Video 
Figure 14 Hardware architecture of the image processing system 
VII. CONCLUSIONS 
A model based dynamic vision system has been pre- 
sented for recognition of partially occluded three-di- 
mensional rigid objects in an application for autono- 
mous road vehicle guidance on German standard 
"Autobahnen". Only image sequences are used as input 
data for the vision system. On the first processing level 
a hypothesis is generated using knowledge about the 
possible location of the appearance of an occluded 
object with an expected shape by matching measured 
features with the internal generic model. Inthe next step 
tracking and motion estimation of the hypothetical ob- 
ject is solved by a recursive estimation algorithm result- 
ing in the state of the object relative to the road ob- 
served. Finally, the verification of the hypothesis is 
performed exploiting knowledge about the possible 
range of the estimated motion parameters and con- 
sistency over time. The validity of the approach pre- 
sented was tested with synthetic image sequences, and 
promising results have been obtained indicating that the 
application to noisy measurement data from real world 
scenes should lead to useful results. 
The main objective of this paper was to introduce a 
systematic method of dealing with partial occlusion of 
objects. For confirmation of the hypothesis it is conceiv- 
able thatthis approach may be extended to checking the 
consistency of an hypothesized model shape; addition- 
ally, it may become more robust by using information 
from the images such as color, texture, segmentation, 
etc. 
REFERENCES 
[Brady 81] M. Brady: "Computer Vision", Noth-Holland 
Publishing Company, Amsterdam, 1981 
[Chellappa et al. 90] R. Chellappa, T.J. Broida: "Recur- 
sive 3D-Motion Estimation from a Monocular Image 
Sequence", IEEE Aerospace and Electronic Systems 
26(4), pp 639-656, 1990 
   
     
   
   
    
     
  
   
   
   
   
    
    
   
    
   
    
  
   
   
   
    
   
   
     
   
    
  
    
  
      
     
    
  
    
[Enkelmann 90] W. Enkelmann: "Obstacle Detection by 
Evaluation of Optical Flow Fields from Image 
Sequences, Proc. of the Computer Vision ECCV 90, pp 
134-138, 1990 
[Brammer, Siffling 75] K. Brammer, G. Siffling: "Kal- 
man-Bucy-Filter, Deterministische Beobachtung und 
stochastische Filterung, Oldenbourg Verlag , 1975 
[Dickmanns, Christians 89] E.D. Dickmanns, T. Chris- 
tians: "Relative State Estimation for Autonomous 
Visual Guidance of Road Vehicles", IAS-2, in Kanade 
et al.(ed), Amsterdam, pp 683-693, 1989 
[Dickmanns et al. 90] E.D. Dickmanns, B. Mysliwetz, T. 
Christians: " Spatio-Temporal Giudance of Autono-. 
mous Vehicles by Computer Vision", IEEE Transac- 
tiona on System, Man and Cybernetics 20 (6), Special 
Issue on unmanned Vehicles and Intelligent Robotic 
Systems, pp 1273-1284 ,1990 
[Dickmanns 91] E.D. Dickmanns: "Dynamic vision for 
locomotion control - An evolutionary path to intel- 
ligence", CCG-Lehrgang SE 3.02, 1991 
[Maybeck 79] P.S. Maybeck: "Stochastic Models, Esti- 
mation and Control Volume 1--2, Academic Press, 
1979, 1982 
[Regensburger, Graefe 90] U. Regensburger, V. 
Graefe: "Object Classification for Obstacle Avoidence", 
Proc. of the SPIE Symposium on Advances in Intel- 
ligent Systems, Boston, pp 112-119, 1990 
[Schick, Dickmanns 91] J. Schick, E.D. Dickmanns: 
"Simultaneous Estimation of 3D Shape and Motion of 
Objects by Computer Vision" IEEE Workshop on 
Visual Motion, Princeton, N.J., 1991 
[Schick 92] J. Schick: "Gleichzeitige Erkennung von 
Form und Bewegung durch Rechnersehen", Disserta- 
tion UniBwM, Fakultát für Luft- und Raumfahrt- 
technik, 1992 
[Solder, Graefe 90] U. Solder, V. Graefe: "Object De- 
tection in Real Time", Proc. of the SPIE Symposium on 
Advances in Intelligent Systems, Boston, pp 104-111, 
1990 
[Thomanek, Dickmanns 92] F. Thomanek, E.D. Dick- 
manns, "Obstacle detection, Tracking and State Estima- 
tion for Autonomous Vehicle Guidance", IEEE/RSJ 
Int. Conf. on Intelligent Robots and Systems, Raleigh 
N.C., 1992