Estimating 3D Polyhedral Building Models by Registering Aerial Images - Advanced Concepts for Intelligent Vision Systems - page 238

Estimating 3D Polyhedral Building Models by
Registering Aerial Images

Fadi Dornaika 1 , 2 and Karim Hammoudi 3

1 University of the Basque Country, San Sebastian, Spain
2 IKERBASQUE, Basque Foundation for Science, Bilbao, Spain
3 Universite Paris-Est, Institut Geographique National, Paris, France

Abstract. We describe a model driven approach for extracting simple
3D polyhedral building models from aerial images. The novelty of the
approach lies in the use of featureless and direct optimization based on
image rawbrightness. The 3D polyhedral model is estimated by opti-
mizing a criterion that combines a global dissimilarity measure and a
gradient score over several aerial images. The proposed approach gives
more accurate 3D reconstruction than feature-based approaches since it
does not involve intermediate noisy data (e.g., the 3D points of a noisy
Digital Elevation Model). We provide experiments and evaluations of
performance. Experimental results show the feasibility and robustness of
the proposed approach.

1 Introduction and Motivation

The extraction of 3D models of buildings from aerial images is currently a very
active research area since it is a key issue in urban planning, virtual reality, and
updating databases for geo-information systems, to name a few [1]. The proposed
methods for building reconstruction differ by the assumption made as well as by
the type of input data. However, one can easily classify these approaches into
two main categories: bottom-up and top-down approaches. In theory, bottom-
up approaches can handle the case where there is no prior knowledge about the
sought building model. However, in the presence of noisy or low resolution data
there is no guarantee that the estimated models will be correct. On the other
hand, top-down approaches rely on some prior knowledge (e.g., using parametric
models). The top-down approaches use the principle of hypothesis-verification
in order to find the best model fit. Both categories have been used with features
that are extracted and matched in at least two images. For roofs, the most used
image features are 2D segments and junctions lines that are converted into 3D
features. The final polyhedral model is then estimated from these 3D features.
Model-based reconstruction techniques were first applied in digital photogram-
metry for the (semi-)automatic reconstruction of buildings in aerial images with
the help of generic building models [2-4]. In this paper, we propose a featureless
approach that extracts simple polyhedral building models from the rawbright-
ness of calibrated aerial images where the footprint of the building in one image