Contributions published at Visualization and MultiMedia Lab (Renato Pajarola)

In den letzten Jahren haben partikelbasierte Methoden zur Simulation von Gasen und Flüssigkeiten in der Computer Graphik an Wichtigkeit gewonnen. Dies da die Repräsentation durch Partikel die Behandlung von freien Oberflächen, Spritzer, Tropfen und komplexen Interaktionen zwischen Objekten erleichtert. Partikelbasierte Methoden weisen jedoch auch Nachteile auf welche das physikalische Verhalten eines Fluids und somit das resultierende visuelle Resultat beeinträchtigen. Obwohl diese Probleme in sozusagen allen partikelbasierten Modellen präsent sind, konzentriert sich diese Dissertation auf die Hauptprobleme der Methode Smoothed Particle Hydrodynamics (SPH). Diese Dissertation beginnt mit einer Einführung in die SPH Methode und erklärt die Schwierigkeit inkompressible Flüssigkeiten zu simulieren. Im Grundmodell von SPH werden Flüssigkeiten durch kompressible Fluide approximiert was zu ungewollten Kompressionsartefakten führt. Obwohl Inkompressibilität erzwungen werden kann, repräsentiert dies den berechenmässig teuersten Teil der Methode, was der Grund ist warum SPH und partikelbasierte Methoden im Allgemeinen weniger geeignet sind um photorealistische Animation von Wasser zu erstellen. In dieser Arbeit präsentieren wir ein neues, inkompressibles Verfahren basierend auf SPH welche Inkompressibilität durch eine Prädiktor-Korrektor Methode erzwingt. Dabei werden die Informationen uber Dichteabweichungen aktiv durch das Fluid propagiert und Druckwerte angepasst, solange bis die Dichtewerte der Partikel einheitlich sind. Mit diesem Ansatz können die Berechnungskosten per Simulationsschritt niedrig gehalten und gleichzeitig ein grosser Simulationszeitschritt verwendet werden. Danach gehen wir auf die Probleme ein welche an den Zwischenflächen von mehreren Fluiden mit unterschiedlicher Dichte entstehen, sowie zwischen Fluiden und festen Objekten. Bei der Simulation von mehreren Fluiden mit dem SPH Grundmodell können Artefakte an der Zwischenfläche beobachtet werden, welche das Verhalten der Fluide negativ beeinflusst. Diese Artefakte sind unphysikalische Oberflächenspannungen sowie numerische Instabilitäten. Diese Dissertation präsentiert ein adaptiertes SPH Modell welches Diskontinuitiäten an den Zwischenflächen von mehreren Fluiden korrekt behandelt und dadurch die Probleme des Grundmodells vermeidet. Des Weiteren prä sentiert diese Arbeit ein einheitliches Modell für die Simulation von Fluiden und festen Objekten um die Interaktion zwischen unterschiedlichen Materialien zu erleichtern. In unserem Modell sind Flüssigkeiten und Gase sowie starre und elastische Körper durch Partikel repräsentiert welche Attribute mit den Objekteigenschaften tragen. Durch das Andern der Attribute können Effekte wie Schmelzen und Erstarren, sowie Vereinigung und Trennung von Objektteilen mit niedrigem Aufwand simuliert werden. Zum Abschluss stellen wir eine neue, effiziente Partikel-Verfeinerungsmethode vor um eine höhere visuelle Qualität beim Rendering von Echtzeit-Flüssigkeiten zu erreichen. Als Ausgangspunkt verwendet unsere Methode die Punktmenge der Simulation und fügt uniform neue Punkte hinzu wobei Oberflächenstrukturen akkurat beibehalten werden. Eine weitere Schwierigkeit von Partikelmethoden ist die Rekonstruktion von glatten Oberflächen. Um dies zu erreichen verwenden wir eine neue Methode, welche den Partikelschwerpunkt der Nachbarschaft bei der Rekonstruktion verwendet, und wir zeigen wie Artefakte in konkaven Regionen erfolgreich vermieden werden können. Particle-based fluid simulations have become popular in computer graphics due to their natural ability to handle free surfaces and interfaces, splashes and droplets, as well as interaction with complex boundaries. However, particle methods have some disadvantageous properties degrading the physical behavior of a simulated fluid and thus the resulting visual quality. Although these problems are present in almost any particle-based fluid solver, this dissertation addresses some of the major problems of the Lagrangian method Smoothed Particle Hydrodynamics (SPH). This thesis starts by reviewing the standard SPH model and its difficulties to satisfy the incompressibility condition. In the standard model, liquids are typically approximated by compressible fluids where pressures are determined by an equation of state, resulting in undesired compression artifacts. Although incompressibility can be enforced, it represents the most expensive part of the whole simulation process and thus renders particle methods less attractive for high quality and photorealistic water animations. In this thesis, we present a novel, incompressible fluid simulation method based on SPH. In our method, incompressibility is enforced by using a prediction-correction scheme to determine the particle pressures. For this, the information about density fluctuations is actively propagated through the fluid and pressure values are updated until the targeted density is sat- isfied. With this approach, the costs per simulation update step can be held low while still being able to use large time steps in the simulation. Next, we shift our attention to the problem of complex interactions between multiple different fluids as well as between fluids and solids. We first focus on the artifacts caused by standard SPH when simulating multiple fluids with high-density ratios. In the standard model, the smoothed quantities of particles near the fluid interface show falsified values and the physical behavior is severely affected, especially if density ratios become large. The artifacts include spurious and unphysical interface tension as well as severe numerical instabilities. In this thesis, we derive a formulation that can handle discontinuities at interfaces of multiple fluids correctly and thus avoids the problems present in standard SPH. With our concepts, an animator has full control over the behavior of multiple interacting fluids. Furthermore, we propose to represent both, fluids and solids, by particles, facilitating the interaction between the different object types. We present a unified simulation model for fluids, rigid, and elastic objects, and show how phase transitions can be modeled by only changing the attribute values of the underlying particles. New effects like merging and splitting due to melting and solidification are demonstrated, and we show that our model is able to handle coarsely sampled and even coplanar particle configurations without further treatment. Finally, we present a novel point refinement method to achieve a higher visual quality of low-resolution fluids. We introduce new algorithms to efficiently upsample an initial point set given by the physical computation. Our method features the ability to accurately preserve surface details and to reach a uniform point distribution. Another challenge is to reconstruct smooth surfaces from the particles. The visualized fluids typically suffer from bumpy surfaces related to the irregular particle distribution. In order to achieve smooth surfaces, this thesis introduces a new surface reconstruction technique based on the center of mass of the particle neighborhood. We show how artifacts in concave regions can be avoided by considering the movement of the center of mass.

A large body of empirical evidence documents that people systematically violate the key axioms of the standard economic theories of choice over time and choice under risk. In response to the evidence, new models have been developed, incorporating hyperbolic time preferences or nonlinear probability weights. These models constitute pronounced departures from standard theory, and, as a consequence, are associated with several practical issues. They often fail at predicting more than one important empirical regularity and, hence, are not able to provide a unifying explanation for anomalous behavior in intertemporal and risky choice. Motivated by these deficiencies, this thesis shows that environmental factors, such as liquidity constraints or inherent uncertainty, can bridge the gap between standard economic theory and effectively observed behavior. Anomalously-looking behavior may not necessarily be caused by exotic preferences, but can naturally arise from decision makers' rational responses to their environments. The predictions made by this approach dovetail nicely with existing empirical findings. Experimental data further support the theory's main conjectures and illustrate that it indeed has significant explanatory power. The results presented have important implications for the design of proper policy interventions and the state of standard economic theory in general.

We present a new layered, hardware-accelerated splatting algorithm for volume rendering. Layered volume splatting features the speed benefits of fast axis-aligned pre-classified sheet-buffer splatting while at the same time exhibiting display quality comparable to high-quality post-classified view-aligned sheet-buffer splatting. Additionally, we enhance the quality by using a more accurate approximation of the volume rendering integral. Commonly, the extinction coefficient of the volume rendering integral is approximated by the first two elements of its Taylor series expansion to allow for simple α-blending. In our approach we use the original, exponential extinction coefficient to achieve a better approximation. In this paper we describe the layered splatting algorithm and how it can be implemented on the GPU. We compare the results in terms of performance and quality to prior state-of-the-art volume splatting methods.

The aim of this work is the development of glooQtviewer, a generic 3D viewer based on OpenGL, a cross-platform application programming interface (API). This work also provides an overview on a few alternative frameworks that are presently available for such a project, and explains the criteria that were applied for selecting those used for this project. Furthermore, this work explores two special effects that can be achieved with OpenGL: shadowing and environment mapping. Finally, the glooQtviewer is discussed in detail and a few examples of rendering are presented.

Shape simplification and re-sampling of underlying point-sampled surfaces under userdefined error bounds is an important and challenging issue. Based on the regular triangulation of the Gaussian sphere and the surface normals mapping onto the Gaussian sphere, a Gaussian sphere based re-sampling scheme is presented that generates a non-uniformly curvature-aware simplification of the given point-sampled model. Owing to the theoretical analysis of shape isophotic error metric for did that Gaussian sphere based sampling, the proposed simplification scheme provides a convenient way to control the re-sampling results under a user-specified error metric bound. The novel algorithm has been implemented and demonstrated on several examples.

null

With the emergence of large-scale point-sampled geometry acquired by high-resolution 3D scanning devices, it has become increasingly important to develop efficient algorithms for processing such models which have abundant geometric details and complex topology in general. As a preprocessing step, surface simplification is important and necessary for the subsequent operations and geometric processing. Owing to adaptive mean-shift clustering scheme, a curvature-aware adaptive re-sampling method is proposed for point-sampled geometry simplification. The generated sampling points are non-uniformly distributed and can account for the local geometric feature in a curvature aware manner, i.e. in the simplified model the sampling points are dense in the high curvature regions, and sparse in the low curvature regions. The proposed method has been implemented and demonstrated by several examples.

We present a simple and efficient method for reconstructing triangulated surfaces from massive oriented point sample datasets. The method combines streaming and parallelization, moving least-squares (MLS) projection, adaptive space subdivision, and regularized isosurface extraction. Besides presenting the overall design and evaluation of the system, our contributions include methods for keeping in-core data structures complexity purely locally output-sensitive and for exploiting both the explicit and implicit data produced by a MLS projector to produce tightly fitting regularized triangulations using a primal isosurface extractor. Our results show that the system is fast, scalable, and accurate. We are able to process models with several hundred million points in about an hour and outperform current fast streaming reconstructors in terms of geometric accuracy.

We propose a robust, feature preserving and user-steerable mesh sampling algorithm, based on the one-to-many mapping of a regular sampling of the Gaussian sphere onto a given manifold surface. Most of the operations are local, and no global information is maintained. For this reason, our algorithm is amenable to a parallel or streaming implementation and is most suitable in situations when it is not possible to hold all the input data in memory at the same time. Using ε-nets, we analyze the sampling method and propose solutions to avoid shortcomings inherent to all localized sampling methods. Further, as a byproduct of our sampling algorithm, a shape approximation is produced. Finally, we demonstrate a streaming implementation that handles large meshes with a small memory footprint.

Continuing improvements in CPU and GPU performances as well as increasing multi-core processor and cluster-based parallelism demand for flexible and scalable parallel rendering solutions that can exploit multipipe hardware accelerated graphics. In fact, to achieve interactive visualization, scalable rendering systems are essential to cope with the rapid growth of data sets. However, parallel rendering systems are non-trivial to develop and often only application specific implementations have been proposed. The task of developing a scalable parallel rendering framework is even more difficult if it should be generic to support various types of data and visualization applications, and at the same time work efficiently on a cluster with distributed graphics cards. In this paper we introduce a novel system called Equalizer, a toolkit for scalable parallel rendering based on OpenGL which provides an application programming interface (API) to develop scalable graphics applications for a wide range of systems ranging from large distributed visualization clusters and multi-processor multipipe graphics systems to single-processor single-pipe desktop machines. We describe the system architecture, the basic API, discuss its advantages over previous approaches, present example configurations and usage scenarios as well as scalability results.

A major drawback in many robotics projects is the dependance on a specific environment and the otherwise uncertain behavior of the hardware. Simple navigation tasks like driving in a straight line can lead to a strong lateral drift over time in an unknown environment. In this paper we propose a fast and simple solution for the lateral drift problem for vision guided robots by real-time scene analysis. Without an environment-specific calibration of the robot’s drive system, we balance the differential drive speed on the fly. Therefore, a feature detector is used on consecutive images. Detected feature points determine the focus of expansion (FOE) that is used for locating and correcting the robot’s lateral drift. Results are presented for an unmodified real-world indoor environment that demonstrate that our method is able to correct most lateral drift, solely based on real-time vision processing.

This paper introduces RASTeR, a GPU based LOD technique for interactive rendering of large terrains based on a paired multi-resolution tree structure. Our approach uses regular height-data blocks and terrain independent triangle patches, which are used to efficiently subdivide the terrain data. At run time, continuous LODs can simply be generated by tiling a limited set of triangle patches, the indices to which are pre-computed, over height-field blocks, thereby minimizing the amount of data to be transferred to the graphics card. RASTeR maintains a constant frame rate through asynchronous and a priori fetching of raw or compressed elevation and texture data. The efficiency of our method is validated by presenting experimental results on large elevation models.

In this bachelor thesis, the terrain rendering system named RASTeR developped by the Vi- sualization and MultiMedia Lab (VMML) of the University of Zurich has to be extended to be able to manage different datasets in one planetary coordinate system. Furthermore, the existing presentation of the terrain on a plane has to be changed to a presentation on a globe, so that the earth can be seen in its natural form. In addition, the existing GUI should be ported from GLUT to Qt and extended by a overflight-mode. The result, an extended implementation of RASTeR, is able to manage numerous datasets at the same time. This is possible because high-resolution tiles can be, when required, connected to a basetree at run-time. The GUI has been successfully ported to Qt and extended by the requested functionality, as well as the data is now presented on a globe. However, the transitions between different tiles are visible and the selection of the appropriate level of detail doesn’t work adequately, which causes sometimes a laggy, sometimes a blurry representation of the terrain. The existing implementation didn’t allow to just extend it to the new behaviour; a big part of the application should have been rebuilt to meet the new requierments. However, it was not possible to completely achieve this within the limits of this thesis.

We present a novel, incompressible fluid simulation method based on the Lagrangian Smoothed Particle Hydrodynamics (SPH) model. In our method, incompressibility is enforced by using a predictioncorrection scheme to determine the particle pressures. For this,the information about density fluctuations is actively propagated through the fluid and pressure values are updated until the targeted density is satisfied. With this approach, we avoid the computational expenses of solving a pressure Poisson equation, while still being able to use large time steps in the simulation. The achieved results show that our predictive-corrective incompressible SPH (PCISPH) method clearly outperforms the commonly used weakly compressible SPH (WCSPH) model by more than an order of magnitude while the computations are in good agreement with the WCSPH results.

The two volume set LNCS 5875 and LNCS 5876 constitutes the refereed proceedings of the 5th International Symposium on Visual Computing, ISVC 2009, held in Las Vegas, NV, USA, in November/December 2009. The 97 revised full papers and 63 poster papers presented together with 40 full and 15 poster papers of 7 special tracks were carefully reviewed and selected from more than 320 submissions. The papers are organized in topical sections on computer graphics; visualization; feature extraction and matching; medical imaging; motion; virtual reality; face processing; reconstruction; detection and tracking; applications; and video analysis and event recognition. The 7 additional special tracks address issues such as object recognition; visual computing for robotics; computational bioimaging; 3D mapping, modeling and surface reconstruction; deformable models: theory and applications; visualization enhanced data analysis for health applications; and optimization for vision, graphics and medical imaging: theory and applications.

null