FIRST EXPERIMENTS WITH CONVERGENT MULTI-IMAGES PHOTOGRAMMETRY WITH 
AUTOMATIC CORRELATION APPLIED TO DIFFERENTIAL RECTIFICATION OF ARCHITECTURAL 
FACADES 
J. García-León 3 *, A. M. Felicísimo 3 , J. J. Martínez b 
a Ingeniería Cartográfica, Geodesia y Fotogrametría, Escuela Politécnica, Universidad de Extremadura, 10071 Cáceres, 
Spain - (jleon, amfeli)@unex.es 
b Expresión Gráfica Arquitectónica, Universidad Politécnica de Cartagena, 30202 Cartagena, Spain - 
juanjo.martinez@upct.es 
Commission VI, WG VI/4 
KEY WORDS: convergent multi-images, automatic correlation, DSM. 
ABSTRACT: 
In architectural photogrammetry, convergent geometry has some advantages over normal geometry. One of them is the reduction of 
the number of images needed for the full coverage of the object. This advantage simplifies network design and increases 
theoretically the accuracy of the method and internal reliability (Mason, 1994). However, the method has still not been widely used 
in practice because, probably the workflow has not been as consolidated as in the case of normal geometry. 
In this article, we present and discuss the workflow to make a Digital Surface Model (DSM) of architectural façades semi- 
automatically and using convergent geometry. The accuracy of DSM is verified using and independent group of checkpoints, 
measured using conventional topographical methods. The influential parameters in the quality of DSM are discussed. 
Finally, the influential factors are discussed and lines of development are proposed in order to overcome the present difficulties. 
The results and discussion are based on a practical example using an historical monument in the medieval city of Câceres (Spain), 
declared a World Heritage City in 1986. 
1. INTRODUCTION 
1.1 Approach to the problem 
Architectural photogrammetry continues to be made with the 
strict limitations of aerial photogrammetry, in which the 
geometry is normal, the base is constant and the scale is 
uniform. These restrictions limit the flexibility now of 
recording data, increasing the number of images and requiring 
true acrobatics in order to resolve hidden areas. (See, for 
instance, Desmond & Bryan, 2001). 
Changing the geometry and data processing strategies would 
solve part of the previously mentioned problems. This work 
presents a group of experiments in which the options chosen 
have been the following: 
• The exposure is not normal but oblique and its geometry 
convergent. 
• The overlapping is large, practically 100%. 
• The images are multiple, up to 11 images. 
There are three main reasons for this design: 
• the oblique image solves the problems of space and 
camera movement. 
• The number of images necessary for the complete 
coverage is less than in the case of normal geometry. 
• The multiple images assure that there will be a repetition, 
which may be taken advantage of in order to obtain 
results that are more precise. 
Data processing is the next problem since the photogrammetric 
applications are designed principally for the treatment of aerial 
case, with normal geometry (Patias & Tsioukas, 1999). In this 
case an application is needed which will allow work to be done 
with a multi-images convergent network and photogrammetry 
in addition to implementing algorithms of stereo-matching for 
automatic Digital Surface Models (DSM) generation. The only 
commercial application that contains these characteristics is the 
extension Orthobase Pro of Erdas Imagine extension. Orthobase 
Pro allows for the adjustment of multiple networks with oblique 
images specifically designed for terrestrial photogrammetry. 
There are object programs that allow adjustments to be made 
with these types of networks but they are not able to execute a 
completely automatic DSM and you can work only with an 
outside of an ensemble of points marked on the object (see, for 
instance, Otepka, et al, 2002) 
1.2 Goals 
Once the net and the computer applications were defined, the 
work focused on a principal objective: achievement of a DSM 
with a maximum density of points and the best precision 
possible. In order to carry this out, experiments were performed 
in which various value combinations were tested for a group of 
Corresponding author.