Content uploaded by Ronald Peikert
Author content
All content in this area was uploaded by Ronald Peikert
Content may be subject to copyright.
Available via license: CC BY-NC-ND 3.0
Content may be subject to copyright.
Available online at www.sciencedirect.com
Procedia Computer Science 7 (2011) 253–255
The European Future Technologies Conference and Exhibition 2011
Current Trends for 4D Space-Time Topology for Semantic Flow
Segmentation
Kreˇ
simir Matkovi´
ca, Alan Leˇ
za, Helwig Hauserb, Armin Pobitzer b, Holger Theisel c,
Alexander Kuhn c, Mathias Ottoc, Ronald Peikertd, Benjamin Schindler d, Raphael Fuchs d
aVRVis Research Center, Vienna, Austria
bUniversity of Bergen, Norway
cUniversity of Magdeburg, Germany
dETH Zurich, Switzerland
1. Introduction
Recent advances in computing and simulation technology promote the simulation of time-dependent flows, i.e., flows
where the velocity field changes over time. The simulation of time-dependent flow is a more realistic approximation of
natural phenomena and it represents an invaluable tool for scientists and practitioners in multiple disciplines, including
meteorology, vehicle design, and medicine. Flow visualization, a subfield of scientific visualization, is one of several
research areas which deal with the analysis of flows. There are many methods for the analysis of steady flows, but
the extension to the time-dependent case is not straight forward. The SemSeg project, a FET-Open project in the 7th
Framework programme, attempts to provide a solution for the semantic segmentation of time-dependent flows. It aims
at the formulation of a sound theoretical mechanism to describe structural features in time-dependent flow. In this
paper, we briefly summarize recent research results from the SemSeg project. Several different approaches are pursued
in the project, including methods based on the finite-time Lyapunov exponent (FTLE), methods based on vector field
topology (VFT), and interactive visual analysis (IVA) methods. Uncertainty visualization and the interactive evaluation
of methods are helping in evaluating the results.
2. Selected Current Research in SemSeg
Vector field topology (VFT) is a well-established methodology for analyzing and visualizing velocity field datasets.
Its power lies in the automatic and parameter-free extraction of flow structures that have proven meaningful in a wide
range of application domains. Its limitation, however, is the restriction to steady flow. This was addressed, in parts, by
using moving frames of reference for the local application of VFT, for example, based on optimality criteria [1].
Uncertain vector field topology. Flow data is usually obtained either by measuring the actual physical process or
by simulation. In both cases the results contain an inherent uncertainty, that evolves from measurement inaccuracies
as well as from different parameter setting or using different simulation models. We present a technique to visualize
the global uncertainty in steady 3D vector fields using a topological approach. We start from an existing approach
for the 2D case and extend this into 3D space. In addition, we develop an acceleration strategy to detect sink and
source distributions. Having these distributions, we use overlaps of their corresponding volumes to identify separating
structures and saddles. As part of the approach, we introduce uncertain saddle and boundary switch connectors and
1877-0509/$ – see front matter © Selection and peer-review under responsibility of FET11 conference organizers and published by Elsevier B.V.
doi:10.1016/j.procs.2011.09.013
254 K. Matkovi´c et al. / Procedia Computer Science 7 (2011) 253–255
provide algorithms to extract them. For the visual representation, we use multiple direct volume renderings and test
our method on a number of synthetic as well as real-world datasets [2]. A popular alternative to VFT is based on the
finite-time Lyapunov exponent (FTLE). The ridges of this scalar quantity indicate flow structures, so-called Lagrangian
coherent structures (LCS). These structures can be visualized without an explicit ridge extraction step. Analyzing LCS
and the related geometrical trajectory properties allows for a deeper insight into the structure of time dependent flow
phenomena.
Studying separation structures based on FTLE. For a thorough analysis such as the study of bifurcations, ridges
are necessary. We compare several ridge definitions and developed a novel method [3] which exploits properties of
FTLE fields. It avoids one order of numerical differentiation and results in higher quality ridge surfaces. Coherent
flow behavior, and separation, as its dual behavior, are important features in flow fields. The currently most common
approach to detect such behavior makes use of the FTLE, which is, informally speaking, the maximal local separation
rate. We developed a filter that distinguishes between separation due to different flow directions and from separation
due to different flow speeds [4]. The filter follows the geometric intuition behind the original definition of FTLE.
Computing the FTLE without the flow map gradient. Existing methods for the computation of FTLE either rely on an
approximation of the flow map gradient, or they use frequent renormalization steps during the integration process. This
poses a number of challenges which are due to the fact that this gradient shows an exponential growing or shrinking
with integration time. We developed a novel method for computing FTLE of 2D unsteady vector fields which uses
exclusively measures that are linearly growing with integration time. Using this approach the evolving FTLE can be
reformulated as an ODE and obtained by a numerical integration of a 7D vector field.
Scale-space aware analysis of time-dependent dynamics. Commonly used feature extraction methods tend to have
a rich response for complex, e.g., turbulent flows. Using classical image processing approaches, based, e.g., on size or
the vicinity of two features, the output is reduced, which, however, does not necessarily respect the underlying physics.
We propose the use of Proper Orthogonal Decomposition (POD) to decompose the flow field according to its kinetic
energy to construct an approximation of the field, representing the largest energy-scales, and apply feature extraction
in the sequel [5]. This guarantees that the large-scale dynamics of the flow are represented in the final output. By using
a scale-space approach to both FTLE computation and ridge extraction [6], we were able to address the dilemma of
either excessive computing time or error due to gradient underestimation.
Interactive Visual Analysis. Besides the above described automatic methods we also use interactive visual analysis
(IVA) in order to understand flow phenomena. As simulation datasets get large, the fully automatic methods are no
longer sufficient. We have developed a pathline explorer, a tool for the interactive visual analysis of time-dependent
flows. The main idea is to compute pathlines and pathline attributes (some of them are scalar and others are functions
of time or of the position along the pathline), and then to interactively explore the new dataset. The first tests were done
using an exhaust manifold case from automotive industry [7]. Interactive visual analysis will be also used to compare
results from various automatic flow segmentation methods.
3. Outlook and Acknowledgments
Based on the here presented research, we are looking forward to new research questions, including the analysis of
tensors on unsteady flows as well as streakline-based approaches.
The SemSeg project acknowledges the financial support of the Future and Emerging Technologies (FET) programme
within the Seventh Framework Programme for Research of the European Commission, under FET-Open grant number
226042. More information about the SemSeg project is available from the URL www.SemSeg.eu.
References
[1] R. Fuchs, J. Kemmler, B. Schindler, F. Sadlo, H. Hauser, R. Peikert, Toward a Lagrangian Vector Field Topology, Computer Graphics Forum 29
(3) (2010) 1163–1172.
[2] M. Otto, T. Germer, H.-C. Hege, H. Theisel, Uncertain 2d vector field topology, Computer Graphics Forum 29 (2).
[3] B. Schindler, R. Peikert, R. Fuchs, H. Theisel, Ridge Concepts for the Visualization of Lagrangian Coherent Structures, in: R. Peikert, H. Hauser,
H. Carr, R. Fuchs (Eds.), Topological Methods in Data Analysis and Visualization II, Springer, 2011.
[4] A. Pobitzer, R. Peikert, R. Fuchs, H. Theisel, H. Hauser, Filtering of FTLE for Visualizing Spatial Separation in Unsteady 3D Flow, in: R.
Peikert, H. Hauser, H. Carr, R. Fuchs (Eds.), Topological Methods in Data Analysis and Visualization II, Springer, 2011.
K. Matkovi´c et al. / Procedia Computer Science 7 (2011) 253–255 255
[5] A. Pobitzer, M. Tutkun, Ø. Andreassen, R. Peikert, R. Fuchs, H. Hauser, Energy-scale aware feature extraction for flow visualization, Computer
Graphics Forum 30 (3).
[6] R. Fuchs, B. Schindler, R. Peikert, Scale-Space Approaches to FTLE Ridges, in: R. Peikert, H. Hauser, H. Carr, R. Fuchs (Eds.), Topological
Methods in Data Analysis and Visualization II, Springer, 2011.
[7] A. Lez, A. Zajic, K. Matkovic, A. Pobitzer, M. Mayer, H. Hauser, Interactive exploration and analysis of pathlines in flow data, in: in: WSCG
2011 Full Papers Proceedings, 2011.