and (8) are preseted in colum 3, 4 and 5. Of course all of 
the results are wrong and the mistakes are too difficult to 
detect by the single criterion itself. The asterisks in 
6th colum indicate the mistakes detected by r€0.50. Anyhow 
all the mistakes can be detected by the dual criterion of 
Organd Pf, (asterisks in colum 7 ). 
Table 4 Gross error detection 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Expected, loro, Matching Results Error Detection 
Pa rallax : r = max | 0 = max | P,= min r Xo5, a AND p, 
Pur m 104 96 | 80 96 0.52 . 
F 73 105 97 | T 97 | 0.63 | . 
P Um» | 105 | 97 | 81 | 97 0.57 . 
[^ 5 | 107 107 rd 91 | 107 0.97 | . 
76 | 108 92 | 92 | 108 | 0.65 * 
76 3 108 EP. us 108 | 116 | 0.62 | , 
Fe | 108 | 116 116 108 0.57 | . 
78 | 110 | 118 118 86 0.55 | . 
81 | 113 121 121 89 "09:491 — ! : 
81 | 113 89 121 89 | * 0.40 | . 
83 | peor {orm 123 91 --—.9.39 4 . 
  
  
  
  
(c) A testing window 
This experiment is carried out for verifying the feasibility 
Of the dual criterion. The photo data ana scanning parameters 
are presented in Table 5. 
Table 5 Photo data and sampling parameters 
  
  
  
  
  
  
  
  
Principle Distence 114.207 mm 
Photo Format 18x 18cm 
Photo Scale = 1:16000 
Scanning Interval in x/y bo/fo A 
Pixel Size 50 x Sur 8) 
CE Testing Window Pha 1024 x 2000 P1 X 1 
Terrain Type Mountain(parallax 8.6mm) - 
  
  
  
  
The epipolar geometry and the hierarchic structure(Makarovi € 
1980) of 6 frequency channels are applied in this experiment. 
Matching is performed along the profiles parallel to epipolar 
lines and with interval of 0.4 mm in photo scale. The paral- 
laxes obtained from previous profile is entered into the 
next as the initial position and matching in a higher frequency 
- 424 =