318 
DEVELOPMENT OF A TEMPLATE FOR AUTOMATIC STEREO MATCHING 
Amir Saeed Homainejad and Mark R. Shortis 
Department of Geomatics 
The University of Melbourne 
Parkville 3052 AUSTRALIA 
Telephone : +61 3 344 6806 
Facsimile : +61 3 347 2916 
Email : homain@sunburn.sli.unimelb.edu.au 
KEY WORDS: Template, Automatic Stereo Matching, Digital Photogrammetry, Close range 
ABSTRACT 
A PhD project has been undertaken at the Department of Geomatics of the University of Melbourne, entitled Real Time 
Photogrammetric Processing. One of the goals of the project is to achieve stereo matching in real time, or on-line photogrammetry. 
Therefore a new template technique has been developed for the purpose of facilitating automatic stereo matching in digital close 
range photogrammetric processing. The structure of templates comprises two parts. The first part is allocated as a control point, and 
the second part is a pattern referring to the template number. Some tests were done in digital close range photogrammetry. Test fields 
including some templates were captured by an off-the-shelf CCD camera at close range. Additionally, an initial test was done within 
aerial photogrammetry by superimposing the templates on a digital aerial photograph. An algorithm which has been developed for 
this project can successfully recognise and detect the control points and the numbers referring to them. The main aim of developing 
the template is use in digital close range photogrammetry for automatic stereo matching, but the biggest advantage of the templates 
is flexibility for use with any type of cluttered scene. 
1. INTRODUCTION 
Since the advent of the era of digital photogrammetry in the early 1980's, photogrammetrists have developed the utilisation of a 
variety of hardware devices and algorithmic approaches which have enhanced digital image processing. These include video 
cameras, digital scanners, frame grabbers, and digital image processing methods. Digital photogrammetry has many advantages in 
contrast with conventional photogrammetry which has led to much research in the field. This paper is not concerned with the 
advantages of digital photogrammetry because these have been covered extensively in the literature, for example, Gruen (1988, 
1992, 1994), Helava (1988) and Homainejad (1992). 
Digital photogrammetry facilitates the method of automatic stereo matching. The main function of stereo matching is the formation 
of a three-dimensional model of an object at a known scale, which can then be measured in order to obtain numerical information in 
the form of XYZ coordinates, or for the compilation of a topographic maps. For the reconstruction of a stereo model from an object, 
it is necessary to have a good distribution of control points on the object and background. In the analog stereo plotter the operator 
manually introduces three rotation angles ( $, Q, K ) and three translations (x, y, z), for at least three points on each photograph 
resulting in a three dimensional model. In the analytical method, the coordinates of at least three points on each photo are measured 
in the photo coordinate system. Then using a mathematical computation method, like bundle adjustment, a mathematical stereo 
model of the object is constructed. Even with many soft copy photogrammetric workstations, it is still necessary that control points 
are measured manually for the orientation process and the creation of the stereo model. 
In digital aerial photogrammetry and digital close range photogrammetry a lot of research and tasks have been undertaken to achieve 
a method for automatic and semiautomatic stereo matching. These methods are categorised into the two major groups of feature 
based matching and area based matching. In feature based matching methods, the most interest is in the extraction of a distinct points 
from two overlapping digital images. For example, Forstner and Gulch (1987) explained a method of distinct points detection and 
extraction based on the Forstner (1986) operator. The distinct points were defined as corners, and centres of circular features such as 
wells and holes. Another method which can be named in this group is the Moravec (1977) operator or Hough transformation which 
was used for lines and curve detection for stereo matching (cf. Nasrabadi 1992). The second method deals with the pixel intensity 
values in the two images. For example, Vollmerhaus and Bildanalyse (1987) explained a method of computation for area based 
matching of pixel values in the images. A good summary of various matching techniques is given in Lemmens (1988). 
One of the goals of this project is to achieve automatic stereo matching in close range photogrammetry. Therefore a new type of 
template has been developed for the automatic recognition and detection of control points in digital images. The aim of developing 
of the template was based on simplifying the process of digital stereo matching, detecting and extracting control points with their 
numbers, and defining the relationship of images and objects. The focus of this paper is on the development of the template. 
There have been many schemes proposed for the automated detection and identification of targets in photogrammetric images. Most 
of these have concentrated on the unambiguous separation of target images from a cluttered background using a combination of 
patterned targets and filtering of the image (van den Heuvel et al 1992, Wiley and Wong 1990). The advantage of the technique 
proposed here is that little or no pre-processing is required, in order to minimise the amount of computation. If the technique is to be 
applied to real time photogrammetry then extremely efficient processing is mandatory. 
In the next section the configuration of the templates will be given. In the third section the results of practical testing are discussed, 
and at the end some conclusions will be given. 
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995