na 1996 
RELATIONAL MATCHING APPLIED TO AUTOMATIC EXTRACTION 
OF GROUND CONTROL IN DIGITAL IMAGES 
Aluir Porfirio Dal Poz 
Assistant Researcher 
Antonio Maria Garcia Tommaselli 
Assistant Professor 
Säo Paulo State University - UNESP 
Campus of Presidente Prudente - Brazil 
E-Mail: ueppr@eu.ansp.br 
Jorge Pimentel Cintra 
Assistant Professor 
University of Säo Paulo - USP 
Polytechnical School of USP - Brazil 
Commission Ill, Working Group 3 
KEY WORDS: PHOTOGRAMMETRY, VISION, AUTOMATION, MATCHING, MODELING, PATTERN 
ABSTRACT 
The aim of this paper is to test the use of relational matching in the automatic extraction of ground control (straight features). 
The identification of ground control on photographs or images is usually carried out by a human operator, who uses his natural 
skills to make interpretations. In Digital Photogrammetry, which uses techniques of digital image processing, the extraction of 
ground control can be automated using relational matching. This matching approach is commonly used in Computer Vision, 
but only recently has been applied by photogrammetrists. It has been recognized its great potential to automate several tasks 
in photogrammetry. The basic principles of the approach and an experiment based on simulated data are presented and 
discussed in this paper. 
1. INTRODUCTION 
Photogrammetric tasks, such as relative orientation of 
images, derivation of digital terrain models (DTM) and 
aerial triangulation can be automated by using the image 
matching technique. The matching method commonly 
used in these applications is the area based matching. 
The basic principle of this technique is the establishment 
of correspondences between patches of the overlapping 
images. 
However, there are other tasks, also based on the 
correspondence principle, where the automation is very 
difficult to be implemented. One of these tasks is the 
automatic extraction of ground control. In such a case, 
the correspondence is performed between an image and 
a symbolic model describing the ground control. Several 
approaches have been proposed recently (Haala and 
Vosselman, 1992; Schickler, 1992; Hellwich and Faig, 
1994). An approach to solve the problem of automatic 
extraction of ground control based on relational matching 
and a heuristic that uses the analytical relation between 
straight features of object space and its homologous in 
the image space is presented in this paper. 
In the next section the basic principles of the approach 
are presented. The section 3 discusses an experiment 
based on simulated data. 
2. THE MATCHING PROCEDURES 
Relational matching is based on the correspondence 
131 
between two relational structures. A relational structure is 
composed of primitives (in our case, straight features) 
and relations among the primitives. 
Three steps can be identified in all relational matching 
approaches: 
. transformation from raster space to entity 
space; 
. transformation from entity space to relational 
space; and 
. matching strategy. 
2.1 Transformation from Raster Space to Entity 
Space 
This transformation is performed by edge detection and 
vectorization. Gradient methods can be used to detect 
edges and Hough method to detect and vectorize straight 
features. Ground control (straight features) is supposed 
to be available in the entity space, i. e. , defined by two 
3D points in the ends located at the straight feature or by 
one 3D point and one 3D normalised vector. 
2.2 Transformation from Entity Space to Relational 
Space 
Transformation from entity space to relational space is 
accomplished by using relational models, commonly 
referred to as relational descriptions in the Computer 
Vision literature. These structures are lists of relations. 
Let Oa be an object and A be the set of its parts. An N- 
ary relation over A is a subset of the Cartesian product 
A^- Ax...xXA (N times) (Shapiro and Haralick, 1987). In 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996 
a a cio ck HB