-60- 
the illumination or ambient light problems. Active sensors (e.g. 
laser scanners) [Besl, 1988, Rioux et al, 1987] avoid these 
limitations by creating features on the surface by controlled 
projection of light. They have the advantage of acquiring dense 
3D points automatically. Recent advances in laser, CCD 
technology, and electronics made possible detailed shape 
measurements with accuracy better than 1 part per 1000 at rates 
exceeding 10,000 points per second. The scanning and imaging 
configuration determine the point density. Many also produce 
organized points, in the form of array or range image, suitable 
for automatic modeling. A single range image is usually not 
sufficient to cover an object or a structure. The amount of 
necessary images depends on the shape of the object, amount of 
self-occlusion and obstacles, and the object size compared to 
the sensor range. The 3D data must be registered in a single 
coordinate system. Several registration techniques are available; 
most are based on the iterative closest point (ICP) approach. 
For the approach to converge to the correct solution, it needs to 
start with the images approximately registered. This will require 
either the knowledge of sensor positions or manual registration 
using features. Once the range images are registered in a single 
coordinate system, they can be used for modeling. This step 
reduces the large number of 3D points into triangular mesh that 
preserves the geometric details and at the same time suitable for 
fast rendering [Curless and Levoy, 1996, Soucy et al, 1995]. In 
this process, the areas where the images overlap must be 
integrated to create a non-redundant mesh. Other requirements 
include filling of holes and removal of any outliers. 
There are two main types of range sensors. The first is 
triangulation-based that projects light in a known direction from 
a known position, and measure the direction of returning light 
through its detected position. The accuracy of measurements 
will of course depend on the triangle base relative to its height. 
Since, for practical reasons, the triangle base is rather short, 
triangulation-based systems have a limited range of less than 10 
meters (most are less than 3 meters). The second sensor type is 
based on the time-of-flight. Those measure the delay between 
emission and detection of the light reflected by the surface, and 
thus the accuracy does not deteriorate rapidly as the range 
increases. Time-of-flight sensors can provide measurements in 
the kilometer range. 
Notwithstanding the advantages of range sensors, we should 
mention some drawbacks. They can be costly, bulky, affected 
by surface reflective properties, and may be complex to operate 
and calibrate. Also a range sensor is intended for a specific 
range, thus one designed for close range may not be suitable for 
long range. Comparative evaluation of image-based and range- 
based methods can be found elsewhere [El-Hakim et al, 1995]. 
2.3 Image-Based Rendering 
In image-based rendering (IBR), images are used directly to 
generate new views for rendering without a geometric model 
[e.g. Kang, 1999]. This has the advantage of creating realistic 
looking virtual environment at speeds independent of scene 
complexity. The technique relies on automatic stereo matching 
that, in the absence of geometric data, requires a large number 
of closely spaced images to succeed. The required computations 
may need high processing power and large memory. Object 
occlusions and discontinuities will also affect the output. The 
ability to move freely into the scene and viewing objects from 
any position will be limited without a geometric model. It is 
therefore unlikely that IBR will be the approach of choice for 
purposes other than visualization. For tourists where general 
visualization is enough, this approach may be adequate, but for 
historians and researchers, and of course for documentation, 
geometric details are needed. 
3. COMBINING MULTIPLE TECHNIQUES 
From the above summary of current techniques, it is obvious 
that none by itself can satisfy all the requirements of culture 
heritage applications. Given that: 
• Although laser scanning will provide all the details, it is 
usually not practical to implement as the only technique 
for every object and structure. Large buildings for example 
will require a large number of scans and produce huge 
number of points even on flat surfaces. 
• Image-based modeling alone will have difficulty with 
irregular and sculpted surfaces. Also it is important to 
develop an approach that requires only a small number of 
widely separated views and at the same time offers a high 
level of automation and be able to deal with occluded and 
unmarked surfaces. 
Figure 1: Combined image-based and laser scanning methods. 
(A) The Abbey of Pomposa. (B) Dazu, China 
Therefore, combining techniques where the basic shapes are 
determined by image-based methods and fine details by laser 
scanning is the logical solution. This is best described by an 
example. In figure 1, most of the structure is easy to model by 
images taken with a 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 enlarged 
sections shown. Those parts are best acquired by a laser scanner 
and added to the global model created from the images. This 
involves matching and integrating local detailed points obtained 
by the scanner to the global model. We measure several 
features, usually six, 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. The transformation parameters are then used to register 
the two coordinate systems of the two data sets. The details of 
each approach and the combined approach will be described 
next.