This diploma thesis deals with the processing of point clouds captured by the Kinect sensor from single position. As part of this thesis an application was designed, which is able to register and reconstruct surface using selected methods. The registration of overlapping frames is based on computation of key points and their FPFH histograms from which the estimation of correspondence is computed. This estimation is then refined and redundant point filtering is performed. Surface is reconstructed from the registered and modified point cloud using Greedy Projection Triangulation. All computations are performed offline. The output of this application is textured polygonial model and an image for texture creation. With assumption of correctly set parameters the results are in a good quality for creation of virtual tours and visualization.

This thesis deals with methods of edema measuring. In my work, I designed hardware and software solutions for a device reconstructing surface of the limb part. Purpose of my work is evaluating a discussing possibilities of this device. For making 3D reconstruction of scene from 2D images I chose reconstruction method Structure from motion. For acquisition of 2D images a device controlled by Arduino platform was constructed, the whole device is realized in program language Matlab. In the end of the thesis is described, how to improve the device for using in real conditions.

No description available.

京都大学 / 新制・課程博士 / 博士(工学) / 甲第23881号 / 工博第4968号 / 新制||工||1776(附属図書館) / 京都大学大学院工学研究科機械理工学専攻 / (主査)教授 蓮尾 昌裕, 教授 松原 厚, 教授 鈴木 基史 / 学位規則第4条第1項該当 / Doctor of Philosophy (Engineering) / Kyoto University / DFAM

No description available.

This dissertation presents an approach to reconstruct degenerate near-planar road surface in three-dimensional (3D) while automatically detect road defects. Three techniques are developed in this dissertation to establish the proposed approach. The first technique is proposed to reconstruct the degenerate near-planar road surface into 3D from one camera. Unlike the traditional Structure from Motion (SfM) technique which has the degeneracy issue for near-planar object 3D reconstruction, the uniqueness of the proposed technique lies in the use of near-planar characteristics of surfaces in the 3D reconstruction process, which solves the degenerate road surface reconstruction problem using only two images. Following the accuracy-enhanced 3D reconstructed road surface, the second technique automatically detects and estimates road surface defects. As the 3D surface is inversely solved from 2D road images, the detection is achieved by jointly identifying irregularities from the 3D road surfaces and the corresponding image information, while clustering road defects and obstacles using a mean-shift algorithm with flat kernel to estimate the depth, size, and location of the defects. To enhance the physics-driven automatic detection reliability, the third technique proposes and incorporates a self-supervised learning structure with data-driven Convolutional Neural Networks (CNN). Different from supervised learning approaches which need labeled training images, the road anomaly detection network is trained by road surface images that are automatically labeled based on the reconstructed 3D surface information. In order to collect clear road surface images on the public road, a road surface monitoring system is designed and integrated for the road surface image capturing and visualization. The proposed approach is evaluated in both simulated environment and through real-world experiments. The parametric study of the proposed approach shows the small error of the 3D road surface reconstruction influenced by different variables such as the image noise, camera orientation, and the vertical movement of the camera in a controlled simulation environment. The comparison with traditional SfM technique and the numerical results of the proposed reconstruction using real-world road surface images then indicate that the proposed approach effectively reconstructs high quality near-planar road surface while automatically detects road defects with high precision, accuracy, and recall rates without the degenerate issue. / Doctor of Philosophy / Road is one of the key infrastructures for ground transportation. A good road surface condition can benefit mainly on three aspects: 1. Avoiding the potential traffic accident caused by road surface defects, such as potholes. 2. Reducing the damage to the vehicle initiated by the bad road surface condition. 3. Improving the driving and riding comfort on a healthy road surface. With all the benefits mentioned above, it is important to examine and check the road surface quality frequently and efficiently to make sure that the road surface is in a healthy condition. In order to detect any road surface defects on public road in time, this dissertation proposes three techniques to tackle the road surface defects detection problem: First, a near-planar road surface three-dimensional (3D) reconstruction technique is proposed. Unlike traditional 3D reconstruction technique, the proposed technique solves the degenerate issue for road surface 3D reconstruction from two images. The degenerate issue appears when the object reconstructed has near-planar surfaces. Second, after getting the accuracy-enhanced 3D road surface reconstruction, this dissertation proposes an automatic defects detection technique using both the 3D reconstructed road surface and the road surface image information. Although physics-based detection using 3D reconstruction and 2D images are reliable and explainable, it needs more time to process these data. To speed up the road surface defects detection task, the third contribution is a technique that proposes a self-supervised learning structure with data-driven Convolutional Neural Networks (CNN). Different from traditional neural network-based detection techniques, the proposed combines the 3D road information with the CNN output to jointly determine the road surface defects region. All the proposed techniques are evaluated using both the simulation and real-world experiments. Results show the efficacy and efficiency of the proposed techniques in this dissertation.

En simulation de fluides, il est très difficile d'obtenir des simulations contenant un haut niveau de détails efficacement dû à la complexité des phénomènes étudiés. Beaucoup de travaux se sont attaqués à ce problème afin de développer de nouvelles techniques permettant d'augmenter la résolution apparente des fluides à plus faibles coûts de calcul. Les nouvelles méthodes adaptatives ou multi-échelles ont permis de grandement améliorer la qualité visuelle des simulations de fumée et de liquides, mais certains problèmes demeurent toujours ouverts et au centre de nombreuses recherches. L'objectif de ce mémoire est d'élaborer une méthode multi-échelles afin d'augmenter la résolution apparente d'une simulation de liquide basée sur un système de particules déjà existante, un type de simulation très populaire grâce à ses propriétés de conservation d'énergie. Une telle méthode permettrait d'obtenir des simulations de résolution apparente élevée à bien moindres coûts de calcul et permettrait ainsi aux artistes d'obtenir un aperçu de leur simulation plus rapidement, tout en ayant un résultat de haute qualité. Nous présentons une méthode permettant de reconstruire une surface d'une telle simulation qui soit encline à la simulation de dynamique de surface afin d'injecter des détails de hautes fréquences occasionnés par la tension de surface. Notre méthode détecte les endroits sous-résolus de la simulation et y injecte de la turbulence grâce à de multiples oscillateurs à différentes fréquences. Les vagues à hautes fréquences injectées sont alors propagées à l'aide d'une simulation d'onde sur la surface. Notre méthode s'applique totalement en tant que post-traitement et préserve ainsi entièrement le comportement général de la simulation d'entrée tout en augmentant nettement la résolution apparente de la surface de celle-ci. / Accurately simulating the behaviour of fluids remains a difficult problem in computer graphics, and performing these simulations at a high level of detail is particularly challenging due to the complexity of the underlying dynamics of a fluid. A recent and significant body of work targets this problem by trying to augment the apparent resolution of an underlying, lower-resolution simulation, instead of performing a more costly simulation at the full-resolution. Adaptive or multi-scale methods in this area have proven successful for simulations of smoke and liquids, but no comprehensive solution exists. The goal of this thesis is to devise a new multi-scale detail-augmentation technique suitable for application atop existing particle-based fluid simulators. Particle simulations of fluid dynamics are a popular, heavily-used alternative to grid-based simulations due to their ability to better preserve energy, and no detail-augmentation techniques have been devised for this class of simulator. As such, our work would permit digital artists to perform more efficient lower-resolution particle simulations of a liquid, and then layer-on a detailed secondary simulation at a negligible cost. To do so, we present a method for reconstructing the surface of a liquid, during the particle simulation, in a manner that is amenable to high-frequency detail injection due to higher-resolution surface tension effects. Our technique detects potentially under-resolved regions on the initial simulation and synthesizes turbulent dynamics with novel multi-frequency oscillators. These dynamics result in a high-frequency wave simulation that is propagated over the (reconstructed) liquid surface. Our algorithm can be applied as a post-process, completely independent of the underlying simulation code, and so it is trivial to integrate in an existing 3D digital content creation pipeline.

Facial modelling and animation has been an active research subject in computer graphics since the 1970s. Due to extremely complex biomechanical structures of human faces and people’s visual familiarity with human faces, modelling and animating realistic human faces is still one of greatest challenges in computer graphics. Since we are so familiar with human faces and very sensitive to unnatural subtle changes in human faces, it usually requires a tremendous amount of artistry and manual work to create a convincing facial model and animation. There is a clear need of developing automatic techniques for facial modelling in order to reduce manual labouring. In order to obtain a realistic facial model of an individual, it is now common to make use of 3D scanners to capture range scans from the individual and then fit a template to the range scans. However, most existing template-fitting methods require manually selected landmarks to warp the template to the range scans. It would be tedious to select landmarks by hand over a large set of range scans. Another way to reduce repeated work is synthesis by reusing existing data. One example is expression cloning, which copies facial expression from one face to another instead of creating them from scratch. This aim of this study is to develop a fully automatic framework for template-based facial modelling, facial expression transferring and facial expression tracking from range scans. In this thesis, the author developed an extension of the iterative closest points (ICP) algorithm, which is able to match a template with range scans in different scales, and a deformable model, which can be used to recover the shapes of range scans and to establish correspondences between facial models. With the registration method and the deformable model, the author proposed a fully automatic approach to reconstructing facial models and textures from range scans without re-quiring any manual interventions. In order to reuse existing data for facial modelling, the author formulated and solved the problem of facial expression transferring in the framework of discrete differential geometry. The author also applied his methods to face tracking for 4D range scans. The results demonstrated the robustness of the registration method and the capabilities of the deformable model. A number of possible directions for future work were pointed out.

Representações computacionais de formas podem ser criadas em ferramentas CAD ou geradas a partir de um objeto físico já existente. Esta última abordagem oferece como vantagens rapidez e fidelidade ao objeto original, que são os aspectos fundamentais em muitas aplicações, como Simulações Numéricas de Equações Diferenciais Parciais e Imagens Médicas. A reconstrução (ou geração de malhas superficiais) a partir de pontos amostrados de uma superfície de um objeto é um problema clássico de representação de formas. Nesta dissertação apresentamos um vasto levantamento bibliográfico deste tipo de reconstrução, classificando e descrevendo os principais trabalhos presentes na literatura. A partir do levantamento bibliográfico, selecionamos um conjunto de algoritmos sobre os quais foram realizadas comparações teóricas e empíricas cujos resultados são apresentados. Para finalizar, apresentamos aplicações de nossas implementações em Simulação Numérica de Equações Diferenciais Parciais e processamento de Imagens / Computational representations of shapes can be developed using CAD applications or created from data acquired from a real physical object. This latter is advantageous with respect to time and fidelity to the original object which are essential to several applications, such as Numerical Simulation of Partial Differential Equations and Medical Imaging. A classical shape representation problem is that of reconstruction (or superficial mesh generation) from points sampled over the surface of an object. In this Master\'s thesis we describe a broad survey of these reconstruction methods. We focus in the classification and characterization of the main algorithms proposed in the literature. From this survey, we selected some algorithms and conducted some theoretical and practical comparisons. We conclude this work describing applications of the algorithms implemented in Numerical Simulations of Differential Partial Equations and Image Processing

Many problems in early vision can be formulated in terms of minimizing an energy or cost function. Examples are shape-from-shading, edge detection, motion analysis, structure from motion and surface interpolation (Poggio, Torre and Koch, 1985). It has been shown that all quadratic variational problems, an important subset of early vision tasks, can be "solved" by linear, analog electrical or chemical networks (Poggio and Koch, 1985). IN a variety of situateions the cost function is non-quadratic, however, for instance in the presence of discontinuities. The use of non-quadratic cost functions raises the question of designing efficient algorithms for computing the optimal solution. Recently, Hopfield and Tank (1985) have shown that networks of nonlinear analog "neurons" can be effective in computing the solution of optimization problems. In this paper, we show how these networks can be generalized to solve the non-convex energy functionals of early vision. We illustrate this approach by implementing a specific network solving the problem of reconstructing a smooth surface while preserving its discontinuities from sparsely sampled data (Geman and Geman, 1984; Marroquin 1984; Terzopoulos 1984). These results suggest a novel computational strategy for solving such problems for both biological and artificial vision systems.