Daniel Meneveaux

Using vanishing points for camera calibration and coarse 3D reconstruction from a single image

In this paper, we show how to calibrate a camera and to recover the geometry and the photometry (textures) of objects from a single image. The aim of this work is to make it possible walkthrough and augment reality in a 3D model reconstructed from a single image. The calibration step does not need any calibration target and makes only four assumptions: (1) the single image contains at least two vanishing points, (2) the length (in 3D space) of one line segment (for determining the translation vector) in the image is known, (3) the principle point is the center of the image, and (4) the aspect ratio is fixed by the user. Each vanishing point is determined from a set of parallel lines. These vanishing points help determine a 3D world coordinate system Ro. After having computed the focal length, the rotation matrix and the translation vector are evaluated in turn for describing the rigid motion between Ro and the camera coordinate system Rc. Next, the reconstruction step consists in placing, rotating, scaling, and translating a rectangular 3D box that must fit at best with the potential objects within the scene as seen through the single image. With each face of a rectangular box, a texture that may contain holes due to invisible parts of certain objects is assigned. We show how the textures are extracted and how these holes are located and filled. Our method has been applied to various real images (pictures scanned from books, photographs) and synthetic images.

A New Partitioning Method for Architectural Environments

Computing global illumination in a moderate time for complex environments and walking through them is one of the challenges in computer graphics. To meet this goal, a preprocessing is necessary. This preprocessing consists in partitioning the environment into cells and determining visibility between these cells. Most of the existing partitioning methods rely on the Binary Space Partitioning technique (BSP) which can be easily applied to axial environments. But for non axial scenes the BSP has an important complexity of O(n^3) in time to construct a tree of size at worst O(n^2), n beeing the total number of input polygons. Moreover this technique entails a too important number of cells which do not necessarily fit with the topology of the environment. We propose in this paper a model-based partitioning method which can be applied to non axial buildings. It results in a few number of cells fitting at best with the environment topology. The problem of visibilty calculation is not addressed in this paper.

Consistency constraints and 3D building reconstruction

Virtual architectural (indoor) scenes are often modeled in 3D for various types of simulation systems. For instance, some authors propose methods dedicated to lighting, heat transfer, acoustic or radio-wave propagation simulations. These methods rely in most cases on a volumetric representation of the environment, with adjacency and incidence relationships. Unfortunately, many buildings data are only given by 2D plans and the 3D needs varies from one application to another. To face these problems, we propose a formal representation of consistency constraints dedicated to building interiors and associated with a topological model. We show that such a representation can be used for: (i) reconstructing 3D models from 2D architectural plans (ii) detecting automatically geometrical, topological and semantical inconsistencies (iii) designing automatic and semi-automatic operations to correct and enrich a 2D plan. All our constraints are homogeneously defined in 2D and 3D, implemented with generalized maps and used in modeling operations. We explain how this model can be successfully used for lighting and radio-wave propagation simulations.

A Framework for Automatically Recovering Object Shape, Reflectance and Light Sources from Calibrated Images

In this paper, we present a complete framework for recovering an object shape, estimating its reflectance properties and light sources from a set of images. The whole process is performed automatically. We use the shape from silhouette approach proposed by R. Szeliski (1993) combined with image pixels for reconstructing a triangular mesh according to the marching cubes algorithm. A classification process identifies regions of the object having the same appearance. For each region, a single point or directional light source is detected. Therefore, we use specular lobes, lambertian regions of the surface or specular highlights seen on images. An identification method jointly (i) decides what light sources are actually significant and (ii) estimates diffuse and specular coefficients for a surface represented by the modified Phong model (Lewis, 1994). In order to validate our algorithm efficiency, we present a case study with various objects, light sources and surface properties. As shown in the results, our system proves accurate even for real objects images obtained with an inexpensive acquisition system.

A Hierarchical Topology-Based Model for Handling Complex Indoor Scenes

This paper presents a topology‐based representation dedicated to complex indoor scenes. It accounts for memory management and performances during modelling, visualization and lighting simulation. We propose to enlarge a topological model (called generalized maps) with multipartition and hierarchy. Multipartition allows the user to group objects together according to semantics. Hierarchy provides a coarse‐to‐fine description of the environment. The topological model we propose has been used for devising a modeller prototype and generating efficient data structure in the context of visualization, global illumination and 1 GHz wave propagation simulation. We presently handle buildings composed of up to one billion triangles.

Out of core photon-mapping for large buildings

This paper describes a new scheme for computing out-of-core global illumination in complex indoor scenes using a photon-mapping approach. Our method makes use of a cells-and-portals representation of the environment for preserving memory coherence and storing rays or photons. We have successfully applied our method to various buildings, composed of up to one billion triangles. As shown in the results, our method requires only a few hundred megabytes of memory for tracing more than 1.6 billion photons in large buildings.

Connectivity compression in an arbitrary dimension

This paper presents a general lossless connectivity compression scheme for manifolds in any dimension with arbitrary cells, orientable or not, with or without borders. Relying on a generic topological model called generalized maps, our method performs a region-growing traversal of its primitive elements while describing connectivity relations with symbols. The set of produced symbols is compressed using standard data compression techniques. These algorithms have been successfully applied to various models (surface, tetrahedral and hexahedral meshes), showing the efficiency and genericity of the proposed scheme.

Synchronisation and Load Balancing for Parallel Hierarchical Radiosity of Complex Scenes on a Heterogeneous Computer Network

In this paper we propose a SPMD parallel hierarchical radiosity algorithm relying on a novel partitioning method which may apply to any kind of architectural scene. This algorithm is based on MPI (Message Passing Interface), a communication library which allows the use of either a heterogeneous set of concurrent computers or a parallel computer or both. The database is stored on a common directory and accessed by all the processors (through NFS in case of a network of computers). As the objective is to handle complex scenes such as building interiors, to cope with the problem of memory size, only a subset of the database resides in memory of each processor. This subset is determined with the help of a partitioning into 3D cells, clustering and visibility calculations. A graph expressing visibility between the resulting clusters is determined, partitioned (with a new method based on classification of K‐means type) and distributed amongst all the processors. Each processor is responsible for gathering energy (using the Gauss‐Seidel method) only for its subset of clusters. In order to reduce the disk transfers due to downloading these subsets of clusters, we use an ordering strategy based on the traveling salesman algorithm. Dynamic load balancing relies on a task stealing approach while termination is detected by configuring the processors into a ring and moving a token around this ring. The parallel iterative resolution is of group iterative type. Its mathematical convergence is proven in the appendix.

Technical Section: Cosine lobes for interactive direct lighting in dynamic scenes

Cosine functions have been widely used for representing Bidirectional Reflection Distribution Functions (BRDF) such as Lambert, Phong and Lafortune models. They are well suited to represent both high and low frequency signals. However, they are difficult to use with visibility and incident radiance. In most systems, the rendering equation terms are thus estimated using various methods. Several interactive rendering systems rather rely on the projection of each term onto orthonormal basis functions such as spherical harmonics or wavelets. These methods are easier to handle since the integration becomes a dot product. However, these functions are also subject to several drawbacks. For instance the number of coefficients is high for the representation of high frequency phenomena; the pre-computation time required for projecting each term of the rendering equation cannot be neglected. This paper demonstrates that cosine lobes can be generalized to visibility and incoming radiance with several advantages. First, cosine lobes do not form an orthonormal basis of functions and the number of parameters remains naturally adapted to the signal. This is very interesting for complex and high frequency functions: glossy BRDF or small light sources for instance. We also use this property for reducing the number of parameters as the computation goes along. Second, Lambert, Phong and Lafortune BRDF models are already used in many rendering systems. Since they already rely on this representation, no transformation into other types of model is necessary. This paper shows how it is possible to rapidly integrate the product of cosine lobes. As a demonstration of our methodology, we propose an interactive rendering system for direct lighting, including soft shadows and spatially varying materials.

Efficient clustering and visibility calculation for global illumination

Using a radiosity method to estimate light inter-reflections within large scenes still remains a difficult task. The two main reasons are: (i) the computations entailed by the radiosity method are time consuming and (ii) the large amount of memory needed is very large. In this paper, we address this problem by proposing a new clustering technique as well as a new method of visibility computation for complex indoor scenes. Our clustering algorithm groups polygons that are close to each other in each room (or corridor) of the building. It relies on a classification method of k-mean type and allows the use of several kinds of distance functions. For each group of polygons (or cluster), we estimate the set of potentially visible clusters with the help of openings such as doors or windows. This computation results in a graph in which the nodes correspond to clusters and the edges express visibility relationships between the corresponding clusters. We use this graph for computing radiosity in complex buildings while reducing both the amount of memory needed and the computing time. Our global illumination method is a MWRA (multi-wavelet radiosity algorithm). Unlike cluster-based radiosity methods, our MWRA does not approximate (but computes accurately) the light energy impinging or leaving a cluster after multiple reflections. We provide results for 3 different test scenes containing a high number of polygons.