Je prŽsenterai un aperu de
l'histoire de quelques problmes en vision par ordinateur gŽomŽtrique, couvrant
trois sicles : calcul de pose, reconstruction tridimensionnelle ˆ partir de
silhouettes, gŽomŽtrie Žpistolaire, tenseur trifocal,
reconstruction projective, auto-calibrage, reconstruction de surfaces de
rŽvolution, reconstruction utilisant des miroirs, etc. Beaucoup de ces
problmes ont ŽtŽ ŽtudiŽs au XIXe sicle, voire plus t™t É
Imagine
an object whose 3D shape -called template-
is known. The object now undergoes
a deformation, taking a deformed shape, and is imaged once by a digital camera. Under which hypotheses can the deformed shape be recovered from
the template and the image? This is
one of the fundamental questions arising
in Shape-from-Template, the process
of reconstructing the 3D shape
of a specific object from a single image and a deformation
prior. I will first present the models and algorithms proposed to study and solve Shape-from-Template over the last decade.
I will then show how Shape-from-Template may facilitate augmented reality in the specific
case of laparoscopic myomectomy,
and shall discuss the specific difficulties of this type of applications.É
I
will talk about recent results from a number of people in my group on Riemannian manifolds in computer vision. In many Vision problems Riemannian manifolds come up as a natural
model. Data related to a problem
can be naturally
represented as a point on a Riemannian
manifold. This talk will give
an intuitive introduction to Riemannian manifolds,
and show how they can be applied in many
situations.
Manifolds
of interest include the
manifold of Positive Definite matrices and the Grassman Manifolds, which have a role in object recognition and
classification, and the Kendall shape manifold, which represents the shape of 2D objects. Of particular interest is the question of when one can define positive-definite kernels on Riemannian manifolds. This would
allow the
application of kernel techniques of SVMs, Kernel FDA, dictionary learning etc directly on the manifold.
David FOFI, Pr. Le2i - IUT
Le Creusot
L'objectif
principal de ces travaux est l'Žlaboration d'un systme de vision pour la
surveillance aŽrienne mettant en Ïuvre des camŽras de
types diffŽrents. Le systme de vision dit hybride est constituŽ d'une camŽra
omnidirectionnelle (fish-eye),
permettant une vision globale de la scne, et d'une camŽra PTZ, permettant la
visŽe et le zoom sur une cible. Il devra pouvoir identifier et suivre des zones
d'intŽrt au sol (cibles) et permettre une reconstruction tridimensionnelle de
la scne ou de ses ŽlŽments par stŽrŽoscopie hybride. Nous aborderons les
problŽmatiques suivantes :
-
suivi de cibles par vision omnidirectionnelle ;
-
auto-calibrage de camŽra PTZ ;
-
suivi prŽdictif par vision hybride.