Catálogo de publicaciones - libros

We present a new method for integrating hierarchical levels of detail (HLOD) with occlusion culling. The algorithm refines the HLOD hierarchy using geometric criteria as well as the occlusion information. For the refinement we use a simple model which takes into account the possible distribution of the visible pixels. The traversal of the HLOD hierarchy is optimized by a new algorithm which uses spatial and temporal coherence of visibility. We predict the HLOD refinement condition for the current frame based on the results from the last frame. This allows an efficient update of the front of termination nodes as well as an efficient scheduling of hardware occlusion queries. Compared to previous approaches, the new method improves on speed as well as image quality. The results indicate that the method is very close to the optimal scheduling of occlussion queries for driving the HLOD refinement.

Given terrain information and user preferences such as desired city size, Autopolis generates commercial and industrial centers, and grows areas surrounding these centers. The program also creates streets and highways, and assigns a land-use for each parcel. Compared with previous methods, Autopolis generates cities with better realism, greater detail in layout, and more design variety.

We present a Virtual Reality application enabling interactive, physically correct simulation of tube-like flexible objects. Our objective was to describe flexible objects by a set of parameters (length, diameter and material constants) instead of rigid geometry (triangle meshes) and to give the user the possibility to add, delete and manipulate those flexible objects in a stereo projected environment in real-time.

A challenge for virtual reality (VR) applications is to increase the realism of an observer’s visual experience. For this purpose the variation of the blur that an observer experiences in his/her vision, while he/she focuses on a particular location, can be mimicked by blurring the VR computer graphics based on a model of the blur. The blur in human vision is the result of a combination of optical and neural vision processes; namely optical refraction, non-uniform retinal sampling, and cortical magnification. Due to the complexity of the phenomenon, apparently no theoretical model of the blur has been published. In this work we model the combined effect by means of a probabilistic model of the human visual system. The results from the models match common vision experience verifying the validity of the underlying theoretical considerations. The implementation of the model for increased realism in virtual reality is illustrated by means of a rendering of a virtual reality scene, which is processed for two different acts of focusing.

Finding efficient and physically based methods to interactively simulate deformable objects is a challenging issue. The most promising methods addressing this issue are based on finite elements and multigrid solvers. However, these multigrid methods still suffer, when used to simulate large deformations, from two pitfalls, depending on the kind of grids hierarchy used. If embedded grids are used, approximating complex geometries becomes difficult, whereas when unstructured grids hierarchy is used, solving speed-up is reduced by the necessity to update coarser levels stiffness matrices. We propose a framework that combines embedded grids solving with fast remeshing. We introduce a new hierarchical mesh generator which can build a hierarchy of topologically embedded grids approximating a complex geometry. We also show how to take advantage of the knowledge of the stiffness matrix sparsity pattern to efficiently update coarse matrices. These methods are tested on interactive simulation of deformable solids undergoing large deformations.

In this paper, we describe a novel algorithm to group, label, identify and perform optical tracking of marker sets, which are grouped into two specific configurations, and whose projective invariant properties will allow obtaining a unique identification for each predefined marker pattern. These configurations are formed by 4 collinear and 5 coplanar markers. This unique identification is used to correctly recognize various and different marker patterns inside the same tracking area, in real time. The algorithm only needs image coordinates of markers to perform the identification of marker patterns. For grouping the dispersed markers that appear in the image, the algorithm uses a “divide and conquer” strategy to segment the image and give some neighborhood reference among markers.

Automatic segmentation of sub-cortical structures has great use in studying various neurodegentative diseases. In this paper, we propose a fully automatic solution to this problem through the utilization of a distribution atlas built from a set of training MR images. Our model consists of two major components: a local likelihood based active contour (LLAC) model and a guiding probabilistic atlas. The former has a very strong ability in standing out the structures that are in low contrast with the surrounding tissues. The latter has the functionality of defining and leading the segmentation procedure to capture the structure of interest. Formulated under the maximum a posterior framework, probabilistic atlas for the structure of interest, e.g. caudate, putamen, can be seamlessly integrated into the level set evolution procedure, and no thresholding step is needed for capturing the target.

The function of a protein is dependent on whether and how it can interact with various ligands. Therefore, an accurate prediction of protein-ligand interactions is paramount to understanding proteins’ biological mechanisms and hence to the development of therapeutic agents. A ligand is most likely to bind in the largest pocket on the surface of the protein. Moreover, it requires that the pocket meets certain structural and geometric criteria that allow the ligand to “anchor” in place by forming stabilizing interactions with the protein. Based on this logic, many geometry-based algorithms have been developed to predict protein-ligand interactions. Here we investigate a geometric-hashing based algorithm – to see how well it distinguishes proteins that do and do not bind a ligand, and propose enhancements that improve its robustness. We also introduce an alternative way of integrating geometric and biochemical properties of multiple binding mechanisms into a single representation.

In this paper we propose a novel approach to ventricular motion estimation and segmentation. Our method is based on a MRF formulation where an optimal intensity-based separation between the endocardium and the rest of the cardiac volume is to be determined. Such a term is defined in the spatiotemporal domain, where the ventricular wall motion is introduced to account for correspondences between the consecutive segmentation maps. The estimation of the deformations is done through a continuous deformation field (FFD) where the displacements of the control points are determined using discrete labeling approach. Principles from linear programming and in particular the Primal/Dual Schema is used to recover the optimal solution in both spaces. Promising experimental results obtained on 13 MR spatiotemporal data sets demonstrate the potentials of our method.

The teniæ coli, three longitudinal bands of muscle on the colon surface, are important anatomical structures for several virtual colonoscopy applications: colon registration, virtual dissection, navigation and polyp matching. The current detection method involves manually tracing one of the teniæ on the colon surface, a long and error-prone process. In this paper, we present a semi-automated detection method, which only requires the user to reconnect a few long segments of teniæ. This method is based on local parameterizations of the colon surface, and on the application of classical image processing algorithms on the graph created by the triangular mesh of the surface. The results, computed on four test cases, are promising.