Catálogo de publicaciones - libros

In this paper, we present a framework for real-time and realistic rendering of large-scale deep ocean surface, which uses LOD scheme and height map to model water surface on CPU and takes full advantage of GPU for vertex and pixel processing. We present multi-resolution nested regular grids, as a LOD scheme, to support free and real-time navigation on large-scale sea area. Previous work neglected the wind changing effect on ocean waves, here we propose a wind model to improve the existing method of ocean wave gen-eration and satisfy the requirement of realistic simulation with wind effect. Experiment results show that our approach is efficient and can achieve interactive rates for the simulation of large-scale deep ocean surface.

It is the key issue of the terrain real-time rendering to reduce the time-spending for the level-of-detail. The concept of triangle live frame is introduced in this paper. Using this live frame, the algorithm reuses per-vertex visibility computation from previously displayed frames efficiently during continuous view-dependent level-of-detail visualizations. This algorithm decreases to a certain extent computing the distance from the viewpoint to vertex at each frame. Experimental results show that the algorithm reduces the time effectively to produce the LOD terrain.

In Virtual Reality (VR) and computer graphic fields, 3-dimension (3D) matter modeling has been developed for many years and successfully applied to many fields. However, researches on 3D energy field modeling are still not enough owing to challenges of full understanding and real-time calculation of invisible energy fields. In the visual information world, energy field modeling is becoming a new research point and should promote relevant research advancement and widen applications. In the paper, after reviewing the 3D object modeling, light field modeling is addressed from three aspects which contain light propagation characteristic, bidirectional reflectance distribution function (BRDF), sunlight transfer process between solar source and observers. Especially the quantitative radiances at ground level and aircraft/space level are presented. According to the sunlight transfer process, a research framework of Remote Sensing Image Simulation (RSIS) is proposed and an experiment is implemented. Our study shows that light field modeling can make invisible energy fields easily-understood and demonstrate an example of multi-sciences integration research.

We present a simple and fast LOD rendering technique for terrain visualization, which sufficiently makes use of the GPU capabilities. At load time, we tile the terrain geometry, compute a discrete set of LODs for each tile analog to mipmaps for texture, and cache index buffers for all the levels. At run time, the LOD level is selected on the CPU and the geometrical morphing algorithm is implemented on the GPU. Additionally, we use coherent hierarchical culling (CHC) algorithm to cull away fully occluded tiles to further accelerate the frame rates. Our approach sufficiently exploits both the spatial and temporal coherence of terrain geometry and reduces the bandwidth requirements. Our method can achieve very high frame rates especially when a player wanders through a valley in games or other applications.

Grass occurs frequently in the forest and the high performance grass generation can greatly enhance the reliability and complexity of virtual forest scenes. So the paper presents a simple method to render fields of grass, animated in the wind in real-time. We employ a series of polygons contain several grass blades to construct a basic grass unit. Animation is achieved by translate the grass units according the “wind vector” and we implement the wind effect such as a strong wind. Furthermore, we employ the rendering optimization to acquire faster frames.

In this paper, we present an algorithm using Gaussian Mixture Model (GMM) for foreground segmentation which can differentiate shadow region from objects with simple criteria. In the algorithm, we have utilized the Improved HLS (IHLS) color space model as the fundamental description for image, instead of using raw RGB data. IHLS color space has an advantage over the standard RGB space to recognize shadow region from object by utilizing luminance and saturation-weighted hue information directly, without any calculation of chrominance and luminance. By exploiting this feature in GMM, we obtain adaptive background model with good sensitivity to color changes and shadow.

This paper proposed a novel geometry compression method using spherical wavelet. Given a manifold triangle mesh with zero genus and arbitrary topology, it is globally parameterized over the unit sphere S2 in E3 firstly. At the same time, by subdividing an icosahedron and projecting all its vertices onto the unit sphere from the center, we can get a spherical triangle mesh with subdivision topology. Then we re-sampling all signals defined on the surface of the original triangle mesh at the vertices of the spherical subdivision mesh and get a set of discrete geometry signals with subdivision topology which can be compressed by using spherical wavelet.

Most existing global illumination methods are computationally very expensive when handling dynamic scenes, in which more than one object is undergoing geometric transformation. Hence, they are not suitable for use in interactive environments. In this paper, we propose a novel approach to address this problem – . It is based on update prioritization and radiosity redistribution to accelerate illuminate recomputation. Our method is developed to support rendering realistic scenes in dynamic virtual environments, where multiple dynamic objects may be moving around. It can handle occluded dynamic objects that are contributing to the scene illumination.

This paper presents a method for dividing the triangle mesh into n-edge mesh and puts forward a new mesh simplification algorithm based on n-edge mesh collapse. An n-edge mesh can be in the form of an edge, a triangle or a quadrangle, it depends on the value of ‘n’. The algorithm utilizes iterative collapse of n-edge mesh to simplify meshes and the surface error approximations are maintained using quadric error metrics. There are n-1 vertices and 2(n-1) faces which have to be collapsed during every simplification, so only few collapses are need when n becomes bigger. And this means, the time of the simplification process can be reduced. Our algorithm contains Garland’s (n=2) [5] and Pan’s (n=3) [11] cases, thus it can be regarded as the summarized algorithm of mesh simplification based on the geometry element collapse. Experimental results demonstrate the different cases, which hold different values of ‘n’ in the algorithm.

Constructing 3-D personalized face model with 2-D face photographs is an active topic in 3-D face modeling research. Presently, almost all methods focus on using a constrained equation to construct 3-D personalized face. Due to the complex structure of human face, these methods can’t delicately represent facial organ surfaces, and the computation cost of these methods is greater. In this paper, a method for reconstructing 3-D personalized face based on multi-layer and multi-region with RBFs is presented. First, the generic face model is reconstructed based on multi-layer and multi-region with RBFs, in which two layers are designed and four facial organs are separated. Then, the edges between layers are smoothed with influence factors of the layers. Finally, the facial texture is mapped onto the reconstructed personalized face model. Experiments show that this method is simple and efficient.