El-Hakim, Sabry 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Sensor Selection & 4— i Data Collection Points in 
Sensor Placement multi- images 
x v 
! ° ° 
Registration Points in single 
> T image on a 
= defined surface 
= i Y 
i Organized Points Unorganized Points Mesh 
# subdivision 
Determine Connectivity Random points 
Y on a defined 
| 
30 Modeling «+ Add Points Se 
  
  
  
  
Figure 2: Summary of the overall procedure 
2.2.1 Registration of Main Images. The images are displayed in the proper order and common points are extracted and 
labeled interactively. If correct scale is required, some distances in the scene are also measured. Control points, if 
available, or data from positioning devices (e.g. GPS), may also be utilized in this step. A bundle adjustment is carried 
out to register the images. In addition to registered images, we also have a number of unorganized scattered 3D points. 
2.2.2 Segmentation, Fitting, and Automatic Point Densification. The 3D points generated so far are not sufficient for 
modeling. They are also unorganized, thus the connectivity, or the topology, is unknown. Three interrelated operations 
are needed in order to add sufficient points and organize them to create a complete 3D model. Segmenting or grouping 
3D points into sets each belonging to a different surface is the first step. Most existing automatic modeling methods 
were developed for organized 3D points, such as the range images obtained from a laser scanner [Soucy et al, 1996], or 
unorganized points belonging to specific types of object [Hoppe et al, 1992]. Unorganized points obtained from features 
on various surfaces on different objects are almost impossible to model automatically since they are subject to many 
possible interpretations. In our approach, the scene is visually divided into surface patches, each is triangulated and 
texture mapped separately. Although this is specified manually by a human operator, it is easy to do since all that is 
required is to draw, with the mouse, a window around the points belonging to the same surface set. Once this is done, 
the modeling will be carried out fully automatic. Each set may be on a different surface, or the same surface may be 
divided into several sets depending on the complexity of its shape. 
   
(A) (B) 
Figure 3: Using existing features to fit a known shape (sphere) then automatically adding any number of points. 
Using any existing features on the surface set, 3D-point computation is first done interactively [Figure 3.A]. These 3D- 
points are then used to compute the function defining the surface, using least squares fitting. The function is in turn used 
to automatically generate new points on the surface [Figure 3.B, half with texture]. On more complex surfaces, we can 
only interpolate between the existing triangles by a subdivision technique [Zorin, 1997]. For partially occluded surfaces, 
a single image can be used to extend the surface. For example, in figure 4, we can extend the floor (4.A) or the side of 
  
206 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 
  
  
the c 
the s 
  
3.2.5 
the £ 
struc 
take 
the : 
that 
mod 
in fi 
by a 
mod 
the « 
mea: 
the i 
of th 
This 
imag 
simi 
regi: 
The 
mod 
imag