anbul 2004 
NTATION 
FRAMES 
rity model, 
Eq. 4 to a 
itions. The 
orm as: 
Jes, and 
and 
he vector 
and Jj: 
; 4nx3n 
unctions in 
orientation 
1) is the 
the image 
quence to 
frame was 
he camera 
lated using 
ve pairs of 
ra station, 
Eq. 5 for 
vw 
(6) 
e exterior 
s follows: 
zl e 
— 
ast image 
NTATION 
FRAMES 
frame, the 
*P of three 
inimum of 
using two 
  
  
  
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004 
camera orientations of the previous image frames. The 
calculation of the GCP of the optical flows is as follows: 
X X, Lu LU 
Y-Y .Q,225 Me iP. 
XXe, C 
2(Z,-Z, AME ATE, 
Wa 
Y — 
i Y 
where U -myx;tm,y,tm,f 
V mox t HV, tmo 
W=m,x, + m,,y; + my, f . The GCP is found with the aid 
of the classical least-squares solution. 
> 
IX. 2. ZV (fy 4D, (8) 
Le À ~-U,_,/W,_, 
Q d «E nm. 
where A= Er and 
10 -U,/W, 
0-1 > V. "nm 
X502 = 2 30 IW, 2 
Y, =2 Z, e. IW, 
= "T ED T . If Eq. 6 is given at 
Xa = Zr ite /W, | 
Yi = ZA fW 
least three GCPs, the camera orientation is calculated as 
follows: 
NEN ep (9) 
After a few iterations of Eq. 10, we can determine the exterior 
orientation of the camera from the third image frame to the last 
image frame as follows: 
= = es E I 
lo, 9, K, X Li Y Z, of F = lai % K, A y Zi of = A 
4. MULTI-BASELINES FOR CREATING VIDEO 
MOSAICS 
As stated above, the 2D-image mosaic technique projects all 
of image frames on a single baseline to create an image with 
wide range. Since it can’t be applied to image frames taken 
from a rotating camera, to solve the drawback, this paper 
proposes a novel method for creating image mosaics in 3-D 
space in case of using an image sequence taken from a 
translating and rotating camera. 
729 
Q Feature point 
Verres 
siurod o1nuay 
Jo yıdap 3Se.10AV 
= 
      
    
  
Object ; 
Baseline -————-—-7- ; 
„Image plane z—— 
(c) 
Fig. 2. Independent baselines by using camera pose and 
average depth of feature points per image frame. 
The core of the proposed method is determining dependent 
baseline to per image frame for creating multi-baselines on 
which all of image frames are projected. The pose of the 
baseline of image frame is the same pose of camera. The 
perpendicular distance between camera focus and baseline is 
evaluated as the average distance of optical flows like Fig. 2(a). 
the spatial transform of optical flows from the world coordinate 
to the coordinate of image plane is following as: 
zr x, Y, 
FET. Zu ists EM dx X ZA 
(10) 
Where (X,, 
pose. Since the pose of baseline is the same pose of image plane, 
in case of Fig. 2(b), the multi-baselines are the thick dot-lines. 
In case that each image frames are projected on itself baseline, 
the result will be Fig. 2(c). 
Y,,Z,) is camera station and (c, , &) is camera 
4.1 THE MODIFICATION OF FAULT BASELINE 
From now, this section only describes the modification of 
multi-baseline in XZ coordination as compensating the reverse 
image rotation of x axis to itself image frame. Since the case of 
Fig. 2(b) based on a camera of which the image motion is the 5 
DOF, (7 5115/7: 9, 7); with. the exception of the image 
rotation of x axis, a. is an ideal state, it is able to create video 
mosaics in 3D space. In most instances, the motion of a camera 
has 6 DOF including the image rotation of y axis which is one 
of the difficult things to create video mosaics like Fig. 3(a) and 
Fig. 4. Therefore there are needed to modify the baseline like 
Fig. 3(b) and Fig. 5.