A Point-Based Surface Reconstruction Pipeline

requires an accurate measurement of the human body as
well as the acquisition of clothing data for which there
is often no CAD model available. Like any measurement ,
the acquired geometry is influenced by random as well
as systematic errors. There is not only noise, outliers
and bad registration of the different depth images,
some parts of the surface are not observed at all.
Single, unconnected point samples are the natural
representation of the data delivered the high speed
body and cloths scanning device. Beside the geometry
information, a set of further values describes a
sampled surface element, or short "surfel".
The talk presents a surfel based surface reconstruction
pipeline, that was explored as an alternative to a rather
"classical" approach that was applied in the
Virtual-Tryon Project.
The problems addressed are simplification, normal
estimation, feature extraction, applications of surfel
clusters like texture acquisition and local parametrization
as well as a global parametrization also computed directly
on the point data. Given a parametrization the whole arsenal
of image processing becomes available. To name a few
applications: surface extrapolation, repair, compression
and meshing are now possible. The different algorithms
form a reconstruction pipeline that was implemented as
a part of my diploma thesis.
Point based methods are a well known alternative in the
rendering community. The talk wants to show some evidence,
that they can also be advantageous for surface reconstruction,
repair and processing.