GEOMETRIC AND RADIOMETRIC MODELING OF 3D SCENES Marco Marcon, Augusto Sarti, Stefano Tubaro Dipartimento di Elettronica e Informazione - Politecnico di Milano Piazza Leonardo Da Vinci, 32, 20133 Milano (ITALY) ABSTRACT Modeling of 3D scenes is a hot topic in Computer Vision from more that thirty years, and probably its history is longer than a century considering also photogrammetry . In the recent years the rapid technological improvements that character- ized the acquisition devices (photo-cameras, video-cameras, ..), illumination devices (lasers, structured light sources) and computational units allowed the application of 3D shape es- timation methods, based on image analysis techniques, in a wide set of applications. Furthermore real-time 3D analysis is becoming a common tool in Virtual and Augmented Reality contexts. Aim of this presentation is a rapid description of re- cent major advances on geometric and radiometric modeling of 3D scenes based on image analysis. Index Terms— Machine Vision, Geometric modeling, 3D scene reconstruction, Photogrammetry. 1. INTRODUCTION In general 3D modeling problem formulation is typically con- nected to a specific application (e.g. biomedical 3D imaging or industrial 3D volume inspection,...) and starts from multi- ple images captured by analog or digital cameras. It is impor- tant to mention that in this type of problems an accurate geo- metrical description of the framed scene is just one aspect of the modelization procedure. For example considering seam- less fusion of real and virtual contents (like those addressed for TV/Film production or for realistic pre-visualization of products, objects, architectural environments as an overlay of a real scene) an accurate description of lighting condition and surface reflection properties of each object present in the scene should be considered. Aspects concerning this topic will be discussed in section 4. In general when a specific 3D modeling application is addressed the preliminary phases that should also be considered can be described as follows: • definition of the requirements and specifications of the application under investigation; • sketch of possible solutions or approaches; • design of an image acquisition model, involving pose planning, sensor design, illumination conditioning etc.; • test of the defined image acquisition model. In the following sections we briefly describes recent results on image analysis for 3D scene modeling. 2. ACQUISITION SET-UP We concentrate our analysis only on passive imaging systems, i.e. systems where no specific lighting devices (like projectors or lasers) are used in the acquisition process. Moreover,in the recent years, some innovative proposals to re-thinking some components of the traditional image acquisition model (e.g. a pinhole projection model with a pre-defined camera mo- tion) leaded to some interesting results connected to virtual view generation, this approach is normally indicated as ”Im- age Based Rendering” [1][2][3][4]. According to [5] we will divide possible acquisition approaches following this scheme: • Visual field: Circular/Non-circular • Focal point(s) associated with each image: Single/Multiple • Acquiring time(s): Single/Multiple • Acquiring pose(s): Single/Multiple Most of the classical scene acquisition approaches can be casted into this classification: e.g. a planar image and a cylindrical image are examples of the non-circular and the circular visual field classes. A pinhole projection model is the example for the class of a planar image associated with a single focal point. For the class of a planar (non- circular) image associated with multiple focal points many examples are available from multi camera set-ups (bi-nocular or multi-ocular) to images where each pixel is associated with a different focal point, (i.e. orthographic projections from pushbroom cameras [6]). Also image based rendering approaches like Light Field [2] and Lumigraph [1] can be classified in this class where the poses of pinhole cameras are arranged to a planar grid layout and assume synchronous acquisition for their applications. In these approaches the generation of novel views is obtained by a re-sampling of the acquired image data which are parameterized as a 4D function. Approaches based on multiple poses of a pin-hole camera represent the most common case in Computer Vision 1604 978-1-4244-4291-1/09/$25.00 ©2009 IEEE ICME 2009