itch 
'om 
inn, 
las 
ase 
the 
the 
ion. 
)an 
n, a 
tual 
ning 
f the 
view 
nta- 
lized 
and 
t al, 
The 
Handmann, Uwe 
  
The prepared data (e.g., bird's-eye view, TTC-estimation, object-list) are passed to the scene analysis for determination 
of global measures and to behavior planning. 
5.2 Scene Analysis 
The scene analysis supports the driver assistance by evaluating the actual traffic condition as well as the scenery. The traffic 
condition is assessed by evaluating the information from scene interpretation. According to the actual traffic condition 
and the planned behavior a risk-factor 
  
n 
| reni 
ae n 
i=1 $ 
for actions is estimated. The risk factor evaluates the sum over all relative velocities |V-e1,i| between observer and objects 
normalized by the distances d; to each of the n objects detected in the scene. The scenario can be determined using 
GPS and street maps for investigating the kind of street, e.g., highway, country road or urban traffic. Depending on these 
scenarios, different objectives have to be taken into consideration. The determined risk factor as well as the classified 
scenario are passed to behavior planning. The scene analysis does not have be performed for every image, as e.g., the 
kind of road does not change abruptly, so the evaluated data are stored in the knowledge base so as to remain available to 
behavior planning. 
6 BEHAVIOR PLANNING 
Behavior planning depends on the given task and on the scene interpretation. Different solutions for the planning task 
are possible. A rule-based fuzzy logic approach is described by (Zhuang et al., 1998) . An expert system is shown 
by (Sukthankar, 1997). At the Institut für Neuroinformatik a system with complex behavior by means of dynamical 
systems for an anthropomorphic robot was realized by (Bergener et al., 1999). A behavior control for autonomous vehicles 
controlling the steering angle and the velocity by neural field dynamics (Amari, 1983) was presented by (Handmann et 
al., 2000b). In the present system, a flow diagram controlled behavior planning realizing an ICC was integrated. 
Behavior planning for the observer results 
in advice to the driver not only based on 
the intention of following the leader but 
also taking into account the safety of the 
own vehicle. This means that the object knowledge 
cannot be followed or might be lost in case | 
of other objects or obstacles endangering 
the observer. The signal behavior for the 
main tasks is described by the flow dia- 
gram shown in fig. 8. 
Information 
from scene interpretation, 
   
    
  
  
Detect: 
New Leader 
  
  
  
At first it has to be determined whether the 
leading object could be detected. If not, a 
new one is searched automatically in the yes 
same lane. A preceding object is recog- 
nized as a leader if it has a consistent tra- 
jectory on the actual lane, is positioned in 
a reasonable (detectable) distance and if it 
could be detected at least once in a cer- \ 
tain number of previous time-steps. If no Follow Signal: 
leader can be found, advice for decelera- Leader safely Lane Change 
tion or no change in action is given ac- 
cording to safety considerations. A safe | 
and comfortable driving velocity has to be Safety and 
achieved by taking the security distance, Comfort Drive 
the TTC to other objects and possible col- 
lision positions as well as a general risk 
factor (eq. 1) into account. In the case 
of leader detection it is of interest whether 
the leader can be located in the same lane. 
  
  
  
À 
Safety and 
Comfort Drive 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Figure 8: Flow diagram of signal behavior for ICC. 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 351