Introduction - Deformable Surface 3D Reconstruction from Monocular Images - page 2

(a) (b) (c)
Figure 1.1: Augmented Reality. (a) A deformed piece of paper. (b) The illustration has been virtually
removed. (c) It has been replaced by the properly deformed and reshaded logo of a conference Pilet et al.
[ 2008 ].
Figure 1.2: The entertainment industry could use 3D reconstruction from video for different appli-
cations. For example, animating the cloth of virtual characters could be guided by video, thus limiting
manual intervention Bhat et al. [ 2003 ]. Courtesy of K. Bhat.
the surgeons' work much harder. As shown in Fig. 1.4 , having a full 3D representation of the
organ's surface recovered from the images, or an augmented view of the organs, would be of
great use. In particular, it could help the surgeons orient themselves more easily and improve
their perception of where their tools are with respect to the relevant surfaces.
In all these cases, using more than one camera greatly simplifies matters by allowing the
use of multi-view stereo techniques. Consequently, there has been increasing interest in relying on
stereo to recover the complex shapes of clothes Starck and Hilton [ 2007 ]. This constitutes a very
hard application, since the folds and wrinkles of clothing produce many self-occlusions, and make
simple matching techniques fail. Various matching techniques have been proposed, such as spherical
matching Starck and Hilton [ 2005 ], as well as different shape representations such as Laplacian
surfaces de Aguiar et al. [ 2007 ]. The resulting motion capture techniques have been successfully ap-
plied with specific markers printed on the garments White et al. [ 2007 ], and, more recently, without
any such markers Bradley et al. [ 2008 ]. However, using multiple cameras also makes the deploy-
ment of the corresponding system much harder since the cameras have to be synchronized and