International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
  
of rectification is illustrated as Fig.3. For close-ranged DC 
images, the coordinate system of object space is set up as Fig.4 
illustrated. The facade textures are rectified onto plumb plane 
accordingly. 
41 The Determination of Vanishing Point and Initial 
Values of DC Angular Parameters. 
Vanishing point V (xp, vj.) is the intersection point of a set of 
straight lines which are parallel in object space. It means that 
each straight line such as line ij belong to this set must be 
passed through the vanishing point as shown in Fig.5. So Eqs.2 
is evolved. 
(2) 
Where; dis dhe 
distance of point / to 
line iV. 
  
  
From Eqs.2 a model 
of adjustment system 
of observations and 
parameters is deduced 
to determine 
vanishing point. Then initial values of DC angular parameters 
are decomposed according to vanishing point geometry (B. 
Caprile. et al., 1990). 
Fig. 5 The constrain of Straight 
lines bundle 
4.2 The Calibration of DC Angular Parameters 
Two algorithms of calibration of DC angular parameters are 
presented here. 
4.2.1 The Calibration Employed the Constraint of Straight 
Lines Bundle: According to vanishing point geometry 
vanishing point coordinate ( xy , vy ) in Eqs.2 is replaced with 
interior and exterior orientation (angular) parameters. Then a 
model of adjustment system of observations and parameters 
employed the constraint of straight lines bundle is deduced to 
calibrate DC angular parameters while interior orientation 
parameters are fixed in the adjustment system. Because the 
initial value of orientation parameters is decomposed from 
vanishing point coordinates, the accuracy is not always high 
owing to the quality of images and the error of line extraction. 
Therefore, It has a small convergent radius that the model of 
adjustment system employed the constraint of straight lines 
bundle. As a result, the model of adjustment system is not 
robust. 
In this article, it’s introduced to control the adjustment system 
that the constraint of known orientation of parallel lines in 
object space. 
4.2.2 The Model of Adjustment System Controlled by the 
Constraint of Known Orientation of Parallel Lines in 
Object Space: There are two group lines parallel to X-axis and 
Y-axis in object space respectively. Their projections in image 
space are involved in the adjustment system, consequently the 
projections of these two group lines in rectified image should 
be parallel to X-axis and Y-axis in object space respectively in 
734 
ideal instance (As illustrated in Fig.3 and Fig.4). Namely Eqs.3 
and 4 are deduced. 
[n X direction: 
X; Hy byf 
  
f bx;*by;-byf — 
  
: (3) 
CX; t €3yj 7 €3/ Cx; tcov; ef 
In Y direction: 
AX t O2 Y — aif Lf 2n Tay, — af (4) 
CX * Co Vg — of Xp +C2Vy — cf 
With: 
x; 3 y; 30x, 
line parallel to X-axis 
(x, À Fm ) Xp 9 y, ): the coordinates of two end points of 
y; ): the coordinates of two end points of 
line parallel to Y-axis 
f : focal length 
a 45 (34 
by b, by |: rotation matrix 
Eqs.3 and 4 can be control condition for the calibration of DC 
angular parameters. Therefore it's deduced that a model of 
adjustment system combined with constraints of straight lines 
bundle and known orientation of parallel lines in object space. 
As the experimental results showing, the model of adjustment 
system deduced here has the characteristics of large convergent 
radius and high stability. 
5. THE AUTOMATIC MOSAIC OF FACADE TEXTURE 
Usually for big buildings, facade texture is covered by several 
images instead of only one image. Therefore to acquire the 
whole facade texture, mosaic process is required. Since facade 
textures are all rectified onto plumb plane, only the 
displacements d, d, in X-axis and Y-axis respectively are 
needed to determine in mosaic process. Thereupon, the question 
is how to determine the displacements d,, d,. 
  
+ 
SF 
cet 
fm. 
I ES ed 275 
1 
t 
d 
    
  
E irr 
mcm | 
"E 
EE 
i 
j 
mmm un zn ud Ÿ 
   
Fig.6 Corresponding points matched 
Firstly, corresponding points are automatically matched 
between each two adjacent images. As Fig.6 illustrated, there 
are 295 corresponding point pairs acquired. Because of the 
similarity of texture, some false point pairs are also obtained. 
Therefore, an optimal point pair should be picked out as mosaic 
point pair by which the displacements d,, d, are determined. 
The strategy of selecting mosaic point pair is then presented. 
Inter. 
  
5.1 
Coef 
| 
| 
| 
| 
| AS w 
| whet 
| the c 
| pair 
| eram 
coe 
| d, ar 
| are 1 
| becai 
| coeft 
of te: 
| 
| 
| 
| 
| 
Henc 
const 
5.2 t 
Coef 
Relat 
corre 
paral 
mean 
belov 
F; 
Q; 
Pi 
| 
| 
| 
| 
| 
| 
| 
| 
| 
| with: 
| 
| 
| 
| 
| 
| 
| 
| 
|