extraction, good enough for certain applications 
[7,8,11]. The required computations include two 
stages. 
- External orientation of a single photograph by 
space resection: the (6 orientation parameters 
are evaluated using well-distributed ground 
control points - Image-to-ground transformation: 
the intersection of the ray, from the image 
point and the camera station, with the DTM 
surface is iteratively evaluated. 
Digital monoplotting requires operators who can 
interpret and simply digitize photographs, similar 
to the method used for manual digitization of 
existing maps. There are three viewing 
possibilities: with the naked eye (using either 
the original or enlarged photographs), with a 
magnifying glass or a stereoscope. 
For change detection, the existing digital data 
which are to be revised are converted to photo 
coordinates (by applying the inverse 
transformation). The transformed digital data are 
then displayed on the graphics screen and visually 
compared with up-to-date photographs; the changes 
are indicated during digitizing. It is also 
possible to plot the transformed digital data on a 
transparent sheet, and then superimpose it on the 
photograph for manual change detection. 
PERFORMANCE EVALUATION 
To evaluate photogrammetric systems, a number of 
items can be considered, such as versatility, 
flexibility, cost, performance, reliability, human 
factors, support requirements, etc [9]: In our 
case, digital monoplotting and stereoplotting 
systems vere experimentally evaluated with respect 
to their accuracy, interpretability and 
time-efficiency. The tests were carried out by 
selecting an area of interest in which there was a 
variety of features portrayed. The data were 
collected and processed by the photogrammetric 
systems and the results were evaluated against a 
source of higher quality information, referred to 
in the sequence as "reference data". 
Geoinformation is related to both time and 
position. Data are collected during a certain 
period of time, and the observed phenomena change 
with time. Aerial photography provides a snapshot 
of the status of the phenomena. Positions can be 
measured after determining the geometry of the 
photographs with respect to the ground. 
Both the boundaries of some features appearing on 
the photographs and/or their attribute information 
can be difficult to extract. Feature extraction 
involves two types of interpretation: delineation 
of the feature and determination of its associated 
attributes, which implies subjectivity. Apart from 
the nature of geoinformation, the equipment, 
methods, scale and quality of the photographs are 
also major factors that influence the accuracy of 
spatial data extracted from aerial photographs. 
Accuracy is defined as the closeness of results of 
computations or estimations to the true values, or 
values accepted as true, and is classified into 
attribute accuracy and positional accuracy [5]. 
Method of determining attribute accuracy 
Attributes are defined here only in relation to 
object type and dimension, such as main road, 
track and path, river, vineyard, etc. Attribute 
accuracy is experimentally quantified by the rate 
of success of feature classification. After being 
digitized, an existing topographic map is used for 
494 
true attribute values, and the rate of success per 
system . is determined by comparing features 
extracted from the photographs by digital 
monoplotting or stereoplotting with those of the 
map. 
Objects such as towers, windmills, etc, appearing 
on the map are difficult or sometimes impossible 
to interpret in medium-scale photographs. Point 
features were therefore omitted in the evaluation. 
The correctness of classification of objects was 
evaluated by comparing the number of objects on 
the map with those extracted from the photographs. 
For this purpose, vector-based GIS software (PC 
Arc/Info) was used, calculating the total length 
of lines per object class and the total area of 
polygons per object class. 
The rate of success, expressed in percentage, was 
computed by dividing the total number of objects 
per class extracted from the photographs by the 
number of objects digitized from the map. 
Positional accuracy evaluation 
To quantify the positional accuracy of digitized 
features, two coverages were overlaid. One was the 
expected higher quality data, in our case the 
existing digital map, and the other was the result 
from either digital monoplotting or stereoplotting 
of the same area and features. In the evaluation, 
linear features were considered, such as man-made 
(well-defined) features (e.g, roads), natural 
features (e.g, rivers), and polygon boundaries 
(e.g, land use boundaries). 
Determination of positional accuracy 
The two coverages, which contain the same linear 
features, were overlaid. If there are no gross 
errors, the linear features should more or less 
coincide. Small deviations and sliver or spurious 
polygons may occur because of different sources 
and methods of digitization, and random digitizing 
errors. 
One way to evaluate accuracy is using the epsilon 
band concept. The epsilon band is intended to 
describe a mean probable location for a line; it 
is an area defined by two parallels to the most 
probable location of the line. The true position 
of the line will occur at some displacement from 
the measured position. Geometrically, the line 
dilates to a sausage-shaped zone, contouring a 
probability density function of the line’s true 
location [4,3]. The width, epsilon, of the band is 
a measure of the uncertainty of the line's 
location; half of this vidth is called the epsilon 
distance. Implementation of this concept requires 
GIS vector-based software with spatial analysis 
capabilities. An epsilon band is formed around the 
reference line and its width is changed until the 
superimposed lines are enveloped by the band or 
only a specified percentage of the points remain 
outside. An accuracy measure is thus obtained. 
Another way to evaluate positional accuracy, and 
which was used in this work in conjunction with 
the epsilon band concept, is the following. The 
line coverage from the test data was re-formatted 
to point coverage (by programming outside the PC 
Arc/Info environment). The line coverage from the 
base‘ data was superimposed on the point coverage. 
The distance from each point to the base line was 
measured. Accuracy was expressed as a standard 
deviation, and an epsilon distance was also 
calculated.