patrick is an Inria researcher since 2008. He is a member of the Aspi group.
Pierre Hellier “Non linear registration of MRI acquisition of different subjects”
In Pierre’s thesis, cosupervised
with Christian Barillot, we have proposed an efficient and robust method for medical image
registration. The method extends in 3D hierarchical motion estimators. The thesis has been defended
in 2000. Pierre was Inria researcher from 2002 to 2011. He is now a researcher at technicolor.
During Souleyman’s Post-doc (2010-2012), we explored wavelets based motion estimator
and large scales optimal control techniques. Souleyman is now assistant professor in Mathematics
at the University of La Rochelle. We still work together on the mathematical analysis of uncertainty
formulation of fluid flow dynamics.
Lionel Oisel “3D dense reconstruction of complex scenes”
In Lionel’s thesis, co-supervised with Luce
Morin, we have proposed a matching technique for the estimation of dense disparity maps. This thesis
has been defended in November 1998. Lionel is now research manager at technicolor, where he leads
a group of 30 peoples on multimedia analysis for the improvement of the content production workflow.
Nicolas Papadakis “Variational techniques for the racking of curves and motion fields”
The aim of Nicolas’s thesis consisted in studying variational techniques defined from data assimilation principles for the tracking of curves and motion fields. The technique has been also extended to the estimation of low order dynamical systems. The thesis has been defended in November of 2007. Nicolas is a CNRS Researcher at the Institut de Mathématiques de Bordeaux.
Christoph Papin “Tracking and characterization of convective cells”
In the thesis of Christophe, we
investigated the tracking of convective clouds systems through level-sets. This thesis co-supervised
with Patrick Bouthemy has been defended in December 1999. Christophe is researcher at Kodak
Valentin Resseguier , " Oceanic models under uncertainty and image assimilation".
Valentin defended is PhD in January 2017. I was granted through an Ifremer/Inria grant and was co-supervised by
B. Chapron and myself. His subject of study concerns the definition of stochastic dynamics models for the upper oceanic streams. We intend
to use high resolution satellite images to calibrate the parameters of the corresponding dynamics under
Cordelia Robinson , "Image assimilation for the numerical recontruction of wake flows".
In the thesis of Cordelia, we proposed a variational assimilation technique allowing the reconstruct the
complete 3d flow behind an unknown obstacle. To that purpose we relied on 3D velocity measurements captured within
several 2D sheets. These observation are coupled within a data assimilation
framework with an accurate Navier Stokes solver. This thesis was co-supervised with Dominique Heitz
Yin Yang "Variational ensemble methods for image assimilation".
Yin thesis is centered on the study of
new optimal control strategies formulated through an ensemble of forecast trajectories. This technique
is computationally efficient and avoids the computation of the adjoint dynamics required in the traditional
variational techniques. The problem is defined in a reduced subspace spanned by the different samples
and introduces a time varying covariance error model. The defense has been defended the 16 of december 2014.