2015

Permanent URI for this collection

https://diglib.eg.org/handle/10.2312/8346

Browse

Browsing 2015 by Issue Date

Filter results by year or month
Now showing 1 - 20 of 32
  • No Thumbnail Available
    Item
    Reinforcement learning in a Multi-agent Framework for Pedestrian Simulation
    (ProQuest, 2014-10-23) Martinez-Gil, Francisco
    This thesis proposes a new approach to pedestrian simulation based on machine learning techniques. Specifically, the work proposes the use of reinforcement learning techniques to build a decision-making modulefor pedestrian navigation. The thesis presents a multi-agent framework in which each agent is an embodied 3D agent calibrated with human features. The virtual worls is also a 3D world in which objects such as walls or doors are placed. The agents perceive their local neighborhood (objects and the rest of agents) and learn to move in this virtual world towards a place inside the environment. The thesis studies different algorithmic approaches based on reinforcement learning and analyzes the results in different scenarios. These scenarios are classic studied situations in the field of pedestrian modelling and simulation (bottlenecks, crossings inside a narrow corridor,...). The results show that the approach is capable of solving successfully the navigational problems. Besides emergent collective behaviors appear such as arch-like grouping around an exit in the bottleneck problem or lanes formation in the crossing inside a corridor scenario. The work opens a new research line in the pedestrian simulation studies which offers advantages as: - The behavioral design is in charge of the learning process and it is not coded by humans. - The agents learn independently different behaviors attending to thheir personal experiencies and interactions with the 3D world. - The learned decision-making module is computationally efficient (because the learned behavior is stored in form of a table or a linear function approximator). The approach has also limitations: - The learned behaviors can not be edited directly, making non trivial the task of implementing authoring tools. - The quality of the learned behaviors is not homogeneous. There are agents that learn very well their task but others do not. -The learned process is not controllable in terms of when and whatis learned in each moment.
  • No Thumbnail Available
    Item
    Interactive Virtual Cutting of Elastically Deformable Bodies
    (Technische Universität München, 2015) Wu, Jun
    This thesis is concerned with computational problems relevant to simulate elastically deformable bodies with changing topology. In particular, I present interactive techniques for the physically-based simulation of cuts in elastic bodies, for the collision detection among these dynamically separating objects, and for the modeling of residual stress in intact soft tissues to simulate the flap shrinkage after cutting. One motivation of this research is to enhance the functionality and performance of surgery simulators, which are becoming accepted in preoperative surgical planning and in surgical skills training, and the potential of which is yet to be fully exploited.
  • No Thumbnail Available
    Item
    Geometric Methods for Realistic Animation of Faces
    (ETH Zurich, 2015) Bermano, Amit Haim
    Realistic facial synthesis is one of the most fundamental problems in computer graphics, and has been sought after for approximately four decades. It is desired in a wide variety of fields, such as character animation for films and advertising, computer games, video teleconferencing, user-interface agents and avatars, and facial surgery planning. Humans, on the other hand, are experts in identifying every detail and every regularity or variation in proportion from one individual to the next. The task of creating a realistic human face is elusive due to this, as well as many other factors. Among which are complex surface details, spatially and temporally varying skin texture and subtle emotions that are conveyed through even more subtle motions. In this thesis, we present the most commonly practiced facial content creation process, and contribute to the quality of each of its steps. The proposed algorithms significantly increase the level of realism attained by each step and therefore substantially reduce the amount of manual labor required for production quality facial content. The thesis contains three parts, each contributing to one step of the facial content creation pipeline. In the first part, we aim at greatly increasing the fidelity of facial performance captures, and present the first method for detailed spatio-temporal reconstruction of eyelids. Easily integrable with existing high quality facial performance capture approaches, this method generates a person-specific, time-varying eyelid reconstruction with anatomically plausible deformations. Our approach is to combine a geometric deformation model with image data, leveraging multi-view stereo, optical flow, contour tracking and wrinkle detection from local skin appearance. Our deformation model serves as a prior that enables reconstruction of eyelids even under strong self-occlusions caused by rolling and folding skin as the eye opens and closes. In the second part, we contribute to the authoring step of the creation process. We present a method for adding fine-scale details and expressiveness to lowresolution art-directed facial performances. Employing a high-resolution facial performance capture system, we augment artist friendly content, such as those created manually using a rig, via marker-based capture, by fitting a morphable model to a video, or through Kinect-based reconstruction. From the high fidelity captured data, our system encodes subtle spatial and temporal deformation details specific to that particular individual, and composes the relevant ones to the desired input animation. The resulting animations exhibit compelling animations with nuances and fine spatial details that match captured performances, while preserving the artistic intent authored by the low-resolution input sequences, outperforming current state-of-the-art in example-based facial animation. The third part of the dissertation proposes to enrich digital facial content by adding a significant sense of presence. Replacing the classic 2D or 3D displaying techniques of digital content, we propose the first complete process for augmenting deforming physical avatars using projector-based illumination. Physical avatars have been long used to give physical presence to a character, both in the field of entertainment and teleconferencing. Using a human-shaped display surface provides depth cues and multiple observers with their own perspectives. Such physical avatars, however, suffer from limited movement and expressiveness due to mechanical constraints. Given an input animation, our system decomposes the motion into low-frequency motion that can be physically reproduced by a robotic head and high-frequency details that are added using projected shading. The result of our system is a highly expressive physical avatar that features facial details and motion otherwise unattainable due to physical constraints.
  • No Thumbnail Available
    Item
    Scalable Exploration of Highly Detailed and Annotated 3D Models
    (2015) Balsa Rodríguez, Marcos
    With the widespread availability of mobile graphics terminals and WebGL-enabled browsers, 3D graphics over the Internet is thriving. Thanks to recent advances in 3D acquisition and modeling systems, high-quality 3D models are becoming increasingly common, and are now potentially available for ubiquitous exploration. In current 3D repositories, such as Blend Swap, 3D Café or Archive3D, 3D models available for download are mostly presented through a few user-selected static images. Online exploration is limited to simple orbiting and/or low-fidelity explorations of simplified models, since photo- realistic rendering quality of complex synthetic environments is still hardly achievable within the real-time constraints of interactive applications, especially on on low-powered mobile devices or script-based Internet browsers. Moreover, navigating inside 3D environments, especially on the now pervasive touch devices, is a non-trivial task, and usability is consistently improved by employing assisted navigation controls. In addition, 3D annotations are often used in order to integrate and enhance the visual information by providing spatially coherent contextual information, typically at the expense of introducing visual cluttering. In this thesis, we focus on e cient representations for interactive exploration and understand- ing of highly detailed 3D meshes on common 3D platforms. For this purpose, we present several approaches exploiting constraints on the data representation for improving the streaming and rendering performance, and camera movement constraints in order to provide scalable navigation methods for interactive exploration of complex 3D environments. Furthermore, we study visualization and interaction techniques to improve the exploration and understanding of complex 3D models by exploiting guided motion control techniques to aid the user in discovering contextual information while avoiding cluttering the visualization. We demonstrate the e ectiveness and scalability of our approaches both in large screen museum installations and in mobile devices, by performing interactive exploration of models ranging from 9Mtriangles to 940Mtriangles.
  • No Thumbnail Available
    Item
    The Use of Geometric Structures in Graphics and Optimization
    (2015) Bus, Norbert
    Real-world data has a large geometric component, exhibiting significant geometric patterns. Therefore exploiting the geometric nature of data to design efficient methods has became a very important topic in several scientific fields, e.g., computational geometry, discrete geometry, computer graphics, computer vision. In this thesis we use geometric structures to design efficient algorithms for problems in two domains, computer graphics and combinatorial optimization. Part I focuses on a geometric data structure called well-separated pair decomposition and its usage for one of the most challenging problems in computer graphics, namely efficient photo-realistic rendering. One solution is the family of many-lights methods that approximate global illumination by individually computing illumination from a large number of virtual point lights (VPLs) placed on surfaces. Considering each VPL individually results in a vast number of calculations. One successful strategy to reduce computations is to group the VPLs into a small number of clusters that are treated as individual lights with respect to each point to be shaded. We use the well-separated pair decomposition of points as a basis for a data structure for pre-computing and compactly storing a set of view independent candidate VPL clusterings showing that a suitable clustering of the VPLs can be efficiently extracted from this data structure. We show that instead of clustering points and/or VPLs independently, what is required is to cluster the product-space of the set of points to be shaded and the set of VPLs based on the induced pairwise illumination. Additionally we propose an adaptive sampling technique to reduce the number of visibility queries for each product-space cluster. Our method handles any light source that can be approximated with virtual point lights (VPLs), highly glossy materials and outperforms previous state-of-the-art methods. Part II focuses on developing new approximation algorithms for a fundamental NP-complete problem in computational geometry. It focuses on the minimum hitting set problem, particularly on the case where given a set of points and a set of disks, we wish to compute the minimum-sized subset of the points that hits all disks. It turns out that efficient algorithms for geometric hitting set rely on a key geometric structure, called eps-net. We give an algorithm that uses only Delaunay triangulations to construct eps-nets of size 13.4/eps and we provide a practical implementation of a technique to calculate hitting sets in near-linear time using small sized eps-nets. Our results yield a 13.4 approximation for the hitting set problem with an algorithm that runs efficiently even on large data sets. For smaller datasets, we present an implementation of the local search technique along with tight approximation bounds for its approximation factor, yielding an (8 + eps)-approximation algorithm with running time O(n^2.34). Our work related to fundamental computational geometry problems also includes a novel dynamic convex hull algorithm for simple polygonal chains handling insertion or deletion of a point in amortized constant time.
  • No Thumbnail Available
    Item
    Algorithms for User-Guided Surface Mappings
    (ETH Zurich, 2015) Diamanti, Olga
    Computing mappings between spaces is a very general problem that appears in various forms in geometry processing. They can be used to provide descriptions or representations of shapes, or place shapes in correspondence. Their applications range from surface modeling and analysis to shape matching, morphing, attribute transfer and deformation. This thesis addresses two particular mapping problems that are of interest in the field, namely inter-surface maps and parameterizations. We focus on methods that are suitable for user-guided applications – we do not consider automatic methods, that do not leave space for the user to control the result. Existing meth- ods for the particular sub-problems that we are studying often either suffer from performance limitations, or cannot guarantee that the produced results align with the user’s intent; we improve upon the state of the art in both those respects. The first problem we study in this thesis is that of inter-surface mapping, with given sparse landmark point correspondences. We found that an efficient solution to this otherwise difficult topic emerges if one reformulates the mapping problem as a problem of finding affine combinations of points on the involved shapes. We extend the notion of standard Euclidean weighted averaging to 3D manifold shapes, and introduce a fast approximation that can be used to solve this problem much faster than the state of the art. We showcase applications of this approach in interactive attribute transfer between shapes. Next, we move on to the problem of surface parameterization. Here, we study the problem from the application point of view of surface remeshing; a popular way to generate a quadrilateral mesh for a given triangular mesh is to first compute a global parameterization, which is guided by a tangent vector field. This field then determines the directions of the quadrilateral edges on the output mesh. In order to design such a direction field, recent methods to tackle the problem are based on integer optimization problems, which often suffer from slow performance and local minima. We reformulate the problem in a way that the field design problem becomes a linear problem. We also add more flexibility by allowing for non- orthogonal directions. Still on the same problem of field-aligned surface parameterizations, we notice that the standard way of producing fields –namely, an optimization only focused iiion field smoothness– does not necessarily guarantee that the resulting quadrilateral meshing will be what the user intended in terms of edge directions. This is due to errors introduced in the post-processing of the field, during the later stages of the remeshing pipeline. This renders such fields suboptimal for user-guided meshing applications. We extend our efficient reformulation of the field design problem to generate fields that are guaranteed to not introduce such further errors, and thus make sure that the users obtain the expected results. Additionally, we allow users more flexible control, by supporting assignment of partial constraints for only some of the directions.
  • No Thumbnail Available
    Item
    3D Reconstruction and Rendering from High Resolution Light Fields
    (ETH Zurich, 2015) Kim, Changil
    This thesis presents a complete processing pipeline of densely sampled, high resolution light fields, from acquisition to rendering. The key components of the pipeline include 3D scene reconstruction, geometry-driven sampling analysis, and controllable multiscopic 3D rendering. The thesis first addresses 3D geometry reconstruction from light fields. We show that dense sampling of a scene attained in light fields allows for more robust and accurate depth estimation without resorting to patch matching and costly global optimization processes. Our algorithm estimates the depth for each and every light ray in the light field with great accuracy, and its pixel-wise depth computation results in particularly favorable quality around depth discontinuities. In fact, most operations are kept localized over small portions of the light field, which by itself is crucial to scalability for higher resolution input and also well suited for efficient parallelized implementations. Resulting reconstructions retain fine details of the scene and exhibit precise localization of object boundaries. While it is the key to the success of our reconstruction algorithm, the dense sampling of light fields entails difficulties when it comes to the acquisition and processing of light fields. This raises a question of optimal sampling density required for faithful geometry reconstruction. Existing works focus more on the alias-free rendering of light fields, and geometry-driven analysis has seen much less research effort. We propose an analysis model for determining sampling locations that are optimal in the sense of high quality geometry reconstruction. This is achieved by analyzing the visibility of scene points and the resolvability of depth and estimating the distribution of reliable estimates over potential sampling locations. A light field with accurate depth information enables an entirely new approach to flexible and controllable 3D rendering. We develop a novel algorithm for multiscopic rendering of light fields which provides great controllability over the perceived depth conveyed in the output. The algorithm synthesizes a pair of stereoscopic images directly from light fields and allows us to control stereoscopic and artistic constraints on a per-pixel basis. It computes non-planar 2D cuts over a light field volume that best meet described constraints by minimizing an energy functional. The output images are synthesized by sampling light rays on the cut surfaces. The algorithm generalizes for multiscopic 3D displays by computing multiple cuts. The resulting algorithms are highly relevant to many application scenarios. It can readily be applied to 3D scene reconstruction and object scanning, depth-assisted segmentation, image-based rendering, and stereoscopic content creation and post-processing, and can also be used to improve the quality of light field rendering that requires depth information such as super-resolution and extended depth of field.
  • No Thumbnail Available
    Item
    Realtime Face Tracking and Animation
    (2015) Bouaziz, Sofien
    Capturing and processing human geometry, appearance, and motion is at the core of computer graphics, computer vision, and human-computer interaction. The high complexity of human geometry and motion dynamics, and the high sensitivity of the human visual system to variations and subtleties in faces and bodies make the 3D acquisition and reconstruction of humans in motion a challenging task. Digital humans are often created through a combination of 3D scanning, appearance acquisition, and motion capture, leading to stunning results in recent feature films. However, these methods typically require complex acquisition systems and substantial manual post-processing. As a result, creating and animating high-quality digital avatars entails long turn-around times and substantial production costs. Recent technological advances in RGB-D devices, such as Microsoft Kinect, brought new hopes for realtime, portable, and affordable systems allowing to capture facial expressions as well as hand and body motions. RGB-D devices typically capture an image and a depth map. This permits to formulate the motion tracking problem as a 2D/3D non-rigid registration of a deformable model to the input data. We introduce a novel face tracking algorithm that combines geometry and texture registration with pre-recorded animation priors in a single optimization. This led to unprecedented face tracking quality on a low cost consumer level device. The main drawback of this approach in the context of consumer applications is the need for an offline user-specific training. Robust and efficient tracking is achieved by building an accurate 3D expression model of the user’s face who is scanned in a predefined set of facial expressions. We extended this approach removing the need of a user-specific training or calibration, or any other form of manual assistance, by modeling online a 3D user-specific dynamic face model. In complement of a realtime face tracking and modeling algorithm, we developed a novel system for animation retargeting that allows learning a high-quality mapping between motion capture data and arbitrary target characters. We addressed one of the main challenges of existing example-based retargeting methods, the need for a large number of accurate training examples to define the correspondence between source and target expression spaces. We showed that this number can be significantly reduced by leveraging the information contained in unlabeled data, i.e. facial expressions in the source or target space without corresponding poses. Finally, we present a novel realtime physics-based animation technique allowing to simulate a large range of deformable materials such as fat, flesh, hair, or muscles. This approach could be used to produce more lifelike animations by enhancing the animated avatars with secondary effects. We believe that the realtime face tracking and animation pipeline presented in this thesis has the potential to inspire numerous future research in the area of computer-generated animation. Already, several ideas presented in thesis have been successfully used in industry and this work gave birth to the startup company faceshift AG.
  • No Thumbnail Available
    Item
    Illustrative Visualization of Medical Data Sets
    (2015-01-29) Lawonn, Kai
    The aim of an illustrative visualization method is to provide a simplified representation of a complex scene or object. Concave and convex regions are emphasized and the surface complexity is reduced by omitting unnecessary information. This abstraction is often preferred over fully illuminated scenes in a multitude of applications. This thesis presents an overview of the state-of-the-art for feature lines. For this, the mathematical background of the differential geometry will be presented. Furthermore, this background will be adapted to triangulated surface meshes. This thesis will also present a smoothing algorithm, which smooths initial curves on triangulated surfaces. Additionally, an evaluation is shown to assess the quality of feature lines on anatomical data sets. Based on the evaluation, two conclusions in the field for medical applications were derived. From this point, this thesis presents two solutions in the field of illustrative visualization for medical data sets. A novel line drawing technique will be presented to illustrate surfaces. According to different requirements, this technique will be extended for molecular surfaces. In the field of vessel visualization with embedded blood flow, an adaptive visualization method will be introduced. This technique will also be extended to animated blood flow. Finally, this thesis shows different illustrative visualization concepts, which can be applied in various fields for depicting surface information.
  • No Thumbnail Available
    Item
    Development and Improvement of Optical Tracking Methods towards Registering the Deformations of 3D Non- Rigid Bodies in Real Time for Augmented Reality Applications
    (Servicio de Publicaciones. Univesidad de Navarra, 2015-03-27) Leizea, Ibai
    Augmented Reality (AR) is a technology that aims to embed virtual objects in the real world, showing to the user the set of objects (virtual and real) as a single world. For that purpose, it is necessary to offer a perfect alignment between virtual and real objects, which increases the effectiveness of AR. The solution to this problem is known as tracking. The object tracking consists in determining at any time the position and orientation of the camera relative to the scene. Optical sensors are most commonly used to overcome the tracking problem due to their low cost implementation. However, it is often difficult to provide robustness, accuracy and low computational cost at the same time. This thesis tackles the improvement and development of the main optical tracking techniques, primarily focused on detecting the deformations of the bodies. First, it has been achieved the tracking of rigid and non-rigid planar surfaces through a monocular camera, and then, the object deformation estimation with a more complex device as a RGB-D camera has been developed. Surface tracking systems such as those based on markers have the problem of not being able to handle occlusions. Thus, this thesis raises a new marker design that offers robustness against occlusions. Furthermore, in order to handle the deformations of surfaces, a solution that recovers the camera pose and the non-rigid surface simultaneously is proposed. Continuing with the deformation handling, it has also developed a robust tracking system for reconstructing the 3D shape of deformable objects using two different physical formulations. One offers a correct physical behaviour with a low computational cost, whereas the other achieves higher levels of accuracy at the expense of higher processing time. In addition, all the presented solutions have the common factor that all are executed in real time, which is a key property for a fluently visual feedback of an AR application.
  • No Thumbnail Available
    Item
    Sampled and Prefiltered Anti-Aliasing on Parallel Hardware
    (2015-05) Thomas, Auzinger
    A fundamental task in computer graphics is the generation of two-dimensional images. Prominent examples are the conversion of text or three-dimensional scenes to formats that can be presented on a raster display. Such a conversion process—often referred to as rasterization or sampling—underlies inherent limitations due to the nature of the output format. This causes not only a loss of information in the rasterization result, which manifests as reduced image sharpness, but also causes corruption of the retained information in form of aliasing artifacts. Commonly observed examples in the final image are staircase artifacts along object silhouettes or Moiré-like patterns. The main focus of this thesis is on the effective removal of such artifacts—a process that is generally referred to as anti-aliasing. This is achieved by removing the offending input information in a filtering step during rasterization. In this thesis, we present different approaches that either minimize computational effort or emphasize output quality. We follow the former objective in the context of an applied scenario from medical visualization. There, we support the investigation of the interiors of blood vessels in complex arrangements by allowing for unrestricted view orientation. Occlusions of overlapping blood vessels are minimized by automatically generating cut-aways with the help of an occlusion cost function. Furthermore, we allow for suitable extensions of the vessel cuts into the surrounding tissue. Utilizing a level of detail approach, these cuts are gradually smoothed with increasing distance from their respective vessels. Since interactive response is a strong requirement for a medical application, we employ fast sample-based anti-aliasing methods in the form of visibility sampling, shading supersampling, and post-process filtering. We then take a step back and develop the theoretical foundations for anti-aliasing methods that follow the second objective of providing the highest degree of output quality. As the main contribution in this context, we present exact anti-aliasing in the form of prefiltering. By computing closed-form solutions of the filter convolution integrals in the continuous domain, we circumvent any issues that are caused by numerical integration and provide mathematically correct results. Together with a parallel hidden-surface elimination, which removes all occluded object parts when rasterizing three-dimensional scenes, we present a ground-truth solution for this setting with exact anti-aliasing. We allow for complex illumination models and perspective-correct shading by combining visibility prefiltering with shading sampling and generate a ground-truth solution for multisampling anti-aliasing. All our aforementioned methods exhibit highly parallel workloads. Throughout the thesis, we present their mapping to massively parallel hardware architectures in the form of graphics processing units. Since our approaches do not map to conventional graphics pipelines, we implement our approach using general-purpose computing concepts. This results in decreased runtime of our methods and makes all of them interactive.
  • No Thumbnail Available
    Item
    Path Sampling Techniques for Efficient Light Transport Simulation
    (2015-06) Georgiev, Iliyan
    Reproducing the interactions between light and matter in a physically accurate way can significantly improve the realistic appearance of synthetic images, however such effects can be very computationally expensive to simulate. Pressed by strict requirements on image quality and visual realism, industrial applications have recently moved away from using legacy rasterization-based rendering solutions to fully embrace physically-based Monte Carlo methods. This dramatic shift has rekindled the interest in developing new and robust light transport simulation algorithms that can efficiently handle a wide range of scenes with complex materials and lighting – a problem that we address in this thesis. State-of-the-art Monte Carlo methods solve the global illumination problem by sampling random light transport paths in the scene via ray tracing. We analyze the efficiency of these methods, devise new path sampling techniques for rendering surface and volumetric light scattering, and develop novel means of leveraging illumination coherence via path reuse. This results in several practical rendering algorithms that produce images with less noise and remain more resilient to variations in the scene configuration than existing methods. The improved efficiency of these algorithms comes from the use of new and diverse sampling techniques, each specialized for handling a different set of lighting effects. Their robustness is due to the adaptive combination of these techniques in a way that preserves their individual strengths.
  • No Thumbnail Available
    Item
    Capacitive Sensing and Communication for Ubiquitous Interaction and Environmental Perception
    (2015-06-08) Grosse-Puppendahl, Tobias
    During the last decade, the functionalities of electronic devices within a living environment constantly increased. Besides the personal computer, now tablet PCs, smart household appliances, and smartwatches enriched the technology landscape. The trend towards an ever-growing number of computing systems has resulted in many highly heterogeneous human-machine interfaces. Users are forced to adapt to technology instead of having the technology adapt to them. Gathering context information about the user is a key factor for improving the interaction experience. Emerging wearable devices show the benefits of sophisticated sensors which make interaction more efficient, natural, and enjoyable. However, many technologies still lack of these desirable properties, motivating me to work towards new ways of sensing a user's actions and thus enriching the context. In my dissertation I follow a human-centric approach which ranges from sensing hand movements to recognizing whole-body interactions with objects. This goal can be approached with a vast variety of novel and existing sensing approaches. I focused on perceiving the environment with quasi-electrostatic fields by making use of capacitive coupling between devices and objects. Following this approach, it is possible to implement interfaces that are able to recognize gestures, body movements and manipulations of the environment at typical distances up to 50cm. These sensors usually have a limited resolution and can be sensitive to other conductive objects or electrical devices that affect electric fields. The technique allows for designing very energy-efficient and high-speed sensors that can be deployed unobtrusively underneath any kind of non-conductive surface. Compared to other sensing techniques, exploiting capacitive coupling also has a low impact on a user's perceived privacy. In this work, I also aim at enhancing the interaction experience with new perceptional capabilities based on capacitive coupling. I follow a bottom-up methodology and begin by presenting two low-level approaches for environmental perception. In order to perceive a user in detail, I present a rapid prototyping toolkit for capacitive proximity sensing. The prototyping toolkit shows significant advancements in terms of temporal and spatial resolution. Due to some limitations, namely the inability to determine the identity and fine-grained manipulations of objects, I contribute a generic method for communications based on capacitive coupling. The method allows for designing highly interactive systems that can exchange information through air and the human body. I furthermore show how human body parts can be recognized from capacitive proximity sensors. The method is able to extract multiple object parameters and track body parts in real-time. I conclude my thesis with contributions in the domain of context-aware devices and explicit gesture-recognition systems.
  • No Thumbnail Available
    Item
    Scene Reconstruction from Multi-Scale Input Data
    (TU Darmstadt ULB, 2015-06-18) Fuhrmann, Simon
    Geometry acquisition of real-world objects by means of 3D scanning or stereo reconstruction constitutes a very important and challenging problem in computer vision. 3D scanners and stereo algorithms usually provide geometry from one viewpoint only, and several of the these scans need to be merged into one consistent representation. Scanner data generally has lower noise levels than stereo methods and the scanning scenario is more controlled. In image-based stereo approaches, the aim is to reconstruct the 3D surface of an object solely from multiple photos of the object. In many cases, the stereo geometry is contaminated with noise and outliers, and exhibits large variations in scale. Approaches that fuse such data into one consistent surface must be resilient to such imperfections. In this thesis, we take a closer look at geometry reconstruction using both scanner data and the more challenging image-based scene reconstruction approaches. In particular, this work focuses on the uncontrolled setting where the input images are not constrained, may be taken with different camera models, under different lighting and weather conditions, and from vastly different points of view. A typical dataset contains many views that observe the scene from an overview perspective, and relatively few views capture small details of the geometry. What results from these datasets are surface samples of the scene with vastly different resolution. As we will show in this thesis, the multi-resolution, or, "multi-scale" nature of the input is a relevant aspect for surface reconstruction, which has rarely been considered in literature yet. Integrating scale as additional information in the reconstruction process can make a substantial difference in surface quality. We develop and study two different approaches for surface reconstruction that are able to cope with the challenges resulting from uncontrolled images. The first approach implements surface reconstruction by fusion of depth maps using a multi-scale hierarchical signed distance function. The hierarchical representation allows fusion of multi-resolution depth maps without mixing geometric information at incompatible scales, which preserves detail in high-resolution regions. An incomplete octree is constructed by incrementally adding triangulated depth maps to the hierarchy, which leads to scattered samples of the multi-resolution signed distance function. A continuous representation of the scattered data is defined by constructing a tetrahedral complex, and a final, highly-adaptive surface is extracted by applying the Marching Tetrahedra algorithm. A second, point-based approach is based on a more abstract, multi-scale implicit function defined as a sum of basis functions. Each input sample contributes a single basis function which is parameterized solely by the sample's attributes, effectively yielding a parameter-free method. Because the scale of each sample controls the size of the basis function, the method automatically adapts to data redundancy for noise reduction and is highly resilient to the quality-degrading effects of low-resolution samples, thus favoring high-resolution surfaces. Furthermore, we present a robust, image-based reconstruction system for surface modeling: MVE, the Multi-View Environment. The implementation provides all steps involved in the pipeline: Calibration and registration of the input images, dense geometry reconstruction by means of stereo, a surface reconstruction step and post-processing, such as remeshing and texturing. In contrast to other software solutions for image-based reconstruction, MVE handles large, uncontrolled, multi-scale datasets as well as input from more controlled capture scenarios. The reason lies in the particular choice of the multi-view stereo and surface reconstruction algorithms. The resulting surfaces are represented using a triangular mesh, which is a piecewise linear approximation to the real surface. The individual triangles are often so small that they barely contribute any geometric information and can be ill-shaped, which can cause numerical problems. A surface remeshing approach is introduced which changes the surface discretization such that more favorable triangles are created. It distributes the vertices of the mesh according to a density function, which is derived from the curvature of the geometry. Such a mesh is better suited for further processing and has reduced storage requirements. We thoroughly compare the developed methods against the state-of-the art and also perform a qualitative evaluation of the two surface reconstruction methods on a wide range of datasets with different properties. The usefulness of the remeshing approach is demonstrated on both scanner and multi-view stereo data.
  • No Thumbnail Available
    Item
    Data-driven Approaches for Interactive Appearance Editing
    (2015-06-22) Nguyen, Chuong H.
    This thesis proposes several techniques for interactive editing of digital content and fast rendering of virtual 3D scenes. Editing of digital content - such as images or 3D scenes - is difficult, requires artistic talent and technical expertise. To alleviate these difficulties, we exploit data-driven approaches that use the easily accessible Internet data (e. g., images, videos, materials) to develop new tools for digital content manipulation. Our proposed techniques allow casual users to achieve high-quality editing by interactively exploring the manipulations without the need to understand the underlying physical models of appearance. First, the thesis presents a fast algorithm for realistic image synthesis of virtual 3D scenes. This serves as the core framework for a new method that allows artists to fine tune the appearance of a rendered 3D scene. Here, artists directly paint the final appearance and the system automatically solves for the material parameters that best match the desired look. Along this line, an example-based material assignment approach is proposed, where the 3D models of a virtual scene can be "materialized" simply by giving a guidance source (image/video). Next, the thesis proposes shape and color subspaces of an object that are learned from a collection of exemplar images. These subspaces can be used to constrain image manipulations to valid shapes and colors, or provide suggestions for manipulations. Finally, data-driven color manifolds which contain colors of a specific context are proposed. Such color manifolds can be used to improve color picking performance, color stylization, compression or white balancing.
  • No Thumbnail Available
    Item
    Interactive Integrated Exploration and Management of Visualization Parameters
    (TU Wien, 2015-07) Mindek, Peter
    Visualization algorithms are parameterized to offer universality in terms of handling various data types, showing different aspects of the visualized data, or producing results useful for domain experts from different fields. Hence, input parameters are an important aspect of the visualization process. Their exploration and management are tasks which enable the visualization reusability, portability, and interdisciplinary communication. With increasing availability of visualization systems, which are suitable for a great variety of tasks, their complexity increases as well. This usually involves many input parameters necessary for the meaningful visualization of data. Multiple input parameters form parameter spaces which are too large to be explored by brute-force. Knowing the properties of a parameter space is often beneficial for improving data visualization. Therefore, it is important for domain experts utilizing data visualization to have tools for automatic parameter specification and for aiding the manual parameter setting. In this thesis, we review existing approaches for parameter-space visualization, exploration, and management. These approaches are used with a great variety of underlying algorithms. We focus on their applicability to visualization algorithms. We propose three methods solving specific problems arising from the fact that the output of a visualization algorithm is an image, which is challenging to process automatically and often needs to be analyzed by a human. First, we propose a method for the exploration of parameter-spaces of visualization algorithms. The method is used to understand effects of combinations of parameters and parts of the internal structure of the visualization algorithms on the final image result. The exploration is carried out by specifying semantics for localized parts of the visualization images in the form of positive and negative examples influenced by a set of input parameters or parts of the visualization algorithm itself. After specifying the localized semantics, global effects of the specified components of the visualization algorithm can be observed. The method itself is independent from the underlying algorithm. Subsequently, we present a method for managing image-space selections in visualizations and automatically link them with the context in which they were created. The context is described by the values of the visualization parameters influencing the output image. The method contains a mechanism for linking additional views to the selections, allowing the user an effective management of the visualization parameters whose effects are localized to certain areas of the visualizations. We present various applications for the method, as well as an implementation in the form of a library, which is ready to be used in existing visualization systems. Our third method is designed to integrate dynamic parameters stored during a multiplayer video game session by the individual participating players. For each player, the changing parameter values of the game describe their view of the gameplay. Integrating these multiple views into a single continuous visual narrative provides means for effective summarization of gameplays, useful for entertainment, or even gameplay analysis purposes by semi-professional or professional players. We demonstrate the utility of our approach on an existing video game by producing a gameplay summary of a multiplayer game session. The proposed method opens possibilities for further research in the areas of storytelling, or at a more abstract level, parameter integration for visual computing algorithms.
  • No Thumbnail Available
    Item
    Sketching free-form poses and movements for expressive character animation
    (2015-07-02) Guay, Martin
    Free-form animation allows for exaggerated and artistic styles of motions such as stretching character limbs and animating imaginary creatures such as dragons. Creating these animations requires tools flexible enough to shape characters into arbitrary poses, and control motion at any instant in time. The current approach to free-form animation is keyframing: a manual task in which animators deform characters at individual instants in time by clicking-and-dragging individual body parts one at a time. While this approach is flexible, it is challenging to create quality animations that follow high-level artistic principles—as keyframing tools only provide localized control both spatially and temporally. When drawing poses and motions, artists rely on different sketch-based abstractions that help fulfill high-level aesthetic and artistic principles. For instance, animators will draw lines of action to create more readable and expressive poses. To coordinate movements, animators will sketch motion abstractions such as semi-circles and loops to coordinate bouncing and rolling motions. Unfortunately, these drawing tools are not part of the free-form animation tool set today. The fact that we cannot use the same artistic tools for drawing when animating 3D characters has an important consequence: 3D animation tools are not involved in the creative process. Instead, animators create by first drawing on paper, and only later are 3D animation tools used to fulfill the pose or animation. The reason we do not have these artistic tools (the line of action, and motion abstractions) in the current animation tool set is because we lack a formal understanding relating the character’s shape—possibly over time—to the drawn abstraction’s shape. Hence the main contribution of this thesis is a formal understanding of pose and motion abstractions (line of action and motion abstractions) together with a set of algorithms that allow using these tools in a free-form setting. As a result, the techniques described in this thesis allow exaggerated poses and movements that may include squash and stretch, and can be used with various character morphologies. These pose and animation drafting tools can be extended. For instance, an animator can sketch and compose different layers of motion on top of one another, add twist around strokes, or turning the strokes into elastic ribbons.
  • No Thumbnail Available
    Item
    Supporting Management of Sensor Networks through Interactive Visual Analysis
    (2015-07-14) Steiger, Martin
    With the increasing capabilities of measurement devices and computing machines, the amount of recorded data grows rapidly. It is so high that manual processing is no longer feasible. The Visual Analytics approach is powerful because it combines the strengths of human recognition and vision system with today’s computing power. Different, but strongly linked visualizations and views provide unique perspectives on the same data elements. The views are linked using position on the screen as well as color, which also plays a secondary role in indicating the degree of similarity. This enables the human recognition system to identify trends and anomalies in a network of measurement readings. As a result, the data analyst has the ability to approach more complex questions such as: are there anomalies in the measurement records? What does the network usually look like? In this work we propose a collection of Visual Analytics approaches to support the user in exploratory search and related tasks in graph data sets. One aspect is graph navigation, where we use the information of existing labels to support the user in analyzing with this data set. Another consideration is the preservation of the user’s mental map, which is supported by smooth transitions between individual keyframes. The later chapters focus on sensor networks, a type of graph data that additionally contains time series data on a per-node basis; this adds an extra dimension of complexity to the problem space. This thesis contributes several techniques to the scientific community in different domains and we summarize them as follows. We begin with an approach for network exploration. This forms the basis for subsequent contributions, as it to supports user in the orientation and the navigation in any kind of network structure. This is achieved by providing a showing only a small subset of the data (in other words: a local graph view). The user expresses interest in a certain area by selecting one of more focus nodes that define the visible subgraph. Visual cues in the form of pointing arrows indicate other areas of the graph that could be relevant for the user. Based on this network exploration paradigm, we present a combination of different techniques that stabilize the layout of such local graph views by reducing acting forces. As a result, the movement of nodes in the node-link diagram is reduced, which reduces the mental effort to track changes on the screen. However, up to this point the approach suffers from one of the most prominent shortcomings of force-directed graph layouts. Little changes in the initial setup, force parameters, or graph topology changes have a strong impact on the visual representation of the drawing. When the user explores the network, the set of visible nodes continuously changes and therefore the layout will look different when an area of the graph is visited a second time. This makes it difficult to identify differences or recognize different drawing as equal in terms of topology. We contribute an approach for the deterministic generation of layouts based on pre-computed layout patches that are stitched at runtime. This ensures that even force-directed layouts are deterministic, allowing the analyst to recognize previously explored areas of the graph. In the next step, we apply these rather general purpose concepts from theory in practical applications. One of the most important network category is that of sensor networks, a type of graph data structure where every node is annotated with a time series. Such networks exist in the form of electric grids and other supply networks. In the wake of distributed and localized energy generation, the analysis of these networks becomes more and more important. We present and discuss a multi-view and multi-perspective environment for network analysis of sensor networks that integrates different data sources. It is then extended into a visualization environment that enables the analyst to track the automated analysis of the processing pipeline of an expert system. As a result, the user can verify the correctness of the system and intervene where necessary. One key issue with expert systems, which typically operate on manually written rules, is that they can deal with explicit statements. They cannot grasp terms such as “uncommon” or “anomalous”. Unfortunately, this is often what the domain experts are looking for. We therefore modify and extend the system into an integrated analysis system for the detection of similar patterns in space and in different granularities of time. Its purpose is to obtain an overview of a large system and to identify hot spots and other anomalies. The idea here is to use similar colors to indicate similar patterns in the network. For that, it is vital to be able to rely on the mapping of time series patterns to color. The Colormap-Explorer supports the analysis and comparison of different implementations of 2D color maps to find the best fit for the task. As soon as the domain expert has identified problems in the networks, he or she might want to take countermeasures to improve the network stability. We present an approach that integrates simulation in the process to perform “What-If” analysis based on an underlying simulation framework. Subsequent runs can be compared to quickly identify differences and discover the effect of changes in the network. The approaches that are presented can be utilized in a large variety of applications and application domains. This enables the domain expert to navigate and explore networks, find key elements such as bridges, and detect spurious trends early.
  • No Thumbnail Available
    Item
    Scheduling activities under spatial and temporal constraints to populate virtual urban environments
    (Université Rennes 1, 2015-07-17) Jorgensen, Carl-Johan
    Crowd simulation models usually aim at producing visually credible crowds with the intent of giving life to virtual environments. Our work focusses on generating statistically consistent behaviours that can be used to pilot crowd simulation models over long periods of time, up to multiple days. In real crowds, people's behaviours mainly depend on the activities they intend to perform. The way this activity is scheduled rely on the close interaction between the environment, space and time constraints associated with the activity and personal characteristics of individuals. Compared to the state of the art, our model better handle this interaction. Our main contributions lie in the domain of activity scheduling and path planning. First, we propose an individual activity scheduling process and its extension to cooperative activity scheduling. Based on descriptions of the environment, of intended activities and of agents' characteristics, these processes generate a task schedule for each agent. Locations where the tasks should be performed are selected and a relaxed agenda is produced. This task schedule is compatible with spatial and temporal constraints associated with the environment and with the intended activity of the agent and of other cooperating agents. It also takes into account the agents personal characteristics, inducing diversity in produced schedules. We show that our model produces schedules statistically coherent with the ones produced by humans in the same situations. Second, we propose a hierarchical path-planning process. It relies on an automatic environment analysis process that produces a semantically coherent hierarchical representation of virtual cities. The hierarchical nature of this representation is used to model different levels of decision making related to path planning. A coarse path is first computed, then refined during navigation when relevant information is available. It enable the agent to seamlessly adapt its path to unexpected events. The proposed model handles long term rational decisions driving the navigation of agents in virtual cities. It considers the strong relationship between time, space and activity to produce more credible agents' behaviours. It can be used to easily populate virtual cities in which observable crowd phenomena emerge from individual activities.
  • No Thumbnail Available
    Item
    Data-driven Modelling of Shape Structure
    (2015-09) Averkiou, Melinos
    In recent years, the study of shape structure has shown great promise, by taking steps towards exposing shape semantics and functionality to algorithms spanning a wide range of areas in computer graphics and vision. By shape structure, we refer to the set of parts that make a shape, the relations between these parts, and the ways in which they correspond and vary between shapes of the same family. These developments have been largely driven by the abundance of 3D data, with collections of 3D models becoming increasingly prominent and websites such as Trimble 3D Warehouse offering millions of free 3D models to the public. The ability to use large amounts of data inside these shape collections for discovering shape structure has made novel approaches to acquisition, modelling, fabrication, and recognition of 3D objects possible. Discovering and modelling the structure of shapes using such data is therefore of great importance. In this thesis we address the problem of discovering and modelling shape structure from large, diverse and unorganized shape collections. Our hypothesis is that by using the large amounts of data inside such shape collections we can discover and model shape structure, and thus use such information to enable structure-aware tools for 3D modelling, including shape exploration, synthesis and editing. We make three key contributions. First, we propose an efficient algorithm for co-aligning large and diverse collections of shapes, to tackle the first challenge in detecting shape structure, which is to place shapes in a common coordinate frame. Then, we introduce a method to parameterize shapes in terms of locations and sizes of their parts, and we demonstrate its application to concurrently exploring a shape collection and synthesizing new shapes. Finally, we define a meta-representation for a shape family, which models the relations of shape parts to capture the main geometric characteristics of the family, and we demonstrate how it can be used to explore shape collections and intelligently edit shapes.