El-Hakim, Sabry 
the cabinet (4.B) by fitting a plane using the corners of the surface. We then use a single image and pick any point on 
the surface. The 3D coordinates of this point can be computed from the image coordinates and the plane parameters. 
  
    
(A) (B) 
Figure 4: Using available features to fit planes and then extrapolate featureless points from single images. 
2.2.3 Adding Range Data. This involves matching and integrating local detailed points obtained by a laser scanner to 
the global model obtained by the image-based method. This is best described by an example. In figure 5, nearly all the 
structure is easy to model with images 
taken by digital camera. However, parts of 
the surface contain fine geometric details 
that will be very difficult or impractical to 
model from images, such as the one shown 
in figure 5.B. Those parts are best acquired 
by a laser scanner and added to the global 
model created from the images. To register 
the detailed model shown in figure 5.C we 
measure several features, usually 6, using 
the images then extract the 3D coordinates 
of the same features from the scanned data. 
This is done interactively using intensity 
images generated by the laser scanner. A 
similarity transformation is then used to 
register the two coordinate systems of the 
two data sets. 
  
  
  
The same idea is applied to integrate 
models reconstructed from different sets of 
images (item 6 in section 1.2 above). 
  
(B) (C) 
Figure 5: (A) Images for global modeling, green boxes are the areas to be 
scanned. (B) The range sensor. (C) The detailed model of area 1. 
3 EXPERIMENTAL RESULTS AND ACCURACY ANALYSIS 
We have extensively tested the performance of our system. Different types of object and environment have been tested 
to demonstrate the concept of automatic point densification, measuring details that show in one image only, and 
integrating different sensor data. Rarely any of the publications describing 3D modeling techniques address the issue of 
geometric accuracy of the model. Here, we quantitatively evaluate the accuracy using directly measured dimensions 
between features or known geometric parameters of objects (for example a radius of a sphere). We used simple objects 
for the accuracy tests, however, this should be sufficient to represent the accuracy for any other object. 
3.1 Example Models 
Figure 6.A, B, and E show examples of models where most of the surfaces are planes. The use of plane fitting and 3D 
from a single image was a must to complete those models. Many of the surfaces had no features except when they join 
other surfaces. This made it impossible to obtain 3D from multiple images. Figure 6.C shows a structure where the 
surfaces are quadrics. To get smooth surfaces, thousands of points were added automatically after fitting quadrics to 
manually extracted points. This model took only 20 minutes to complete. Figure 6.D shows a structure consisting of 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 207