Eq.(2.1) the distance along the line of 8 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
a m sight has been decreased by 7% along the num 
s = frame border - by 23% and along the val 
: distant border - by 1% approximately. Deo 
8 Here W,-0.2. num 
= max 
E Enlarged by three times approximately 
: frame difference fields of processing num 
8 = fragments are shown in Fig.6. Before the Thi 
a i subtraction the fragments have been Bha 
g preliminaryly combined in accordance with 
a \ the serial correlation algorithm (Barnea, Thu 
3 1972). Correlation processing areas are abs 
i drawn by white lines in Fig.4. In Fig.6 
S the fragment interframe displacement is 
8 - À one pixel along Y datum line and two 
s | 6 N pixels along X one. 
z 
E 3 > You can see in Fig.6 that the real speed Rar 
i N of moving pattern is less than design one B a 
# 3 = which causes partial pattern shading. 175 
à fo] 
ms F PT T 
T "m ees M Re DR, Fragment frame difference fields have Baa 
been filtered by trunckated to pattern Pu 
P A 3 i area constant impuls response filter. We 
eS A ye Satis use extremal filter outputs to decision LION 
ho tion: (Fire ob) 02 > typical landscape function calculation. Decision function Dud 
Broken lines is experimental and smooth Telus Sor each fragments are shown in Cla 
lines is approcsimated curves. T Wil 
Hol 
Ide 
Sen 
pp. 
  
Fig.6 The result of preliminary combined fragments subtraction. 
Table 3. Extremal filter outputs 
and decision function values 
  
Fragment number 
  
  
Features 
I 2 3 4 5 6 
oh 3.51 1.0] «5 1 0.6 d 4.5 1.2.7 
min 
on 5.3 | 8.0.1..8.9 12] 78] 47 
TOX 
P,,(X) | 00001 |.00012|.00000| .00032| .00001 | .00001 
Ban . 00000 | .00001 | .00000 | .00001 | .00000 | .00000 
X) .00004 | .00009| . 00000, .03250| . 00002 | „00012 
PU) .00379| .00036| .02089| .00015, . 00868, . 00063 
P nx . 00004 | .00000 | .0001 4 | .00006 | .00003 | .00003 
(x) .00283 | .0081 4 | .00037 | .00251 | .00064 | .001 21 
  
  
  
  
  
  
  
432