Laser-based survey techniques gives a possibility to scan a large number of points in a short period of time. However, the creation of a 3D model basing on such results remains a cumbersome and uneasy task. We present a method that aims at making this task easier. It proposes, on one hand, to monitor the survey by the image by means of a linkage between the scanner and the camera, and in addition to use morphological models of the objects being surveyed.These models expresses pieces of the architectural knowledge that history reveal. We will hereafter focus on this second aspect. The authors are involved in a research project conducted in the framework of RIAM, i.e. net of Research and Innovation in Audio-visual and Multimedia.

1 – IMAGE BASED DIGITIZING: 3D Rendering tasks are now made easier thanks to the numerous 3D graphics cards available on the marketplace. But automatic 3D objects digitizing is the key challenge for today 3D computer graphics development. The goal of this paper is to present research results and new avenues towards direct 3D digitizing with multi-views cameras, in order to design 3D objects as simply as 2D scanning. At present, classical epipolar techniques and disparity extraction don’t give quite convincing results. Moreover, 3D laser based digitizing technique provide a huge polygon number on their outputs. Most objects and scenes are now digitized by 3D classic modeling techniques, which is a very old, expensive, not automatic, and time consuming method. Recent advances in low level vision analysis gave interesting results related to Automatic neurofocalisation, and perceptive grouping, mainly thanks to Professor Burnod researches in the field of Neuro Visual hardware models. These techniques are based on multi scale hyper complex filters such as Gaussian, Laplacian derivatives, at the first vision level, and learned filters at the upper levels. These filters are able to compute highly informative points or zones on natural scenes, such as: vertices, vectors, lips, eyes, mouth etc.... These image analysis tools are giving Neuro localization output information. They simulate the low level attractiveness of the Neuro-Visual system in the brain: Neurons can be considered as hardware filters, as they perform sum of products. Moreover, such biological models can perform the perceptive grouping process, which merge similar textured regions, more efficiently that classic segmentation algorithms. These information can give key elements for: a) Finding corresponding points on epipolar lines: and achieve better Z depth estimation. b) Helping a meshing tool for generating an optimized mesh. c) Help the rendering algorithm to remove unnecessary details of the scene. 2 – MESH SIMPLIFICATION: Biological models can be used also for reducing mesh complexity, according to the degree of importance of each picture area. 3 – ENHANCED RENDERING TECHNIQUES: High quality rendering techniques have now important limitations in terms of displayed polygons per frame. Low cost hardware cards achieve now 1 to 2 millions polygons per second. But large, complex databases, such as architecture, environment need more polygons. But it is quite unnecessary to display such polygon number when the viewer is watching only a limited part of his environment. Moreover, we must adapt the polygon number according to the solid angle of each observed object. But how can we select important and non important area? Visual models can bring a response to this problem. 4 – MERGING IMAGE ANALYSIS AND SYNTHESIS: A new direct bridge could be created between image analysis and synthesis, (as in MPEG4-SNHC goals), by the use of filter-based data structures instead of splines/polygons. Such data base could facilitate the gap between 3D objects and image processing elements. (The terms used in MPEG4/SNHC is VOP for video object planes: It’s the first time that an image element other than contour, edge, etc., is considered as an object. Possibility of automatic symbolization, as a tool for bridging the gap between 3D analysis and synthesis thanks to the Cortical Transform. Significant results have also been achieved in very low bit rate Coding (at 30 Kbits/s and less), with more acceptable aspects which fits with visual System requirements. Moreover, neural tools seem to be generic and well adapted to the various input styles of TV sequences, thanks to their learning capabilities & architecture dedicated to all kinds of 2D & 3D Analysis-Synthesis processing.

ABSTRACT: Laser scanning enables to measure ahuge number of 3D points in a few time. However, the creation of a 3D model from these data remains an involving work. We introduce a method to get this modeling task easier. Our approach is twofold. On the one hand, a camera is coupled to the laser scanner in order to drive the scan by image analysis. On the other hand, architectural knowledge is digitized in a library of parametric components.

Le releve laser donne la possibilite de scanner un grand nombre de points en peu de temps. Cependant, la creation d'un modele 3D a partir de ces donnees reste un travail important. Nous presentons une methode pour faciliter cette tâche de modelisation. Notre approche est basee sur deux etapes. La premiere consiste a coupler le scanner laser a une camera afin de guider le releve par l'image. La deuxieme etape consiste a modeliser les connaissances architecturales via une bibliotheque d'objets parametres.