Fig. 7: Reconstructed limb orientations 
image 
signal 
  
measurement: 
edge searches 
Y 
correspondences e 
! 
  
  
  
  
  
  
  
4. 
representing edges of thighs and shanks can be seen; 
they are used to calculate the leg skeletons. 
In the initialization phase, the operation of image 
analysis with the m d of forming the inductive step 
towards an initial object hypothesis still needs some 
improvement. Provided that a valid object model has 
been found, feature tracking subsequently can be 
supported by the process model that helps in 
systematically deriving search regions and test features 
from one hypothesis. The measured features either 
verify the supposition, and the model will be updated, 
or falsify it with the rejected model being replaced by 
another one. In the future, the search area will be 
confined by road recognition because only in the 
vicinity of lanes objects are presumed to be obstacles. 
32. State estimation 
The second part of the analysis task resides in the 
estimation process incorporating the two major steps 
of recursive estimation [Kalman 60, Maybeck 79]: 
correction and prediction (see figure 8). Estimation 1s 
supported and updated by pre-interpreted features 
edge regions 
Meme hh hh hh hh heh hh hh hh hh heh IMRHIIHHÍHS 
1 
expected edges 
  
  
  
comparison: 
subtraction, assessment 
  
  
  
geometry 
movements 
prediction (for t,k 4- 1): 
separate state extrapolations 
correction, model adjustment (at t,k): 
decoupled innovations of 
pose (position, orientation) 
visibility expectations, 
pruning 
prediction error x^' 
tate estimates x^ 
predictions x* 
visible 3-D edges 
  
| perspective mapping, 
| presumed edge areas 
  
  
  
  
  
  
  
Fig. 8: Estimation procedure