Real-Time Mono- and Multi-Volume Rendering of Large Medical Datasets on Standard PC Hardware
(Grimm, 2005) Grimm, Soeren
Direkte Volumenvisualisierung ist eine effizienteMethode um komplexe Strukturen volumetrischer Datens¨atze zu untersuchen. Der Hauptvorteil, verglichen zur normalen Oberfl¨achenvisualisierung, ist die M¨oglichkeit halbtransparente Visualisierungen zu generieren, dadurch erh¨alt man mehr Informationen ¨uber die r¨aumlichen Zusammenh¨ange verschiedener Strukturen. Um solch halb transparenten Visualisierungen zu erzeugen, muss eine enorme Datenmenge abgearbeitet werden. Durch immer leistungsf¨ahigere Prozessoren und durch verbesserte Aufnahmeger¨ate werden diese Datenmengen immer gr¨oßer. Diese enormen Datenmengen stellen eine große Herausforderung f¨ur derzeitige 3D Rendering Architekturen und Algorithmen dar. Die immer gr¨oßeren Datenmengen erh¨ohen die Nachfrage an 3D Visualisierung. Die herk¨ommliche 2D Visualisierung erreicht, durch die enorme Anzahl von Schichtbildern, bereits die Grenze der Benutzbarkeit. Die 3D Visualisierung wird immer mehr als eine alternative unterst¨utzende Methode von Schichtbild-Untersuchungen großer medizinischer Datens¨atze eingesetzt. In dieser Dissertation werden effiziente Verfahren zur Handhabung und dem Rendern großer volumetrischer Daten vorgestellt, welche zu einer signifi- kanten Geschwindigkeitssteigerung, durch bessere Ausn¨utzung von Prozessorund Speicher-bandbreite, fuehren. Zuerst werden Verfahren zum Adressieren und Verarbeiten eines Cache-effizienten Speicher-Layouts fuer große volumetrische Daten vorgestellt. Diese Verfahren dienen als Basis f¨ur ein komplettes hochqualitatives Raycasting-System, welches in der Lage ist große Daten bis zu 3 GB, eine Limitierung des virtuellen Adressbereichs heutiger Betriebssysteme, zu handhaben. Die Hauptbeschleunigungs-Komponenten dieses System sind eine raffinierte Caching Methode f¨ur die Gradientenberechnung in Verbindung mit einer hybriden Technik zum Ueberspringen und Entfernen transparenter Regionen, wodurch die Menge der zu verarbeitenden Daten signifikant reduziert wird. Dieses System wird dann so erweitert, dass effizientes Verarbeiten mehrerer volumetrischer Datens¨atze moeglich ist. Ein Beschleunigungsverfahren zum Rendern von Szenen, die aus mehreren volumetrischen Datensaetzen bestehen, wird vorgestellt. Die Grundidee des Verfahrens basiert auf der Unterscheidung zwischen Regionen, in denen sich mehrere Objekte ueberschneiden, die eine teure Verarbeitung erfordern, und Regionen in denen sich nur ein Objekt befindet und somit eine effizienteres Verarbeiten erlauben. Weiterhin werden V-Objects, ein Konzept zur Modellierung von Szenen, welche aus mehreren volumetrischen Objekten bestehen, vorgestellt. Es wird gezeigt, dass das Konzept der V-Objects zusammen mit direkter Volumenvisualisierung eine viel versprechende Methode zur Visualisierung medizinischer Daten ist und das es eine weiterte Moeglichkeit zum Untersuchen und Erkunden der Daten bietet. Im zweiten Teil der Dissertation wird eine Alternative zur gitter-basierten Volumengraphik vorgestellt: Vots, eine punkt-basierte Representation volumetrischer Daten. Es ist ein neues Primitive zur Modellierung, Verarbeitung und Rendern von volumetrischen Daten. Ein neues Paradigma wird praesentiert durch die Umwandlung der Daten von einer diskreten zu einer impliziten Darstellung., Direct Volume Visualization is an efficient technique to explore complex structures within volumetric data. Its main advantage, compared to standard 3D surface rendering, is the ability to perform semitransparent rendering in order to provide more information about spatial relationships of different structures. Semitransparent rendering requires to process a huge amount of data. The size of volumetric data is rapidly increasing, on the one hand due to the boost of processing power in the past years, and on the other hand due to improved capabilities of newer acquisition devices. This large data presents a challenge to current rendering architectures and techniques. The enormous data sizes introduce a growing demand for interactive 3D visualization. Conventional slicing methods already reach their limit of usability due to the enormous amount of slices. 3D visualization is more and more explored as an attractive alternative additional method for examinations of large medical data to support the necessary 2D examination. Within this dissertation a set of approaches to handle and render large volumetric data is developed, enabling significant performance improvements due to a much better utilization of the CPUs processing power and available memory bandwidth. At first, highly efficient approaches for addressing and processing of a cache efficient memory layout for volumetric data are presented. These approaches serve as a base for a full-blown high-quality raycasting system, capable of handling large data up to 3GB, a limitation imposed by the virtual address space of current consumer operating systems. The core acceleration techniques of this system are a refined caching scheme for gradient estimation in conjunction with a hybrid skipping and removal of transparent regions to reduce the amount of data to be processed. This system is extended so that efficient processing of multiple large data sets is possible. An acceleration technique for direct volume rendering of scenes, composed of multiple volumetric objects, is developed.
Accelerating the Rendering Process Using Impostors
(Jeschke, 2005) Jeschke, Stefan
Die Darstellung dreidimensionaler geometrischer Modelle zur Erzeugung glaubw ürdiger Bilder ist in der Computergrafik ein Gebiet von großem Interesse. Eine besondere Herausforderung ist hierbei die Erstellung einer flüssigen Animation mit mindestens 60 Bildern pro Sekunde für sehr komplexe Modelle. Anwendungen hierfür finden sich in vielfältigen Bereichen wie zum Beispiel in Schiffs-, Fahr-, oder Flugsimulationen, virtuellen Realit¨aten oder auch Computerspielen. Obwohl gebräuchliche Grafikhardware in den vergangenen Jahren an Leistung stark zugenommen hat, wachsen die Ansprüche nach realistischeren und damit komplexeren Modellen in noch höherem Maße. Diese Dissertation beschäftigt sich mit einem Ansatz zur beschleunigten Ausgabe solch komplexer Modelle. Es wird ausgenutzt, daß sich das Erscheinungsbild insbesondere entfernter Szenenteile über mehrere Ausgabebilder kaum ver- ändert. Diese Szenenteile werden durch vorberechnete bildbasierte Repräsentationen, sogenannte Imposter, ersetzt. Imposter bieten den Vorteil der schnelleren Darstellbarkeit bei gleichem oder zumindest ähnlichem Erscheinungsbild für einen räumlich abgegrenzten Bereich, dem sogenannten Sichtbereich. In bisherigen Ansätzen hierzu wurde jedoch die visuelle Qualit¨at der Impostor (d.h. die visuelle Unterscheidbarkeit zur Originalgeometrie) für den Sichtbereich nur unter sehr hohem Aufwand sichergestellt, und die Speicherplatzanforderungen für alle Imposter einer Szene sind oft unerw¨unscht hoch. Diese Punkte sind beim heutigen Stand der Impostertechnik als Hauptprobleme einer breiten Anwendbarkeit zu sehen. In dieser Arbeit wurden zwei neue Impostortechniken entwickelt, die auf einer Einteilung des zu repräsentierenden Szenenteils in Bildschichten mit unterschiedlichem Abstand zum Betrachter basieren. Durch die Einf¨uhrung spezieller Fehlermetriken wird die visuelle Qualität der Imposter für einen großen Sichtbereich quantifizierbar und garantierbar. Gleichzeitig können unsichtbare Szenenteile ef- fizient entfernt werden, was den Speicherbedarf für die Repr¨asentation verringert. Dabei werden keinerlei Informationen über die Struktur des originalen Szenenteils benötigt. Bei der einen Technik wird jede Bildschicht separat mit Geometrieinformation verknüpft. Hierdurch wird eine schnelle Impostererstellung sowie ein sehr geringer Speicherbedarf f¨ur entfernte Szenenteile erreicht. Bei der anderen Technik erfolgt die Verkn¨upfung der Geometrieinformation unabh¨angig von den Bildschichten. Dies reduziert die geometrische Komplexit¨at und den ben¨otigten Speicherplatz f¨ur die Repr¨asentation nahe gelegener Objekte wesentlich. Der zweite Teil der Arbeit besch¨aftigt sich mit dem effizienten Einsatz von Impostern. Das Ziel ist hierbei, durch den Impostereinsatz eine Mindestbildwiederholrate für jeden möglichen Blickpunkt in einer Szene zu garantieren und gleichzeitig den Speicherplatzbedarf f¨ur alle Imposter zu minimieren. Dazu wurde ein Algorithmus entwickelt, der automatisch Imposter und dazugeh¨orige Sichtbereiche so auswählt, daß nur solche Szenenteile als Imposter für jeden Blickpunkt repräsentiert werden, die sich hierf¨ur besonders eignen. Speziell wird dabei der Fehler bisheriger Ansätze vermieden, für benachbarte Sichtbereiche mehrere sehr ähnliche Imposter für entfernte Objekte zu generieren. Außerdem werden parallel zu Impostern weitere Darstellungsbeschleunigungsverfahren eingesetzt, was den Speicheraufwand f¨ur die Imposter weiter reduziert. Der Algorithmus ist dabei so allgemein gehalten, daß ein effizienter Impostereinsatz in beliebigen dreidimensionalen Szenen sowie auch mit unterschiedlichen Impostertechniken erm¨oglicht wird. Zusammenfassend erlauben die entwickelten Techniken and Algorithmen eine fl¨ussige Animation bei einer garantierbaren Mindestausgabebildqualit¨at sowie einen wesentlich geringeren Speicheraufwand f¨ur Imposter in einer Szene. Dies erlaubt den Einsatz von Impostern in verschiedenen Szenen und Applikationen, in denen diese Technik bisher nicht anwendbar war. - The interactive rendering of three-dimensional geometric models is a research area of big interest in computer graphics. The generation of a fluent animation for complex models, consisting of multiple million primitives, with more than 60 frames per second is a special challenge. Possible applications include ship-, driving- and flight simulators, virtual reality and computer games. Although the performance of common computer graphics hardware has dramatically increased in recent years, the demand for more realism and complexity in common scenes is growing even faster. This dissertation is about one approach for accelerating the rendering of such complex scenes. We take advantage of the fact that the appearance of distant scene parts hardly changes for several successive output images. Those scene parts are replaced by precomputed image-based representations, so-called impostors. Impostors are very fast to render while maintaining the appearance of the scene part as long as the viewer moves within a bounded viewing region, a so-called view cell. However, unsolved problems of impostors are the support of a satisfying visual quality with reasonable computational effort for the impostor generation, as well as very high memory requirements for impostors for common scenes. Until today, these problems are the main reason why impostors are hardly used for rendering acceleration. This thesis presents two new impostor techniques that are based on partitioning the scene part to be represented into image layers with different distances to the observer. A new error metric allows a guarantee for a minimum visual quality of an impostor even for large view cells. Furthermore, invisible scene parts are efficiently excluded from the representation without requiring any knowledge about the scene structure, which provides a more compact representation. One of the techniques combines every image layer separately with geometric information. This allows a fast generation of memory-efficient impostors for distant scene parts. In the other technique, the geometry is independent from the depth layers, which allows a compact representation for near scene parts. The second part of this work is about the efficient usage of impostors for a given scene. The goal is to guarantee a minimum frame rate for every view within the scene while at the same time minimizing the memory requirements for all impostors. The presented algorithm automatically selects impostors and view cells so that for every view, only the most suitable scene parts are represented as impostors. Previous approaches generated numerous similar impostors for neighboring view cells, thus wasting memory. The new algorithm overcomes this problem. The simultaneous use of additional acceleration techniques further reduces the required impostor memory and allows making best use of all available techniques at the same time. The approach is general in the sense that it can handle arbitrary scenes and a broad range of impostor techniques, and the acceleration provided by the impostors can be adapted to the bottlenecks of different rendering systems. In summary, the provided techniques and algorithms dramatically reduce the required impostor memory and simultaneously guarantee a minimum output image quality. This makes impostors useful for numerous scenes and applications where they could hardly be used before.
Generative Mesh Modeling
(Havemann, Nov 2005) Havemann, Sven
Die generative Modellierung ist ein alternativer Ansatz zur Beschreibung von dreidimensionaler Form. Zugrunde liegt die Idee, ein Modell nicht wie üblich durch eine Ansammlung geometrischer Primitive (Dreiecke, Punkte, NURBS-Patches) zu beschreiben, sondern durch Funktionen. Der Paradigmenwechsel von Objekten zu Geometrie-erzeugenden Operationen ermöglicht es, prozedurale Modelle auch prozedural zu repräsentieren. Statt das Resultat eines 3D-Konstruktionsprozesses zu speichern, kann so der Konstruktionsprozess selber repräsentiert werden. Der generative Ansatz eröffnet unter anderem gänzlich neue Perspektiven für das Wissensmanagement im 3D-Bereich. Er ermöglicht etwa, auf einen Fundus bereits gelöster Konstruktions-Aufgaben zurückzugreifen, um sie in ähnlichen, aber leicht variierten Situationen wiederverwenden zu können. Das Konstruktions-Wissen kann dazu in Form von Bibliotheken parametrisierter, Domänen-spezifischer Modellier-Werkzeuge gesammelt werden. Konkret wird dazu eine neue allgemeine Modell-Beschreibungs-Sprache vorgeschlagen, die
Interactive 3D Flow Visualization Based on Textures and Geometric Primitives
(Laramee, 2004) Laramee, Robert S.
This thesis presents research in the area of flow visualization. The theoretical framework is based on the notion that flow visualization methodology can be classified into four main areas: direct, geometric, texture-based, and feature-based flow visualization. Our work focuses on the direct, geometric, and texture-based categories, with special emphasis on texture-based approaches. After presenting the state-of-the-art, we discuss a technique for resampling of CFD simulation data. The resampling tool addresses both the perceptual problems resulting from a brute force hedgehog visualization and flow field coverage problems. These challenges are handled by giving the user control of the resolution of the resampling grid in object space and giving the user precise control of where to place the vector glyphs. Afterward, we present a novel technique for visualization of unsteady flow on surfaces from computational fluid dynamics. The method generates dense representations of time-dependent vector fields with high spatio-temporal correlation. While the 3D vector fields are associated with arbitrary triangular surface meshes, the generation and advection of texture properties is confined to image space. Frame rates of up to 60 frames per second are realized by exploiting graphics card hardware. We apply this algorithm to unsteady flow on boundary surfaces of, large, complex meshes from computational fluid dynamics composed of more than 200,000 polygons, dynamic meshes with time-dependent geometry and topology, as well as medical data. We also apply texture-based flow visualization techniques to isosurfaces. The result is a combination of two well known scientific visualization techniques, namely iso-surfacing and texture-based flow visualization, into a useful hybrid approach. Next we describe our collection of geometric flow visualization techniques including oriented streamlines, streamlets, a streamrunner tool, streamcomets, and a real-time animated streamline technique. We place special emphasis on necessary measures required in order for geometric techniques to be applicable to real-world data sets. In order to demonstrate the use of all techniques, we apply our direct, geometric, and texture-based flow visualization techniques to investigate swirl and tumble motion, two flow patterns found commonly in computational fluid dynamics (CFD). Our work presents a visual analysis of these motions across three spatial domains: 2D slices, 2.5D surfaces, and 3D.
Virtual Endoscopy for Preoperative Planning and Training of Endonasal Transsphenoidal Pituitary Surgery
(Neubauer, June 2005) Neubauer, Andre
Virtual endoscopy is the navigation of a virtual camera through anatomy, computationally reconstructed from radiological data. Virtual endoscopy mimics physical minimally invasive surgical activity and is used for diagnosis (e.g., the detection of colon polyps), planning of endoscopic interventions, postoperative assessment of surgical success and training for inexperienced endoscopists. This thesis introduces STEPS, a virtual endoscopy system designed as a planning and training tool for endonasal transsphenoidal pituitary surgery, a method used to minimally invasively remove tumors of the pituitary gland. A rigid endoscope is inserted into the nose and advanced through intracranial cavities towards the bony wall covering the pituitary gland. This bone is then opened and the tumor is removed. STEPS reconstructs the boundaries of the investigated cavities primarily using iso-surfacing in original CT data. Presegmented objects of interest can be added to the scene and displayed behind the semi-transparent isosurface. These objects (e.g., the tumor, the pituitary gland and important blood vessels) provide an augmented picture of patient anatomy to guide the surgeon, aid in planning the endoscopic approach, and help the user find an ideal site for actual surgical activity. Visual information can be further enhanced by display of rigid structures beyond the isosurface. The user can freely decide upon colors and lighting conditions. All rendering techniques applied by STEPS are completely CPU-based, ensuring a high amount of flexibility and hardware-independence. Nevertheless, rendering is efficient resulting in smooth interaction. STEPS allows free navigation through the nasal and paranasal anatomy, but can also be used to simulate the movement parameters of the rigid endoscopes. This includes simulation of surgical instruments and haptic feedback. Furthermore, STEPS allows the simulation of angled endoscopes and the simulation of barrel distortion as exhibited by real endoscope optics. This thesis gives an overview about existing techniques and applications of virtual endoscopy, introduces the field of application, and, in detail, describes STEPS, the required preprocessing, the rendering techniques and the user interface.
From Image-based Motion Analysis to Free-Viewpoint Video
(Theobalt, Christian, 2005) Theobalt, Christian
The problems of capturing real-world scenes with cameras and automatically analyzing the visible motion have traditionally been in the focus of computer vision research. The photo-realistic rendition of dynamic real-world scenes, on the other hand, is a problem that has been investigated in the field of computer graphics. In this thesis, we demonstrate that the joint solution to all three of these problems enables the creation of powerful new tools that are beneficial for both research disciplines. Analysis and rendition of real-world scenes with human actors are amongst the most challenging problems. In this thesis we present new algorithmic recipes to attack them. The dissertation consists of three parts: In part I, we present novel solutions to two fundamental problems of human motion analysis. Firstly, we demonstrate a novel hybrid approach for markerfree human motion capture from multiple video streams. Thereafter, a new algorithm for automatic non-intrusive estimation of kinematic body models of arbitrary moving subjects from video is detailed. In part II of the thesis, we demonstrate that a marker-free motion capture approach makes possible the model-based reconstruction of free-viewpoint videos of human actors from only a handful of video streams. The estimated 3D videos enable the photo-realistic real-time rendition of a dynamic scene from arbitrary novel viewpoints. Texture information from video is not only applied to generate a realistic surface appearance, but also to improve the precision of the motion estimation scheme. The commitment to a generic body model also allows us to reconstruct a time-varying reflectance description of an actor`s body surface which allows us to realistically render the free-viewpoint videos under arbitrary lighting conditions. A novel method to capture high-speed large scale motion using regular still cameras and the principle of multi-exposure photography is described in part III. The fundamental principles underlying the methods in this thesis are not only applicable to humans but to a much larger class of subjects. It is demonstrated that, in conjunction, our proposed algorithmic recipes serve as building blocks for the next generation of immersive 3D visual media.
New Techniques for the Modeling, Processing and Visualization of Surfaces and Volumes
(Rössl, Christian, 2005-07-20) Rössl, Christian
With the advent of powerful 3D acquisition technology, there is agrowing demand for the modeling, processing, and visualization ofsurfaces and volumes. The proposed methods must be efficient androbust, and they must be able to extract the essential structure ofthe data and to easily and quickly convey the most significantinformation to a human observer. Independent of the specific nature ofthe data, the following fundamental problems can be identified: shapereconstruction from discrete samples, data analysis, and datacompression.This thesis presents several novel solutions to these problems forsurfaces (Part I) and volumes (Part II). For surfaces, we adopt thewell-known triangle mesh representation and develop new algorithms fordiscrete curvature estimation, detection of feature lines, andline-art rendering (Chapter 3), for connectivity encoding (Chapter 4),and for topology preserving compression of 2D vector fields (Chapter5). For volumes, that are often given as discrete samples, we base ourapproach for reconstruction and visualization on the use of newtrivariate spline spaces on a certain tetrahedral partition. We studythe properties of the new spline spaces (Chapter 7) and presentefficient algorithms for reconstruction and visualization byiso-surface rendering for both, regularly (Chapter 8) and irregularly(Chapter 9) distributed data samples.
Importance-Driven Expressive Visualization
(Viola, 2005) Viola, Ivan
In this thesis several expressive visualization techniques for volumetric data are presented. The key idea is to classify the underlying data according to its prominence on the resulting visualization by importance value. The importance property drives the visualization pipeline to emphasize the most prominent features and to suppress the less relevant ones. The suppression can be realized globally, so the whole object is suppressed, or locally. A local modulation generates cut-away and ghosted views because the suppression of less relevant features occurs only on the part where the occlusion of more important features appears. Features within the volumetric data are classified according to a new dimension denoted as object importance. This property determines which structures should be readily discernible and which structures are less important. Next, for each feature various representations (levels of sparseness) from a dense to a sparse depiction are defined. Levels of sparseness define a spectrum of optical properties or rendering styles. The resulting image is generated by ray-casting and combining the intersected features proportional to their importance. An additional step to traditional volume rendering evaluates the areas of occlusion and assigns a particular level of sparseness. This step is denoted as importance compositing. Advanced schemes for importance compositing determine the resulting visibility of features and if the resulting visibility distribution does not correspond to the importance distribution different levels of sparseness are selected. The applicability of importance-driven visualization is demonstrated on several examples from medical diagnostics scenarios, flow visualization, and interactive illustrative visualization.
Towards Real-Time Novel View Synthesis Using Visual Hulls
(2005) Li, Ming
This thesis discusses fast novel view synthesis from multiple images taken from different viewpoints. We propose several new algorithms that take advantage of modern graphics hardware to create novel views. Although different approaches are explored, one geometry representation, the visual hull, is employed throughout our work. First the visual hull plays an auxiliary role and assists in reconstruction of depth maps that are utilized for novel view synthesis. Then we treat the visual hull as the principal geometry representation of scene objects. A hardwareaccelerated approach is presented to reconstruct and render visual hulls directly from a set of silhouette images. The reconstruction is embedded in the rendering process and accomplished with an alpha map trimming technique. We go on by combining this technique with hardware-accelerated CSG reconstruction to improve the rendering quality of visual hulls. Finally, photometric information is exploited to overcome an inherent limitation of the visual hull. All algorithms are implemented on a distributed system. Novel views are generated at interactive or real-time frame rates.
Faces And Hands- Modeling and Animating Anatomical and Photorealistic Models with Regard to the Communicative Competence of Virtual Humans
(2005) Albrecht, Irene
In order to be believable, virtual human characters must be able to communicate in a human-like fashion realistically. This dissertation contributes to improving and automating several aspects of virtual conversations. We have proposed techniques to add non-verbal speech-related facial expressions to audiovisual speech, such as head nods for of emphasis. During conversation, humans experience shades of emotions much more frequently than the strong Ekmanian basic emotions. This prompted us to develop a method that interpolates between facial expressions of emotions to create new ones based on an emotion model. In the area of facial modeling, we have presented a system to generate plausible 3D face models from vague mental images. It makes use of a morphable model of faces and exploits correlations among facial features. The hands also play a major role in human communication. Since the basis for every realistic animation of gestures must be a convincing model of the hand, we devised a physics-based anatomical hand model, where a hybrid muscle model drives the animations. The model was used to visualize complex hand movement captured using multi-exposure photography.

Browse

Recent Submissions

Results Per Page

Sort Options