El-Hakim, Sabry
° Active 3D Sensors Ensure:
* Video/ Digital Still * Complete Coverage
Data Collection * Sensor Fusion * Correct geometry
* Positioning Devices * Correct texture
j * Existing models or geometric data
* Automatic or Operator in The Loop Ensure:
Registration * Sequential or global *Min error propagation
* Positioning Devices *Convergence A
i * Organized points Ensure:
* Unorganized points * Details
Modeling * Multi-resolution * Realism
* High rendering speed
Figure 1: The main steps for creating 3D models from real scenes.
Passive image-based methods, mostly based on photogrammetry, have been developed for specific applications such as
architecture [Debevec et al, 1996]. Those needing user interaction have matured to a level where commercial software
is now available (e.g. Photomodeler [http://photomodeler.com] and ShapeCapture [ http://www.shapequest.com]).
Approaches that automatically acquire 3D points from a sequence of images at unknown locations, using projective
reconstruction, are available [Faugeras, 1992 and Polleyfeys et al, 1999]. These methods require images taken close to
each other (short baseline) in order for the automatic correspondence to work. This makes them more noise sensitive
Active sensors (e.g. laser scanners) have the advantage of acquiring dense 3D points automatically [Beraldin et al, 1999
and Sequeira et al, 1999]. They also produce organized points suitable for automatic modeling [Soucy et al, 1996].
However, the sensors can be costly, bulky, and affected by surface reflective properties. Also a range sensor is usually
designed for a specific range, thus a sensor designed for a close range can not be used for a long range. Active methods
can be combined with passive methods to take advantage of the strength of each [El-Hakim et al, 1998].
For relatively simple objects, Structures, or environments, most existing methods will work at varying degrees of
automation, level of details, effort, cost, and accuracy. Many researchers have presented examples of those types of
model in the past 5 years. However, when it comes to complex environments the only proven methods so far are those
using positioning devices, CAD or existing models and operator in the loop. Limited examples are available, most in
urban or city modeling [Coorg et al, 1999 and Gruen et al, 1998] and as-built factory models [Chapman et al, 1998].
1.2 Summary Of The Approach
The proposed approach attempts to integrate several techniques, each of which individually may not be general enough,
into a coherent system that has a wider range of applications. It is based on the following key ideas:
1. An easy to use interactive system can be more effective than the existing automatic methods. Some aspects of 3D
modeling, such as connectivity between points, can not be done based on the information extracted from the image
alone. Full automation often imposes restrictions on the type of object being reconstructed [Bailard et all, 1999],
and may limit accuracy. Human intervention, at the critical
from a single image.
4. The system should be able to integrate data from different types of sensor. With the existing technology, it is not
possible to capture all details with one type of sensor. Laser scanning is ideal for featureless sculptured surfaces
and geometrically complex shapes but may not be practical or feasible for large structures. On the other hand
image-based methods can model large structures, however may not be able to acquire details on unmarked surfaces.
204 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000.
Wi
on
pa
an
tri:
ne
sta
2.1