e
je
nt
pe
ISPRS Commission III, Vol.34, Part 3A , Photogrammetric Computer Vision“, Graz, 2002
Gyro sensors and/or other navigation systems. Range data
obtained in different viewpoints are registered and integrated,
and a completed model of urban environment is reconstructed.
There are several drawbacks of stationary systems. First, in data
acquisition, successive range views have to keep a degree of
overlay, so that location and direction of viewpoints can be
traced (or refined) by registering range data. Planning for
viewpoints and directions in data acquisition becomes difficult
when measuring large and complicated scene, since a balance
between the degree of overlay and the number of viewpoints has
to be decided according to both target objects and registration
method. Secondly, there is still no registration method that
could succeed in automatically registering range data of all
kinds. When the number of range views increases, registration
while keeping necessary accuracy becomes difficult. Updating
stationary systems to moving platform ones (called vehicle-
borne system) for reconstructing 3D model of large real scene is
very important.
Konno et al.[11] developed a sensor system by mounting three
single-row laser range scanners on a vehicle with a high
accurate navigation system. In this research, we propose a
prototype of reconstructing the urban outdoor environment from
the output of the vehicle-borne sensor system.
2. OUTLINE OF THE RESEARCH
In this chapter, we first briefly introduce the sensor system, and
its data output, then state the problems, and finally outline the
concept of the research.
2.1 Sensor system
In the sensor system developed by Konno et al.[11], three laser
range scanners (LD-As) are mounted on a measure vehicle
(GeoMaster), which has been equipped with a
GPS/INS/Odometer based navigation unit (see Figure 1). LD-A,
produced by IBEO Lasertechnik, is a single-row laser range
scanner. It has a profiling rate of 10Hz, a maximal range
distance of 100 meters, and a measurement error of 3cm. In
each profiling (scan line), 1200 range distances are measured
equally in 300 degrees, where 60 degrees of blind area exists
due to hardware configuration. Reason for using three LD-As is
to reduce occlusions by trees and other obstacles. As the vehicle
moves ahead, LD-As keep profiling the surroundings on three
different vertical planes (cross-section). Meanwhile, the
navigation unit outputs the vehicle's location coordinates (x, y,
Z) and orientation angles ( , , ) in world coordinate system
at the moment of each laser scanning, so that all range distances
(range points) in LD-A’s local coordinate system at the moment
of measurement can be geo-referenced to the world coordinate
system. Range points of different LD-As are recorded in
different output files (views) in the order of measurement
sequence.
2.2 Problem statement
This research focus on generating a surface representation of
urban out-door environment using the range outputs of the
above sensor system. Surface reconstruction from dense range
data has been studied for decades. Soucy and Laurendeau, [16]
and Turk and Levoy, [18] exploited the connectivity of
structured range points. Hoppe, ef al [8] proposed a method of
generating an implicit surface from unorganized points using
volumetric representation and marching cube algorithm. Curless
and Levoy [2], Wheeler et al.[19] are the hybrids of the above
two methods, where implicit surface method is exploited to
integrate structured range views. Most of the researches assume
that all the range points are on or near an implicit surface, and
they are clear or only have a systematic error. However this is
not always true in urban out-door environment.
A ow
AY XN CN fili
/ly A Laser profiling
/
\
| Ve ME
| "d TA GPS antenna
ue
OW ~
Moving ahead
(a) Sensors’ alignment (b) Pictures of the vehicle
Figure 1. Sensor System
Except the irregular points that reflected by passing cars and
pedestrians, window glasses and trees are two major difficulties
in the modelling of urban out-door environment by laser
scanning. Some window glasses are penetrative to laser beam,
subsequently yields range measurement of unknown indoor
objects, which are beyond our interest. While some window
glasses give mirror reflection, so that yields black holes (no
data) on building surface. Trees are of complicated shape and
plenty of occlusions. Laser scanning of a tree yields a cloud of
scatter points, which are not implying a surface but a volume. It
is obvious that modelling a surface-structured object, such as
building and road surfaces, should be conducted in a different
level with that of a volume-structured object, such as trees.
However trees are always near to and block the measurement of
building, borderline between them is always confusing.
2.3 Outline of the research
Reconstructing a surface representation of urban out-door
objects is conducted in two procedures. First, range points are
classified into six groups, ie. the measurement of vertical
(building) surface, false window area, road surface, other kinds
of surface, tree and unknown objects. False window area
(briefly referred to “window area” in the following sections)
implies the penetrative or mirror-like window area, of which
range values are corrected and interpolated using the
measurement on surrounding vertical building surfaces. In this
research, window area that does not yield erroneous or false
reflections is not our research interest. They are regarded as a
part of vertical building surface. Secondly, volumetric
representation and marching cube algorithm are exploited since
it is easy in generating a model of desired level of detail, which
is required in many 3D GIS applications. The scheme for
generating volumetric representation and the algorithm for
computing signed distance are defined differently, where iso-
surfaces are computed for surface-structured objects, i.e.
A - 413