Miguel Sainz

Point-based rendering techniques

The increasing popularity of points as rendering primitives has led to a variety of different rendering algorithms, and the different implementations compare like apples to oranges. In this paper, we revisit and compare a number of recently developed point-based rendering implementations within a common testbed. Also we briefly summarize a few proposed hierarchical multiresolution point data structures and discuss their advantages. Based on a common view-dependent level-of-detail (LOD) rendering framework, we then examine different hardware accelerated point rendering algorithms. Experimental results are given with respect to performance timing and rendering quality for the different approaches. Additionally, we also compare the point-based rendering techniques to a basic triangle mesh approach.

Confetti: object-space point blending and splatting

We present Confetti, a novel point-based rendering approach based on object-space point interpolation of densely sampled surfaces. We introduce the concept of a transformation-invariant covariance matrix of a set of points which can efficiently be used to determine splat sizes in a multiresolution point hierarchy. We also analyze continuous point interpolation in object-space and we define a new class of parameterized blending kernels as well as a normalization procedure to achieve smooth blending. Furthermore, we present a hardware accelerated rendering algorithm based on texture mapping and /spl alpha/-blending as well as programmable vertex and pixel-shaders.

Points reloaded: point-based rendering revisited

The increasing popularity of points as rendering primitives has led to a variety of different rendering algorithms, and in particular the different implementations compare like apples to oranges. In this paper we revisit a number of recently developed point-based rendering implementations. We briefly summarize a few proposed hierarchical multiresolution point data structures and their advantages. Based on a common multiresolution framework we then describe and examine different hardware accelerated point rendering algorithms. Experimental results are given with respect to performance timing and rendering quality for the different approaches.

DMESH: FAST DEPTH-IMAGE MESHING AND WARPING

In this paper we present a novel and efficient depth-image representation and warping technique called DMesh which is based on a piece-wise linear approximation of the depth-image as a textured and simplified triangle mesh. We describe the application of a hierarchical multiresolution triangulation method to generate adaptively triangulated depth-meshes efficiently from reference depth-images, discuss depth-mesh segmentation methods to avoid occlusion artifacts and propose a new hardware accelerated depth-image rendering technique that supports per-pixel weighted blending of multiple depth-images in real-time. Applications of our technique include image-based object representations and the use of depth-images in large scale walk-through visualization systems.

Recovering 3D metric structure and motion from multiple uncalibrated cameras

An optimized linear factorization method for recovering both the 3D geometry of a scene and the camera parameters from multiple uncalibrated images is presented. In a first step, we recover a projective approximation using a well-known iterative approach. Then we are able to upgrade from a projective to a Euclidean structure by computing the projective distortion matrix in a way that is analogous to estimating the absolute quadric. Using singular value decomposition (SVD) as the main tool, and from a study of the ranks of the matrices involved in the process, we are able to enforce an accurate Euclidean reconstruction. Moreover, in contrast to other approaches, our process is essentially a linear one and does not require an initial estimation of the solution. Examples of synthetic and real data reconstructions are presented.

Automatic creation of three-dimensional avatars

Highly accurate avatars of humans promise a new level of realism in engineering and entertainment applications, including areas such as computer animated movies, computer game development interactive virtual environments and tele-presence. In order to provide high-quality avatars, new techniques for the automatic acquisition and creation are required. A framework for the capture and construction of arbitrary avatars from image data is presented in this paper. Avatars are automatically reconstructed from multiple static images of a human subject by utilizing image information to reshape a synthetic three-dimensional articulated reference model. A pipeline is presented that combines a set of hardware-accelerated stages into one seamless system. Primary stages in this pipeline include pose estimation, skeleton fitting, body part segmentation, geometry construction and coloring, leading to avatars that can be animated and included into interactive environments. The presented system removes traditional constraints in the initial pose of the captured subject by using silhouette-based modification techniques in combination with a reference model. Results can be obtained in near-real time with very limited user intervention.

Camera calibration of long image sequences with the presence of occlusions

Camera calibration is a critical problem in applications such as augmented reality and image based model reconstruction. When constructing a 3D model of an object from an uncalibrated video sequence, large amounts of frames and self occlusions of parts of the object are common and difficult problems. In this paper we present a fast and robust algorithm that uses a divide and conquer strategy to split the video sequence into sub-sequences containing only the most relevant frames. Then a robust stratified linear based algorithm is able to calibrate each of the subsequences to a metric structure and finally the subsequences are merged together and a final nonlinear optimization refines the solution. Examples of real data reconstructions are presented.

Object-space point blending and splatting

We present a novel point-based rendering approach based on object-space point interpolation. We introduce the concept of a transformation-invariant covariance matrix of a set of points to efficiently determine splat sizes in a multiresolution hierarchy. We analyze continuous point interpolation in object-space, and define a new class of parametrized blending kernels to achieve smooth blending. Furthermore, we present a hardware accelerated rendering algorithm based on α-texture mapping and α-blending.

Fast Body-Cloth simulation with moving humanoids

In this paper we present a very fast method for body-cloth animation. The usual bottle-neck in cloth simulation performance is collision detection, which becomes more difficult to solve when a complex geometry, like a human body, is involved. Recent image based methods, that use depth images to detect collisions, usually relays on CPU for collision correction. In our work we implement a GPU based simulation that takes care both of cloth simulation and body-cloth collisions when the humanoid is moving. Our solution is based on a hierarchic depth map structure. A high frame rate is obtained with both structured and unstructured cloth meshes with thousands of particles.

Photorealistic Image Based Objects from Uncalibrated Images

In this paper we present a complete pipeline for image based modeling of objects using a camcorder or digital camera. Our system takes an uncalibrated sequence of images recorded around a scene, it automatically selects a subset of keyframes and then it recovers the underlying 3D structure and camera path. The following step is a volumetric scene reconstruction performed using a hardware accelerated voxel carving approach. From the recovered voxelized volume we obtain the depth images for each of the reference views and then we triangulate them following a Restricted Quadtree meshing scheme. During rendering, we use a highly optimized approach to combine the available information from multiple overlapped reference images generating a generate a full 3D photo-realistic reconstruction. The final reconstructed models can be rendered in real time very efficiently, making them very suitable to be used to enrich the content of large virtual environments. CR Categories: I.3.5 [Computer Graphics]: Computational Geometry and Object Modeling—; I.4.8 [Image Processing and Computer Vision]: Scene Analysis—;