ISPRS Commission III, Vol.34, Part 3A ,,Photogrammetric Computer Vision", Graz, 2002
Algorithm ComputeEdgeIntersectionPoint
Input: E(c;,c;)
Input: range points X — (x) falling into V, and V
Output: p
;
For xe X
calculate d, (x) and p; GO
1 1 p (X)
OD & , P; = — An
l0 . ges
4.2 Iso-surface of volume-structured objects
A surrounding surface is generated between filled and empty
voxels, so that enclosing the volume-structured objects. Each
voxel might have one of two states, §_=-1,1 for filled and empty
respectively. Generation of surrounding surface is also break
down into two steps, i.e. compute the state of each voxel, and
calculate the intersection points on the edges that bridging
different states.
Computation of point state: Let Bi be the number of range
points falling into the voxel of index (ij,k). After smoothing
operation of
id J+1 k+1
BASE X Du 3)
i,j,k 27 ii, jj kk
di=i—l jj=j—1 kk=k—1
state § is assigned by
=1 if n, j ST,
S i «| o (4)
1 otherwise
Computation of edge intersection point: Suppose the center
points € of V, and e of V, are of different states. We define
the intersection point p, on edge E(c, ,C;) as the middle point
of p; - (c, +e,)/2-
5. EXPERIMENTS AND DISCUSSIONS
An experiment is conducted in a real outdoor environment,
Roppongi Campus of the Univ. of Tokyo. A map of the testing
site and vehicle trajectory is shown in Figure 6. The
measurement vehicle ran at a speed of about 10km/h, and 500
scan lines were measured by each LD-A. In Figure 7, range data
from different LD-As are shown in different colours, while red,
green, blue represents the centre, right, left LD-A respectively.
Range data measured by each LD-A are showed in Figure 8 (left
column) in intensity values. From both Figure 7 and 8, it can be
found that the objects are measured simultaneously from three
different directions to reduce occlusions efficiently. All the
range points are geo-referenced into a world coordinate system
using to the navigation data. The result is shown in Figure 9. It
can be found that there are many windows on the building
surface, some, not all, of the laser beams penetrate window
glasses and the sensor got the reflections of unknown indoor
objects. In addition, it can be found that there are many
irregular points in sky, which might be caused by direct sunlight
and sensor's systematic error. Range points from different LD-
As are classified in separate procedures into vertical building
surface, road surface, other surface, window, tree and others.
Classification result is shown in Figure 8 (right column). There
are totally 319743 valid range points, where 35.4% are
discriminated as the measurement of vertical building surface,
42.61% are road surface, 3.16% are other surface, 4.57% are
window, 9.7% are trees. Other points that do not belong to any
of the above groups, which hold about 4.51%, are classified to
the group of unknown objects. Irregular points as shown in
Figure 9 are also classified to this group. They are discarded in
our system. A surface model is generated using volumetric
modelling and marching cube method as shown in Figure 10.
Voxel sizes of surface-structured and volume-structured objects
are assigned 0.3m. As we generate a surrounding surface to
enclose volume-structured objects, the surface swells with voxel
size. On the other hand, if the voxel size for surface-structured
objects is set too small, not only black holes, but also distorted
surfaces might be generated in the area of windows, rain pipes
and corners, so that manual modifications are required.
6. CONCLUSIONS AND FUTURE WORK
In this paper, a method is presented to generate surface model
of urban out-door environment using vehicle-borne laser range
scanners. Range points are classified into six groups, so that
erroneous measurement are corrected or discarded, and surface-
structured and volume-structured objects are modelled in
different strategies. Volumetric modelling and marching cube
method are exploited in this research, where an estimate for
signed distance is proposed. Through an experiment, it is
demonstrated that urban out-door environment can be
reconstructed with high automation and efficiency using our
method. Future study will be addressed on improving the
accuracy of classification, and extracting and modelling other
urban features like parking cars, telegram poles, traffic signals,
and so on.
REFERENCE
[1] ChenY., G.Medioni, Object Modelling by Registration of
Multiple Range Images, Image and Vision Computing,Vol. 10,
No. 3, Apr. pp.145-155, 1992.
[2] Curless, B., M.Levoy, A volumetric method for building
complex models from range images, Proc. of SIGGRAPH,
pp.303-312, ACM, 1996.
[3] Debevec, P.E., C.J.Taylor, J.Malik, Modeling and Rendering
Architecture from Photographs: A hybrid geometry- and image-
based approach, SIGGRAPH1996.
[4] El-Hakim, S.F., C.Brenner, G.Roth, A Multi-sensor Approach to
Creating Accurate Virtual Environments, JSPRS Journal of
Photogrammetry &Remote Sensing, vol.53, pp.379-391, 1998.
[5] Gruen, A,, TOBAGO - a semi-automated approach for the
generation of 3-D building models, ISPRS Journal of
Photogrammetry and Remote Sensing, vol.53, issue 2, pp.108-
118, 1998.
[6] Higuchi, K., M.Hebert, and K.Ikeuchi, Building 3-D Models
from Unregistered Range Images, Graphical Models and Image
Processing, vol.57, No.4, pp.315-333, 1995.
[7] Hoover, A, et.al., An Experimental Comparison of Range Image
Segmentation Algorithms, in IEEE Trans. on Pattern Analysis &
Machine Intelligence, vol. 18(7), pp. 1-17, July 1996.
[8] Hoppe, H., et al., Surface reconstruction from unorganized
points. Proc. of SIGGRAPH, 26(2):71-78, 1992.
[9] http://city.les.mit.edu//city.html, MIT City Scanning Project:
Fully Automated Model Acquisition in Urban Areas.
[10] http://vis-www.cs.umass.edu/projects/radius/radius.html, The U.
Mass RADIUS Project.
[11] Konno, T, et al, A New Approach to Mobile Mapping for
Automated Reconstruction of Urban 3D Model, proc. of Int.
Workshop on Urban Multi-Media/3D Mapping, 2000.
[12] Uehara, M., H. Zen, From Digital Map to 3D Map: Creating 3D
Map by Motion Stereo Utilizing 2D Map, IAPR Workshop on
Machine Vision Applications, pp.592-595, 2000.
A - 416