Geometry and Texture Recovery of Scenes of Large Scale

Abstract

This paper presents a systematic approach to the problem of photorealistic 3-D model acquisition from the combination of range and image sensing. The input is a sequence of unregistered range scans of the scene and a sequence of unregistered 2-D photographs of the same scene. The output is a true texture-mapped geometric model of the scene. We believe that the developed modules are of vital importance for a flexible photorealistic 3-D model acquisition system. Segmentation algorithms simplify the dense datasets and provide stable features of interest which can be used for registration purposes. Solid modeling provides geometrically correct 3-D models. Finally, the automated range to an image registration algorithm can increase the flexibility of the system by decoupling the slow geometry recovery process from the image acquisition process; the camera does not have to be precalibrated and rigidly attached to the range sensor. The system is comprehensive in that it addresses all phases of the modeling problem with a particular emphasis on automating the entire process interaction.