Authors:
Van-Toan Cao
;
Trung-Thien Tran
and
Denis Laurendeau
Affiliation:
Faculty of Science and Engineering and Laval University, Canada
Keyword(s):
Non-rigid Registration, Deformation, Correspondence, Alignment, Parameterization.
Related
Ontology
Subjects/Areas/Topics:
Computer Vision, Visualization and Computer Graphics
;
Geometric Computing
;
Geometry and Modeling
;
Modeling and Algorithms
;
Scene and Object Modeling
;
Surface Modeling
Abstract:
In this paper we present a method for non-rigid registration of meshes. The method aligns two surfaces of
deformable objects by automatically separating the deformation into a single global transformation and other
local deformations. The local deformations are found by applying the deformation model proposed by Sumner
(Sumner et al., 2007) that previous methods have used to register two surfaces. However, we specify a rigid
transformation for each node of the deformation graph using a rotation matrix and a translation vector. With
this model, unit quaternions of rotation matrices are paramterized using an homeomorphic relation between
the 4D unit sphere and the 3D projective space. Therefore, the number of unknowns is reduced by half
compared to the original models based on affine transformations and the optimization process is less complex.
We demonstrate the efficiency of the proposed method by aligning the surfaces of data sets without any prior
knowledge and assumptions about th
e deformation between the two surfaces.
(More)