1-1-1
A ROBUST METHOD FOR REGISTERING 2.5D LASER RANGE IMAGES OF
URBAN OBJECTS
Huijing ZHAO , Ryosuke SHIBASAKI
Center for Spatial Information Science, Univ. of Tokyo
Email: chou@skl.iis.u-tokyo.ac.jp
Commission II, Working Group I
KEYWORDS : laser range image, registration, urban 3D object
ABSTRACT
In this paper, a method for registering 2.5D laser range image of urban object is proposed. The problem is solved in two
levels, pair-wise registration to find the local matching between successive views and multiple views’ registration to
obtain global matching. A kind of Z-lmage is introduced. Pair-wise registration using Z-lmage is developed, where 4
transformation parameters, a horizontal rotation angle and three translation parameters, are tracked. In multiple views’
registration, accumulation of estimation error in pair-wise registration is solved as a least square minimization problem.
Through an outdoor experiment for reconstructing 3D model of the buildings at IIS, University of Tokyo, it is demonstrated
that the registration method has robustness and accuracy in the sense of automation.
1 INTRODUCTION
As the development of mobile navigation system, ITS
(intelligent transportation system) and other 3D GIS
applications, complete and accurate urban spatial
database is becoming strongly demanded. Reconstructing
3D urban objects using ground-based acquisition
techniques are attracting more and more attentions.
Several researches using CCD camera have
demonstrated that 3D information can be extracted using
motion (S.Ozawa et.al. 1998) and stereo (M.Okutomi and
T.Kanade 1993) vision techniques. However, the
difficulties in reliable stereo matching, distortion from
limited resolution and unstable geometry of CCD cameras
are the major obstacles to the accuracy of the method. On
the other hand, recent years have witnessed a substantial
development of range collecting techniques. Researches
using range sensors can be founded in C.Thorpe
ef.a/.1998 in a mobile navigation system, B.Kamgar-Parsi
et.al. 1991 to obtain a map of ocean floor, Y.Chen and
G.Medioni 1992, G.Champleboux et.al. 1992, H.Shum
ei.a/.1994, K.Ng et.al. 1998 to construct 3D models for
indoor objects. Efficiency and accuracy of range sensors
serving for spatial object’s reconstruction has been
demonstrated. In recent years, with the development of
laser technique, range finder using eye safe laser, with
long range measurement and high speed is being used
for urban purpose. One of the major topics in
reconstructing spatial object using range sensor is the
registration of range images of different configuration. This
paper contributes to a practical method for the registration
of 2.5D laser range images of urban objects.
To obtain laser range images of urban object, a laser
range finder is installed on a vehicle or directly on the
ground. Once a view of laser range image is measured,
the sensor is moved to the next viewpoint. A three
dimensional registration method using planar faces has
been investigated in our previous work (Zhao and
Shibasaki, 1997). Robustness of method is insufficient in
urban outdoor environment, since enough number of
planar faces is difficult to be extracted for the following
reasons. First, laser range measure in urban area has
high uncertainty, which might be caused by trees, window
glasses, passing cars and pedestrians. Secondly, objects
other than interest, such as window frame, electric cable,
