e» 
Fuse, Takashi 
  
Labels are considered to be consistent if they represent nearly the same displacements, 
  
xg — Ax y? t (Ay - Ay)? €T. (17) 
L" in the equations (15) and (16) is a set of labels which satisfy equation (17). When vehicle a; does not have the label 
," Which satisfy equation (17), degree of consistency is P(A) —0, otherwise PA) 20. Vehicle a, are selected 
by satisfying equation (18). 
  
JG xy to» sn. (18) 
For opposite label  ', degree of consistency is defined by (16). 
Improvement of label probabilities is accomplished by applying following equation, 
= poe 
pe” (A) = A. (=12,......, Lp), (19) 
vy iUo 
lel; 
where 
/(new) .. p(old) 
pena en y, e 
and 
PE, ) = BON 4 + BBO) + COP (A) 0-12, D). Q1) 
For opposite label probabilities, 
ed Ao) 
Qo (Au )= J — (I71,2,....... 9L. p), (22) 
: $0 Ou) 
l'e Li 
where 
Qi" Q',)- Q?^ 0), Gn 
and 
Que (A ) = + (A, XA + BOB (A )+ Cpeto Ay )) ; (=1,2, es ee 1). (24) 
Parameters A, B and C in equation (21) and (24) are positive constants which influence the convergence characteristics 
of the model. The role of 4 is to delay the total suppression of unlikely labels. The role of B and C is to determine 
the rate of convergence. The larger the value of B and C relative to À, the faster will be the convergence of similarity 
assignments. 
The complete procedure to estimate the most likely correspondence for each vehicle can be summarized as follows. 
Each possible correspondence is assigned an initial probability. These probabilities are iteratively refined using 
equation from (19) to (24). This procedure is repeated until the probabilities reach steady states, but in practice we 
may need to arbitrarily stop it at 50 iterations (Barnard and Thompson, 1980, Takagi and Shimoda, 1991). 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 281