| = Signal manager 
wr aH 
m Signal processor E [ 
  
  
  
  
  
  
Coherent LL 
  
  
É 
  
RS - 232 
  
  
    
IRAGA-780 | 
  
Fig.3. The functional scheme of experimental 
complex 
We choose the experimental position on the 
height of 30m at the distance to the objects of 
150«2000m. As the example we choose the track 
ZIL-131. Some experimental results (TV, LL and IR 
images respectively) are shown on fig.4-5. 
  
Fig.4 Truck at the distance about 500m 
  
  
  
Fig.5 Truck at the distance about 1500m 
4. MAIN RESULTS FOR FRAME BASED 
ANALYSIS OF MULTISENSOR IMAGE 
aha áÁ— 
SEQUENCES 
4.1. Base algorithms for multisensor image 
processing 
In order to optimise the multisensor image 
sequences analysis on the level of processing we 
suggest several approaches to use the multisensor 
information to increase the reliability of target 
recognition in automatic manner: 
e fusion on pixels level, 
e fusion on the level of feature vectors; 
388 
e fusion on the level of soft-vectors. 
For fusion on pixels level we have used 
Pytiev correlation theory and desiered the special 
algorythms [8]. Also, our software includes about 25 
different features’ extractor  algorythms and 
designed the special approach to select the 
invariant feautures to the concrete problem. Due to 
the soft-vector level, there are two approaches to 
evidence fusion from different sensors. First of 
them is based on classical Bayessian theory, and 
the second one - modified Dempster-Schaffer 
approach, which was successfully used in our 
software[4-7]. 
4.2. Software realization 
  
The first subject of discussion on the 
problem of software realization is the determination 
of principle hardware platform. Any designer must 
find the compromise between requirement 
characteristics of the computer system and 
economical indexes. 
The designed software is based on the 
conception of Frame Object-Oriented Programming 
[2]. This technique is based on the new type of 
programming objects - "software frames" (the term 
frame had taken from Artificial Intelligence). From 
the program point of view they are the generalizing 
of the well-known term "class" from the Object 
Oriented Programming (OOP). 
Frames are encapsulate links - special data 
structures, which determines the behavior of the 
frame. There are links of two types: "synchronies' 
and "information". Any frame action could be carry 
out as the reaction on the message from the other 
frame, which is connected with it by corresponding 
"input" synchronic link. Moreover, any external data 
are available for reading or writing only by "input" or 
“output” information links and only when the 
"frames-masters" are “agree” with it. Frames 
couldn't process independently or "see" any external 
data. Frames work only together. 
In terms of OOP we could say, that any 
frame is the store place for the object, which 
contains the other objects, which are data, 
procedures, links, etc. So, frame, even if it hasnt 
definite members, has the possibility of access to 
them. Any copy of existing objects have to be 
placed in special "box" - slot - of frame structure. 
Our approach suppose to present any 
processing scheme as a Frame Processing Set 
(FPS). FPS is an asynchronic set of soft-frames, 
which are connected by links (fig.6). Any FPS must 
include Data Frames, Processing Frames and 
Control Frames. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996 
  
  
In 
image Se 
about 30 
  
Fig.7. Ba 
  
  
  
  
Fi 
RESULT: 
This pape 
in the fi 
We've d 
Processit 
implemer 
Support 
processin 
special 
behavior