fixed interval, so one pixel height of image data is just depends 
on traveling velocity of the Line Camera. ( Table 2. ) 
  
LINE CAMERA 
  
LIGHT 
Railway 
  
  
SMOOTH SURFACE 
  
  
  
LINE CAMERA 
Railway 
  
  
ROUGH SURFACE 
  
  
  
  
  
  
Figure 4. Illumination for taking Flaw Image 
  
RAILWAY SURFACE 
Gauge Corner Side ( Inner side of tracks ) 
NEGATIVE IMAGE DATA 
GED MODEL 
   
    
   
   
    
  
  
Fine Flaws 
(0.1mm - 0.3mm 
2 
Virtical resolution will be changed 
by the velocity of the Line Camera 
movement , So the fine crack will 
be recorded like as a Bar Image . 
1 Scanning AREA of Line Camera 
Polished Part like as a mirror by wheels : Dark part 
Rusty Part ( wheels not touch ) : Bright part 
  
  
  
Figure 5. Flaw image mode in Recorded Image Data 
When the Line Camera travels at a speed of 20 kilometers per 
an hour ( 5.55 meters per a second ), a vertical resolution of 
each scan line is approximately 0.56 millimeters. In such 
situation, as almost of fine flaw width is less than 0.3 
millimeters, it is impossible to distinguish each flaw from the 
Line Camera image data. 
However, the purpose of measurement is not to count the 
number of flaws, but to measure the length of flaws form the 
gauge corner side. So, if it is possible to distinguish flaw 
part form polished part, it means achieved the purpose. 
RESULT OF FIELD TEST 
Field test had held on actual rail way track of TOHOKU 
Shinkansen in mid night. — Figure 6. and Figure 7. are Picture 
of original recorded image data which had taken at field test at 
a speed of 20 kilometers per an hour and processed image data, 
It is able to distinguish between rusty part and polished part 
from the image of few flaw part of original data. And also 
possible to find flaw part in polished part from the image of 
cranky part of original data. 
Table 3. is the other result of field test. At this case, the 
image data had taken with three kind of illuminate strength. 
This result says the illumination not in right condition is easily 
causes incorrect result of recognition. In right condition of 
illumination, result value of analysis close to read data. 
Table 3. Compare with Real Value and Calculated Value 
( With three kinds of Illumination) 
  
Result by Computer analysis (mm) 
Too Bright| Standard | Too Dark 
test1 | test2 test1 | test2 | test! | test2 
14K990m ( Real Flaw length : 20mm [median] ) / 
14K 997m - 14K 998m| 49] 53] 19] 15] 58] -59 
14K 996m - 14K 997m 0 Si 120 16 57| 59 
14K 995m - 14K 996m 5 6 9| 16| 56| 59 
14K 994m -14K 995m| 52| 33] 22| 16| 55| 59 
14K 993m - 14K 994m| 53| 34| 20| 18| 58| 59 
14K 992m -14K 993m| 51| 48| 19| 18| 59| 59 
14K970m ( Real Flaw length : 20mm [median] ) 
14K 972m - 14K 973m| 51 4| 19] 34 22 
14K 971m - 14K 972m 6| 50, 17} 31] 21 
14K 970m - 14K 971m 5] s1] 20] 33] 18 
14K 969m -14K 970m] 50| 51| 20} 32] 17 
14K 968m - 14K 969m 5 3|. 20]  20| 15 
14K 967m - 14K 968m 9 di 13) 31) 14; 11 
Test1 : Go to Far Side Test2 : Come from Far Side 
Velocity of the line camera movement : 20 km / hour 
Too Bright : Too strong Illuminated 
Standard : Standard Illuminated 
Too Dark : Too weak Illuminated 
  
LOCATION 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
O|OC|OJ|O|O 
  
  
  
  
  
  
  
  
  
  
CONCLUSION 
There are confirmed, 
(1) Impossible to distinguish each flaw on rail surface with the 
present line sensor camera system. 
(2) The flaw area on rail surface is able to detect from the 
image data which is recorded by the line sensor camera. 
The greatest problem is to shortening the processing time on 
the engineering work station. Still now, it takes about Five 
days to process the data of 60 kilometers length. ( Approxi- 
mately 55G bytes ) But the progress of performance of 
computers is still going on, so it is expected that this system is 
possible to use in practically in the near future. 
Hereafter there are some another plan with this system, that an 
expansion of items of track materials, for automatic inspection 
such as damage rail surface or another part of tracks. 
398 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996 
  
ABSTRACT 
Bccur acy 
The INCH 
dimensic 
à set ol 
Monte { 
scienti: 
part of 
reductit 
some of 
terrest: 
identifi 
iNagara. 
spe th: 
zigsulat: 
indicate 
be pass. 
oí vari 
However. 
some di 
closely 
possible 
The sul 
outside 
number 1 
aea 
and her 
substan 
observa! 
such se 
Applica 
photogr 
possibl 
intreasi