Catálogo de publicaciones - libros

Subject to disturbance, a human can carry out many balance strategies, changing its posture. Virtual human animation is a challenging problem. In the present paper, we introduce a new dynamic balance control of virtual humans with multiple non coplanar frictional contacts. We formulate a constrained optimization problem (Quadratic Programming). Our virtual human can autonomously manage its balance without following imposed trajectories, in all kind of environments (floor, stairway) while being disturbed by external forces. In contrast to classical methods based on ZMP, it can use its hands to keep its balance by pressing an inclined wall. In every case, it fits its posture to ensure the best balance.

We present a novel multi-modal haptic interface for sketching and exploring the structure and properties of mathematical knots. Our interface derives from the familiar pencil-and-paper process of drawing 2D knot diagrams to facilitate the creation and exploration of mathematical knots; however, with a touch-based interface, users can also leverage their physical intuition by seeing, touching, and feeling the knots. The pure haptic component provides an intuitive interaction model for exploring knots, focusing on resolving the apparent conflict between the continuous structure of the actual knot and the visual discontinuities at occlusion boundaries. The auditory component adds redundant cues that emphasize the traditional knot crossings, where the haptic proxy crosses a visual disruption in the graphics image. Our paradigm enhances and extends traditional 2D sketching methods by exploiting both touch and sound to assist in building clearer mental models of geometry such as knot structures.

Visual sensitivity is constantly adjusting to the current visual context through processes of adaptation. These adaptive changes strongly affect all perceptual judgments and optimize visual coding for the specific properties of the scenes before us. As a result human observers “see” better when they are first allowed to adapt to a specific context. The basic form of the response changes resulting from adaptation have been studied extensively and many are known in broad outline. Here we consider the advantages of applying these changes to images, to simulate the processes of adaptation that normally occur within the observer. Matching images to the observer may obviate the need for some forms of perceptual learning and offers a number of potential benefits for interacting with visual displays.

This paper presents a new method for efficient computation of indirect lighting in local lighting environments. We establish a separation between the BRDF and the irradiance for each vertex, such that during runtime we are able to quickly reconstruct per vertex irradiance. In reconstructing irradiance, we establish an important relationship between three components: stratified irradiance shared across vertices, the fast wavelet transform, and a wavelet-based nonlinearly approximated inner product. By nonlinearly approximating the BRDF for each vertex, we demonstrate how stratified irradiance has spatial independence in the 2D Haar wavelet domain, in turn allowing for large extents of irradiance samples contributing to many vertices. By expressing irradiance in terms of shared scaling coefficients, we introduce an efficient algorithm for evaluating the inner product between the irradiance and the BRDF. Our system is tailored towards the interactive rendering of static but geometrically complex models which exhibit complex reflectance materials, capable of interactive lighting and interactive view under frame rates of 2-6 fps, ran entirely on a single CPU.

Simulation of 3D clouds is an important component in realistic modeling of outdoor scenes. In this paper, we propose a novel physically driven cumulus cloud simulation algorithm based on the similarity approach. By using the similarity approach, the overall cloud characteristics is captured with a set of constant parameters, which in turn enables decoupling of the 3D cloud simulation into the 1D vertical and the 2D horizontal simulations. As a result, the proposed cloud simulation algorithm greatly facilitates computing efficiency, general shape control and wind effect simulation, while yielding realistic visual quality.

Early culling of the invisible geometric primitives in a complex scene is valuable for efficiency in the conventional rendering pipeline.This may reduce the number of geometric primitives that will be processed in the rest of the pipeline. In this paper, we propose a conservative six degrees of freedom (DoF) incremental occlusion culling method called Delta-Horizon (H).H method is based on constructing an occlusion horizon (OH), which is a set of connected lines passing just above all visible primitives, for culling the invisible primitives beyond. Utilizing the coherence of occluders enables the incremental update of OH in consecutive frames. Although H method may work in image space, we utilize polar coordinates and build OH in object space. This not only facilitates a quick update of OH, but also overcomes the drawbacks of previous OH methods such as viewing in six DoF, occlusion culling within up-to-360-degree viewing frustums in one pass and camera zooming without additional cost.

To extract useful information from high-dimensional geometric or structural data, we must find low-dimensional projections that are informative and interesting to look at. The conventional, manual-interaction methods used for this purpose are ineffective when the dimensionality of the data is high, or when the geometric models are complex. Standard methods for determining useful low-dimensional views are either limited to discrete data, or to geometric information embedded in at most three dimensions. Since geometric data embedded in dimensions above three have distinct characteristics and visualization requirements, finding directly applicable techniques is a challenge. We present a comprehensive framework for exploring high-dimensional geometric data motivated by projection pursuit techniques. Our approach augments manual exploration by generating sets of salient views that optimize a customizable family of geometry-sensitive measures. These views serve to reveal novel facets of complex manifolds and equations.

We describe a method of extracting a simplified contour surface along with detailed normal maps from volume data in one fast and integrated process. A robust dual contouring algorithm is used for efficiently extracting a high-quality “crack-free” simplified surface from volume data. As each polygon is generated, the normal map is simultaneously generated. An underlying octree data structure reduces the search space required for high to low resolution surface normal mapping. The process quickly yields simplified meshes fitted with normal maps that accurately resemble their complex equivalents.