International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B5. Istanbul 2004 Internatior 
SIENNE 
  
1 mm left. 1 mm forward 
2500 D T pem 
  
  
  
  
  
  
  
  
  
  
2000} [ + There are 
a 5 " one point 
1500F "I + : 
be stitch: 
1000 4 & projective 
s a Q 
500} + y 
Q © -6 a Xi 
3 X9 = — * 
ot e 1 amg’ 
-500 
$ © or the rc 
- 1000} 4 ; 
: s images pr 
-1500r No + 
SA A © 
: : The trans 
7 E 2 2324 D di 7 5 we aa Tha A 0 TT men 2000. T7. 2000 be solved 
Figure 2. Targets for camera adapter adjustment. points ar 
Figure 4. Deformations (x 50) of the 1 mm deviated set. the same 
overlappi 
S 2 mm left, 2 mm forward parameter 
Ed 7 ; TE using the 
2000} Xo adjusted 
* © level difl 
af A i ta 
E (Szeliski, 
1000 }- b & 4 
€ a e In the oth 
so s e NN on image 
or e. | either sin 
model is 
-800F 4 
= ® 
-100l I a k 
1000 
© =D 
n hj 
-1500r e 6- 
FOO) rrr ns ere treet en me ee ms rt Sf ron reat 
7300 -2000 71000 0 1000 2000 3000 where a a 
Figure 3. One of the four panoramic image mosaics. The size of unknown 
the image is 7219 x 6004 pixels. Figure 5. Deformations (x 50) of the 2 mm deviated set. optimal r 
grev valu 
Next, the locations of 19 checkpoints were measured on all ° $ mm left, 5 mm forward rotations 
; : : i 2500 : T : e re 
mosaics using least squares matching. In some checkpoints the be projec 
signal to noise ratio was so small that they had to be measured 2000 - ; \ f small, the 
manually. In Figures 4, 5 and 6 the differences between the bis % i images th 
: R. 1500 zi 
non-deviated set and the 1 mm, 2 mm and 5 mm deviated sets 
are shown, respectively. Deformations on the image edges are 1000 | ? 1 Accordin; 
bigger than in the middle of the image. The reason is clear; xl $m E more reli 
transforming of the concentric images to one common plane 3 9 L5 images al 
amplifies the errors of the outermost images. The average of " process t: 
deformation of the 2 mm deviated set is slightly smaller than the : exactly cc 
deformation of 1 mm deviated set (see Table 1), but the 5 mm 2 x exactly c 
deviated set is clearly worse. One would expect that the 2 mm ~1000}- s 1 deformati 
deviated image set deviates more than 1 mm deviated, but this is m € 2 created ty 
not always the case, because the magnitude of the eccentricity is € ru we 
: fac : en i ACCOUL sviati > zoo esce à à Ed checkpoir 
not the only factor to be taken into account. Deviating the 2000 — Sis E # oi = an = KP | 
camera to different directions would have produced different relative t 
Both new 
image an 
images w 
deformations. Also the photographed objects influence the 
deformations. The more depth differences there are in the 
objects, the less deviation is allowed in concentricity (Luhmann 
Figure 6. Deformations (x 50) of the 5 mm deviated set. 
  
  
  
  
  
  
  
  
  
  
et al., 2003). As can be seen later also the stitching order affects image. F 
; ; Sn set | mm 2mm 5mm anticlocka 
to the deformations. Because of the complexity of the E XI PUE = : 
: ; : is : i mean 1.9858 pix 1.6347 pix 2.7576 pix image. TI 
deformation of the panoramic image mosaic the previous results : : AAT : = 
; : ne. std. 1.3273 pix 0.9176 pix 2.0554 pix 2. It can l 
are just suggestive. So even small eccentricities can cause : 6.4973 pix 3.7748 ni 7.1805 pix 
deformations of several pixels. ax PIRE EE La {ace ie 
Table 1. Statistics of "the deformations caused by different deformati 
eccentricities. experienc 
a certain 
For exan