The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
836
2. BACKGROUND
The last fifty years have witnessed a rapid evolution of the data
used in Geomatics applications. This evolution was inspired by
the development of (a) fast digital computers, (b) new high
resolution satellites, (c) the development of integrated,
navigation sensors/systems, and (d) general technological
advancements. As a result, numerous types of sensory data have
become available and are classified according to:
1- Color information (e.g. color and pan images).
2- Spectral bands (multi/hyper spectral).
3- Recording media (analogue and digital).
4- Sensor Dimensionality (frame, line, and panoramic).
5- Platform/sensor location (satellite, aerial, terrestrial).
6- Physical nature of the collected data (optical, Non-optical
data like LiDAR (Light Detection And Ranging)).
The topic of multi-sensor data fusion has received over the
years a lot of attention by various parties, see for example Hall
and Llinas, 1997. As noted by (Hong, 1999), these techniques
have the overall objectives of achieving numerous benefits,
such as:
• Robust operational performance
• Extended spatial/temporal coverage
• Reduced ambiguity
• Increased confidence
• Improved detection performance
• Enhanced resolution (spatial/temporal)
• Increased dimensionality
Many researchers have proposed different fusion schemes for
integrating different sensory data. This fusion process can be
broadly classified into optical-to-optical, optical-to-nonoptical,
and nonoptical-to-nonoptical.
To name few works on data fusion, Lee et al., 2002, presented a
study on performing aerial triangulation using frame, push
broom, and whisky broom cameras thus applying the concept of
multi-sensor aerial triangulation (MSAT). Lee and Choi, 2004,
merged the terrestrial images and terrestrial lidar cloud points to
reconstruct 3D GIS building models with more realistic
geometric and radiometric properties. Iavaronea and Vagnersb,
2004, generated complete and accurate solid model for Toronto
city hall in Toronto by combining aerial and tripod-mounted
LiDAR data. In fusion literature, there is almost no attempt
towards the fusion of airborne and land-based mobile mapping
systems imagery/navigation data.
3. POTENTIAL OF THE MMS DATA FUSION
Mobile mapping concept has yielded to a breakthrough to the
Geomatics applications and opened new avenues for both
research and industry visions. Mobile mapping systems (MMS)
can be installed on different platforms. Regardless the platform,
MMS operate by attaching the navigation sensor to mapping
sensors in a rigid connection. The navigation sensors provide
sufficient information for estimating the position and the
orientation of the mapping sensor with respect to the world
coordinate frame (El-Sheimy, 1996). While, mapping sensors
are used to locate object features with respect to their local
coordinate system. By combining the navigation and mapping
sensor data, the mapping process can be done in an efficient,
economic manner and almost everywhere. The mapping sensors
may include, but not limited to, frame/line/panoramic digital
camera(s), laser scanners, and different Radar sensors. Thus,
MMS can, in principle, provide multiple mapping data sources
collected at the same time.
As stated earlier that images captured by (LMMS) and (AMMS)
are different in the sense of direction, scale, coverage,
hidden/visible features. In LMMS, the direction of the camera
axis is almost horizontal. While in AMMS, the camera optical
axis is usually looking down. Additionally, in LMMS the
features of interest are usually traced in images captured at 20-
50m away from the camera while, in AMMS the camera-object
distance ranges from 300m to few several kilometres above the
earth’s surface. Therefore, one aerial image may cover the
operational range of several hundreds of LMMS image sets. In
AMMS, the images are earth-top views which contain mainly
building roofs and road planes. LMMS’ images are
complementary to AMMS’ images as they include the building
sides and more road details.
With data iusion of images captured by different kinds of
sensors (multi-modality), 2D images coming from the same
vehicle, aerial images, an important issue will be the correct
geo-referencing of data, which means a good localization
(position, orientation) in a common terrestrial frame before
model creation.
Based on the above discussion, the integration between the data
captured by AMMS and LMMS is of high potential hence both
image/navigation data sets are complementary and can be
integrated to complete the picture about the earth’s surface. As
a photogrammetric rule of thumb, the mapping accuracy is
relatively worth in the direction of the camera axis due to
intersection geometry. Hence, the camera axis in the LMMS
and AMMS are almost perpendicular. A combined solution,
from both systems observations improves the 3D quality of the
object space model. In (Hassan et al., 2006b), a
photogrammetric strategy for bridging LMMS during GPS
signal outage spans has been introduced. The framework has
been developed based on photogrammetric reconstruction using
LMMS and point features only. The new proposed integration
framework, of both data sets with different matching entities
(i.e. lines and points), can make the reconstruction of the
LMMS trajectory easier and more practical. Finally, as a major
potential of this frame work, if the aerial images involved have
no or poor georeference information-as huge image data base
still exists-the georeference information can be recovered
efficiently. The georeference of the aerial images, in this case,
will be mainly based on the 3D control lane line extracted from
LMMS (Cheng et al., 2007).
The proposed three applications for the investigated iusion
scenario serve different community. Enhancing the 3D object
space accuracy is a pure mapping application. Additionally,
aiding land-based systems in urban blocks is one of the
navigation community’s concerns. Finally, georefencing
airborne images using the proposed iusion scheme presents a
fast and efficient solution for many photogrammetric
applications.
4. MATHEMATICAL MODEL
Traditional designs of photogrammetric adjustment frameworks
have been implemented based on indirect orientation concept,