nbul 2004 
onnects at 
|, Grün & 
category, 
be more 
ing linear 
> tracking 
nan & de 
onsists of 
' the line 
algorithm 
, and the 
oint must 
ar feature 
ssary user 
the linear 
or control 
rrect the 
>r that, the 
following 
etermined 
file serves 
Then the 
feature Is 
n and the 
he current 
sed on a 
file that is 
predicted 
r the new 
opped e.g. 
It stops in 
n. 
ched, it is 
proposed. 
t can be 
acking has 
t be found 
tion result 
ed points. 
ry, a new 
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
    
  
reference profile can be specified and the tracking of the line can 
be continued. Where the linear feature has been extracted 
completely, it is closed, and the next line segment can be 
captured. Figure 6 shows a typical situation during the 
acquisition of a road network. 
  
Interactive 
initialization 
i 
Determination of 
the reference profile 
| 
No Prediction of 
the next point 
i 
Measurement of 
the next point 
| 
Fusion of prediction 
and measurement 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
    
  
   
Stop 
tracking? 
  
    
  
Task dependent 
interaction 
  
  
  
Figure 5. Flow chart of the line following algorithm 
  
Figure 6. Line extraction of a road intersection. The different 
road segments are topologically connected by a node. 
The width of the line is automatically determined based on the 
assumption that the linear feature is bordered by more or less 
parallel grey value edges. 
Before the line is stored in the database, it is smoothed and the 
number of vertexes is reduced using the algorithm presented in 
Ramer (1972). 
S. RESULTS 
The semi-automatic feature extraction software was designed to 
operate as a vector data acquisition and update component in a 
distributed GIS environment. We decided to use the OpenGIS 
standard GML for vector data import and export, because it 
enables the integration into an open GIS infrastructure. GML 
can be applied for interoperability between different 
organisations and companies, which has already been 
cessfully tested (Feldman & Curtis 2003). 
The feature extraction is preferably done in digital orthophotos 
for the capture of 2D GIS vector data. In addition, the software 
is available for the capture of 3D features using oriented aerial 
imagery. Here the automation part consists currently of the on- 
line z measurement functionality which automatically derives 
the height of each vertex point of a line feature or the contour of 
an area feature. The algorithms have been extensively tested 
with IKONOS 2 and IRS satellite imagery as well as with 
orthophotos. Figure 7 shows the result of extracted VMapl 
features from an orthophoto with 50 cm pixel size. 
  
Figure 7. Extracted line and area features from an orthophoto 
6. CONCLUSIONS 
inJECT is one of the very few software developments in 
automated feature extraction that have been implemented as a 
commercial system. The software has been substantially 
extended with new automation modules and a strong GIS 
interface. The software basis of inJECT has been proven to be 
an excellent platform to add and test external software modules 
that increase the automation level significantly. By adapting the 
OGC-defined GML standards the field for future applications is 
wide spread. The developed GIS interface to Dynamo and 
GeoMedia opens the window to many applications in those 
fields. Our solution has major advantages compared to many 
GIS data acquisition packages that do not allow such user- 
friendly data capture and updating, especially not in 3D. It has