The International Archives of the Phutogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
field survey prior to data acquisition and their establishment is
typically expensive and time-consuming. Also, for many
terrestrial surveys, the establishment of sufficient control points
is virtually impossible. For example, consider the control
requirements to map an entire city using close-range
photogrammetry. Finally, for some mapping sensors such as
laser scanners or push-broom CCD arrays (Line CCD camera),
it is difficult or impossible to establish control. The use of these
sensors is not practical unless direct-georeferencing is
performed.
1.2 Related Works
The first operational land-based MMS was developed by the
Centre for Mapping at the Ohio State University. Their system
(called GPSVan) integrated a code-only GPS receiver, two
digital CCD cameras, two colour video cameras and several
dead reckoning sensors (Goad, 1991; Novak, 1991). All
components were mounted on a van. The GPS provided the
position of the van and the images from the CCD cameras were
used to determine the positions of points relative to the van. The
dead reckoning sensors, which consisted of two gyroscopes and
an odometer (wheel counter) on each of the front wheels, were
primarily used to bridge GPS signal outages. These sensors
were also used to provide orientation information for the
exposure stations; however, there is little - if any - published
information that examines the orientation accuracy of the
GPSVan’s dead-reckoning sensors and the poor accuracy of
similar sensors suggests that the orientation information they
provided would have been of marginal quality at best. The two
video cameras were used solely for archival purposes and to aid
attribute identification - no relative positioning was performed
from the video imagery. Using bundle-adjustment techniques
with relative-orientation constraints, the GPSVan was able to
achieve relative object space accuracies of approximately 10cm.
Unfortunately, because only carrier-smoothed code-differential
GPS was used, absolute object-space accuracies were limited to
1-3 m.
GPSVan successfully illustrated how land-based multi-sensor
systems could improve the efficiency of GIS and mapping data
collection. However, ©Due to poor navigation sensors, the
absolute accuracy of the object space points was too poor for
many applications, especially when compared with competing
technologies. ©Because GPSVan were simple stereovision
system as its mapping sensor, automatic extraction of objects is
difficult or impossible. Therefore, further developments of land-
based mobile mapping system focused on these two points ®to
increase absolute object space accuracies by using more high
accuracy hardware or using more sophisticated processing
techniques; ©to equip more mapping sensors for enabling more
flexible data collection , better imaging configuration of objects
and easier object extraction and recognition automatically.
Based on the two points, the following expression is extended
by introducing contributions of various research groups.
Works to improve/stabilize mapping accuracy
The obvious techniques for improving absolute object space
accuracy were to use carrier-phase GPS, while the obvious
choice accuracy dead-reckoning sensor was a high precision
IMU. The use of an IMU has an additional advantage over other
types of dead-reckoning sensors in that it also provides high-
accuracy orientation information for the exposure stations.
Later land-based MMS added dual-frequency carrier-phase
differential GPS, more accurate IMUs, and more sophisticated
processing techniques for navigation sensor data. Typical
example system is the VISAT™ system (Schwarz et al., 1993).
VISAT had absolute object space accuracies that had previously
been unattainable (about 30 cm with absolute accuracy).
This 30Cm accuracy looks very excellent, but this accuracy is
not direct from navigation sensor, and it should be do “image-
based bundle adjustment” to improve georeferencing accuracy
of image like triangulation of aerial photogrammetry (Schwarz
et al., 1993). So it can be said that bundle adjustment is efficient
method to improve accuracy, but it is difficult to do it full
automatically duo to fallible image matching for tie point.
The other intelligent techniques for improving absolute object
space accuracy were to use more sophic data processing of
multiple navigation sensors like GPS, IMU, and ODO. This
works can be found in Geomobil (Talaya 2004) and
StreetMapper (Hunter 2006). The Geomobil system use
POS/LV navigation sensor and it’s corresponding software-
Pospac, so that the absolute accuracy can reach to 30 cm.. The
StreetMapper use TERRAcontrol navigation sensor system by
IGI, and the accuracy also can get the same level of Geomobil.
But, however, this level accuracy is should keep system in good
GPS signal condition.
Works to extract object as automatically as possible
For more flexible data collection, better imaging configuration
of objects and easier object extraction and recognition
automatically, the obvious methods is to equip more kind type
of sensor such as laser sensor, which can get 3D geometry
information of object surface (see Reed et al 1996, and Talaya
2004). But because there are also disadvantages of laser sensor,
the further development focus on fusion of CCD camera, which
can get color information, and laser. This kind of research can
be found in Zhao 2003. GeoMaster™ in research of Zhao 2003
was notable because of fusion of laser range finder data and
CCD line camera. Where previous land-based MMS were
simple stereovision systems employing only two forward facing
cameras, GeoMaster™ have 6 CCD line camera and 3 laser
scanners that enable more flexible data collection and better
imaging geometry. But Due to easy geometry deformation of
line camera, automatic recognition of object becomes difficult.
The mobile systems which are equipped with laser sensor and
vision sensor also can be found in following research groups,
such as Geomobil (Talaya 2004), StreetMapper (Hunter 2006)
and Waseda system (Ishikawa 2006). But, however, these
systems are just to use two kinds of data for different object, not
to fuse them for detect and extract same object for automatic
processing by overcome the drawbacks of individual sensor.
But it is obvious that fusion of multiple mapping sensors is
efficient method to detect and extract object because data fusion
processing can overcome the drawbacks of camera or laser
sensor.
1.3 Our Research Objective
Based on the above discussion, the primary objective of this
paper is to develop a serial of methods for collecting more
accuracy, higher density (called as high-definition) spatial
information data of road objects with more rapid and less costly
by fusing vehicle-based navigation data, vision data and laser
range data.
The system based on our developed methods can overcome the
drawbacks of current mobile mapping systems - namely not
high position accuracy and not high ratio of automatic mapping