Project Description:
Recent development of 3D scanners allows us to obtain
data of surfaces effectively. The collected data can be used to construct
3D digital surface models useful for a wide variety of applications, such
as face recognition, medical research, computer graphic, etc. For most situations,
a single scan is not enough to produce a complete model of the 3D object.
In order to obtain information of the entire object, several measurements
have to be taken from different viewpoints around the object. These multiple
scans from different directions are then integrated into one surface model.
The integration process mainly consists of two procedures:(1) surface alignment
or registration that matches overlapping areas of the multiple scans; (2)
surface merging that merges all scans to create a complete surface model.
These processes are commonly done by a qualified person whose tasks include
manual processing/segmentation of the input range images, picking up feature
points manually on every surface scans, surface registration and merging
of scans semi-automatically, visual verification of the resulting surface
model. These processes have to be repeated over and over until the resulting
merged surface does not have folds, holes and misalignment. It is indeed
a very time-consuming process. Therefore, it is of interest to look for an
automatic algorithm to
register and merge surface scans together. In this project, we propose
an automatic method to merge 3D attribute meshes using conformal parameterization.
Geodesic patch extraction is firstly carried out
to determine the overlapping region and remove the redundant region.
The surfaces are then parameterized conformally onto 2D rectangular domains.
Surface registration is performed on the conformal parameter domain to match
curvatures and the conformal factor. Finally, we propose the Partition of
Unity Merging algorithm to merge the surfaces and obtain a smooth surface
model.
Publication:
Result: