and its classification into the initial symbolic meaning 
prevails. The figure movement parameters over time 
are considered characteristic for a conveyed message 
by gestures as part of nonverbal communication. 
The idea is also to investigate how far methods of 
image generation and visualisation (computer graphics 
Ima can positively influence the results that can be 
acchieved in the recognition field (computer vision 
area). In order to point out these separate paths, this 
paper splits into two major parts, a synthesis section 
and an analysis section. In the first one an animation 
tool for creating synthetic image sequences for the 
application domain is described. It uses three different 
model classes for the calculation of movement 
trajectories. The second part deals with both state 
estimation, which is explained in detail, and image 
analysis (as a prerequisite for obtaining measurement 
values). 
1.3. System organization 
The work that has been accomplished is reflected by 
four coordinated partial process modules all being 
implemented on a SGI graphics worksiation as a 
simulation facility. Figure 2 gives an overview over the 
main signal flow. Control values and sychronizing data 
are exchanged by the UNIX message passing 
mechanism. 
real world 
  
   
  
  
  
   
   
  
  
   
   
    
  
  
  
   
  
  
  
  
  
   
   
  
   
   
  
The first two processes serve as sources for pixel data: 
An image sequence recall process is able to display 
various stored CCD-camera images onto a screen for 
purposes of testing image processing algorithms, 
checking their robustness, and for preparing 
applications outside the simulation environment. 
Secondly, an animation process (described in the 
synsthesis section) synthetically generates image 
sequences with figures and background objects under 
user control. It meets the requirements for handy test 
signals, for having a performance measure for the 
estimation procedure, and for obtaining multiple 
training material for future classificators. Figure 3 
gives an impression of used input images. 
The next two modules belong to the analysis part of the 
recognition task and are described in detail later in this 
aper: 
Tho image processing module is the mediator between 
pixel source and state estimation and picks up screen 
pixel data and searches for edges as basic image 
features. It sends information marking search areas to 
the imaging process and measurement values to the 
state estimation. This module is the basis for early 
recognition, proper initialization of state estimation 
ang it supplies the necessary features for estimator 
update. 
The estimation process contains the generic models, 
synchronizes and controls all other processes and may 
receive reference values from the animation side. 
  
  
  
movement 
  
  
  
  
  
model 
  
  
  
  
  
  
  
  
selection correction 
  
  
  
dynamic 
model 
comparison 
  
  
process prediction 
model 
  
  
  
geometric 
object model 
  
igeom. view- : 
:ino model 
  
  
  
  
  
  
Fig. 1: Signal relation and basic idea 
  
  
signal flow 
A » control flow