Automatic 3D modeling of the urban landscape

In this paper we present a fully automatic system for building 3D models of urban areas at the street level. We propose a novel approach for the accurate estimation of the scale consistent camera pose given two previous images. We employ a new method for global optimization and use a novel sampling technique for primitive fitting to obtain an efficient representation of the scene. We combine these techniques to obtain a 3D representation based on textured planar patches using a single handheld digital reflex camera (DSLR).