Information
- Publication Type: Master Thesis
- Workgroup(s)/Project(s): not specified
- Date: 2025
- TU Wien Library: AC17561836
- Open Access: yes
- First Supervisor: Renata Georgia Raidou
- Pages: 129
- Keywords: Comparative Visualization, Point Clouds, Optimal Transport, Visual Shape Analytics, Ensemble Visualization, Shape Space, Animation, Rendering, Earth Mover's Distance
Abstract
Advances in measurement technologies and 3D vision have significantly enhanced the speed and precision with which real-world objects and landscapes can be captured and reconstructed. These virtual reconstructions are relevant for surveying applications and are often encoded as point clouds, i.e., a set of 3D coordinates, possibly accompanied by additional attributes like colors or normals. Often, reconstructions of objects or landscapes are acquired over time to monitor their changes. Intuitive visualization that allows one to comprehend the shifts over time in such reconstructions could be of help, but the vast size of the data imposes challenges on comparative visualization pipelines. On the other hand, it is simpler than ever to amass numerous reconstructions of real-world objects, even for novice users. Still, outside of computationally intensive algorithms tailored to applications for the medical domain, there is a gap in approaches that allow for comparing differences within ensembles of shapes. Available algorithms outside of medicine are built upon nearest neighbor queries, which do not scale well to complex shapes and lack guidance for the comparison. Extensive ensembles of spatial data need to be delivered in a structured way to avoid time-intensive manual ordering when there is no chronological ordering implied or known. We designed and implemented a framework to support the comparative visualization of ensembles of point clouds. By utilizing the mature mathematical framework of optimal transport, we circumvent shortcomings of commonly employed nearest neighbor-based approaches and allow our method to compare a whole ensemble of reconstructions in a comprehensive representation. If there is no inherent ordering, our method enables the automatic arrangement of individual point clouds, establishing their relationships and simplifying the analysis process. We derive additional metrics about the whole ensemble, which are then used to enrich the visualization and help to detect patterns of variation within the data. By leveraging fast GPU-based implementations, we enable a smooth transition between displayed point clouds in an animation and offer visual aids that highlight the characteristics of each shape and how these change. Our method processes the data fast and provides comprehensive means to browse through a large ensemble of point clouds.
Additional Files and Images
Weblinks
BibTeX
@mastersthesis{schindler-2025-ssm,
title = "Shape shifting : a multiscale optimal transport approach to
3D point cloud comparison",
author = "Marwin Schindler",
year = "2025",
abstract = "Advances in measurement technologies and 3D vision have
significantly enhanced the speed and precision with which
real-world objects and landscapes can be captured and
reconstructed. These virtual reconstructions are relevant
for surveying applications and are often encoded as point
clouds, i.e., a set of 3D coordinates, possibly accompanied
by additional attributes like colors or normals. Often,
reconstructions of objects or landscapes are acquired over
time to monitor their changes. Intuitive visualization that
allows one to comprehend the shifts over time in such
reconstructions could be of help, but the vast size of the
data imposes challenges on comparative visualization
pipelines. On the other hand, it is simpler than ever to
amass numerous reconstructions of real-world objects, even
for novice users. Still, outside of computationally
intensive algorithms tailored to applications for the
medical domain, there is a gap in approaches that allow for
comparing differences within ensembles of shapes. Available
algorithms outside of medicine are built upon nearest
neighbor queries, which do not scale well to complex shapes
and lack guidance for the comparison. Extensive ensembles of
spatial data need to be delivered in a structured way to
avoid time-intensive manual ordering when there is no
chronological ordering implied or known. We designed and
implemented a framework to support the comparative
visualization of ensembles of point clouds. By utilizing the
mature mathematical framework of optimal transport, we
circumvent shortcomings of commonly employed nearest
neighbor-based approaches and allow our method to compare a
whole ensemble of reconstructions in a comprehensive
representation. If there is no inherent ordering, our method
enables the automatic arrangement of individual point
clouds, establishing their relationships and simplifying the
analysis process. We derive additional metrics about the
whole ensemble, which are then used to enrich the
visualization and help to detect patterns of variation
within the data. By leveraging fast GPU-based
implementations, we enable a smooth transition between
displayed point clouds in an animation and offer visual aids
that highlight the characteristics of each shape and how
these change. Our method processes the data fast and
provides comprehensive means to browse through a large
ensemble of point clouds.",
pages = "129",
address = "Favoritenstrasse 9-11/E193-02, A-1040 Vienna, Austria",
school = "Research Unit of Computer Graphics, Institute of Visual
Computing and Human-Centered Technology, Faculty of
Informatics, TU Wien",
keywords = "Comparative Visualization, Point Clouds, Optimal Transport,
Visual Shape Analytics, Ensemble Visualization, Shape Space,
Animation, Rendering, Earth Mover's Distance",
URL = "https://www.cg.tuwien.ac.at/research/publications/2025/schindler-2025-ssm/",
}
thesis