possibility of extensions. This is obtained by a modular 
design and the use of standards for coding. These 
standards are currently missing for image visualization 
and user interface, although the common orientation 
towards X-windows and Motif enviroments. 
The use of a client/server architecture is another way to 
allow portability, this allows the use of the digital 
photogrammetric system even with huge 
photogrammetric data, relying on a central server. 
The program is written in the ANSI C language to take 
advantage either of the flexibility of the C programming 
language and of the ANSI standardization. 
The graphical interface, that realizes both image 
visualization and user interaction, uses the X- 
windows/Motif system, that is generally available on Unix 
system. 
The software is divided into different modules for: 
e image acquisition and preprocessing; 
e user interaction (both input and output); 
e numerical computation. 
Each module has been tested as a separate program to 
evaluate both correctness and efficency. 
3. OPERATIONS WITH THE SYSTEM 
The operations with the system are divided into 
orientation and plotting. 
The orientation is performed into three steps: 
e the operator searchs for approximate homologous 
points in all the loaded images and for control points; 
e the system refines the seach of the homologous 
points by a least square image matching; 
e the global orientation is performed by bundles 
adjustment. 
The procedure determines orientation parameters for 
each image and ground coordinates of the points with 
their standard deviations. It is possible to fix the 
reference system either using known ground points 
coordinates or setting proper constrains on the image 
orientation parameters. In this way it is possible to set an 
arbitrary reference system. 
The actual plotting of points in the object space (i.e. the 
determination of object point coordinates) is carried out 
using a procedure that determines parameters in the 
object space by means of suitable observation equations 
described in the following paragraph. The operator 
chooses a point in the first image moving a cursor on the 
screen by the mouse and gives an approximate value for 
the elevation Z of the point. The system automatically 
gives the planimetric object space coordinates of the 
point and determines its height. 
If the system is computing the coordinates of a linear set 
of points the operator follows the line on the first image: 
the system computes in real time the 3D ground point 
coordinates and records them with a fixed spatial interval. 
The ground points coordinate determination is 
monoscopic in the choice of the points but tridimentional 
in the results: the operator selects points on one image 
and the system "sees" the other images to give the 
coordinates in the object space reference system. 
Obviously image orientation must be known at this stage. 
420 
All the system functions are controlled by the operator in 
two ways: 
e with the preparation of input files; 
e With the interaction during the system operations. 
Input files are: 
e a session description file, which contains all the 
default values for the parameters and the choices for 
the different options; 
e a file containing approximate or actual (if available) 
orientation parameters for all the images; 
e the image files. 
A set of initial parameters is set up in the session 
description file, these include: 
e input and output file names; 
e least square matching parameters: dimensions of the 
local matching window, approximate values and 
constrains for the geometric and  radiometric 
trasformation parameters; 
e ground points coordinates determination parameters, 
such as approximate slope and object space 
discretization steps; 
e image orientation parameters: minimum and 
maximum number of iterations and convergence test 
settings. 
Each system operation records its results in a specific file 
that can be used as input for the following steps. 
During system operations all user interactions are driven 
by an user-frendly interface; it's possible to change all the 
parameters’ values at run time. 
4. MATHEMATICAL MODELS AND COMPUTATIONAL 
PROCEDURES 
4.1 General informations 
The matamatical models used for the implementation of 
the system are partly very classical and well known and 
partly more innovative. 
The least square image matching in the image space is 
now a standard procedure, and it will not be described 
here. It is used by the system to refine the coordinates of 
the homologous points before computing the orientation. 
Bilinear finite elements are used to interpolate the image 
density (or gray values). The image gradient is evaluated 
by means of finite differences and interpolated in the 
same manner. This procedure is not based on an overall 
consistent model, but the results are acceptable for 
practical purpose and the procedure is very simple. 
The equations for the final plotting are based on a direct 
definition of unknown parameters in the object space and 
will be described in the next point. 
4.2 The equations of local image matching in the 
object space approach. 
The object surface is locally approximated by a plane, 
therefore the parametric equation of the surface itself in 
the neighborhood of a point P is: 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996 
All 
(3)