Research Interests

Includes, but is not limited to, the following topics: 3D modeling from images, shape from silhouettes, visual hulls, segmentation, silhouette extraction, tracking, computer graphics and rendering, statistical modeling, virtual reality, computational geometry, real-time computing, parallel and distributed computing...

Masters Research

I have obtained a Masters Degree (DEA) in Computer Science, Computer Imagery, and Computer Vision, at the GRAVIR research lab at the INRIA Rhône-Alpes (Grenoble, France), with the iMAGIS team, with supervisors Edmond Boyer (MOVI team) and Jean-Marc Hasenfratz (iMAGIS team). During the masters, entitled Reconstruction 3D en temps-réel pour la réalité virtuelle (Real-time 3D Reconstruction for Virtual Reality Applications), I implemeted and tested several 3D shape from silhouette algorithms and learned about Visual Hull properties and algorithms. The goal is to create, from several calibrated cameras which see the silhouettes of objects in the scene, a virtual shape model of those objects. Such models give a virtual representation of real objects, which can be rendered and manipulated in the context of a virtual, computer generated world, a specialized virtual reality technique referred to as Mixed Reality. I have implemented several reconstruction algorithms, and proposed a new technique to optimally vectorize 2D silhouettes, a necessary step before many geometric, surface type reconstructions.

PhD Research

I am completing a PhD with topics similar to the masters', with the MOVI team in the GRAVIR lab at the INRIA Rhônes-Alpes (Grenoble, France), with advisors Edmond Boyer and Radu Horaud. First, I took part in the creation and evaluation of a new reconstruction method from silhouettes, which has been published in 2003 (CVPR, A Hybrid Approach for Computing the Visual Hull of Complex Objects). I then proposed a new shape-from-silhouette method, which computes the visual hull as an exact viewing cone intersection, which was published in 2003 (BMVC, Exact Polyhedral Visual Hulls). Being also interested in the real-time aspect for 3D reconstruction, I participated in the creation of parallel reconstruction methods on a dedicated PC grid, based on the previous methods. This work has lead to two publications in a conference and a workshop (IPT 2004, Marker-less Real Time 3D Modeling for Virtual Reality and 3DSensor-CVPR 2004 A Distributed Approach for Real-Time 3D Modeling). Recently, I have investigated how silhouette formation can be modelled in images and the impact this has on 3D reconstruction. This has lead me to propose a new statistical 3D reconstruction method from uncertain silhouettes, to appear in ICCV 2005. During the PhD, I also promoted my work in France, through two national conference publications (Orasis 2003, Une approche hybride pour calculer l'enveloppe visuelle d'objets complexes, and Orasis 2005 Modélisation 3D temps-réelle distribuée). The new family of statistical shape reconstruction method seems promising and leaves many possibilities open for further exploration. This is why I wish to continue with this topic and all connected research topics, which includes 2D and 3D segmentation, tracking of rigid or deformable objects.