of E g is conducted in a hill climbing method.
Table 1. Line segment extraction is conducted using the Z-
Image without ground points. From Table 1, it is obvious
that line extraction of #0~#4 in both views has rather high
reliability, while #6,#7 in viewl and #4,#6,#8,#9 in view2
are wrong extractions. In Table 1, we can find unreliable
line segments always have larger values on o n or g d
than reliable ones, whereas there’s no obvious difference
in regression variance between good and poor extractions.
This means o n and o d has better performance in
characterizing the reliability of line parameter estimation.
Figure 3. Parameters that are involved in multiple views’
registration, v., v and v k stand for the viewpoints of
view #/, view #j and view #k. d ij is the distances between
V i and v . p is the angles from vy k to VVj. e u
is the angle from X-axis of view #i’s coordinate system to
V.V , a. is the angle from the X-axis of view #i’s
coordinate system to view #j’s.
4 EXPERIMENT RESULT AND DISCUSSION
In this section, we will discuss two sets of experiment
results. First, we will tabulate some of the results obtained
in registering two views of laser range image using Z-
Image to testify the méthodologie framework for pair-wise
registration. Secondly, we will discuss an experiment
where 29 views of overlapping laser range images are
registered to construct a spatial model for the buildings in
IIS, Univ. of Tokyo. The experiment aims at examine
robustness of pair-wise registration, and testifying the
efficiency of the method in solving the error accumulation
problem. Laser range images used in the experiment are
measured with the range of 360° in horizontal rotation
and 60° in vertical rotation, with the resolution of 2 sample
per horizontal degree and 1 sample per vertical degree.
Laser range measurement has an accuracy of ±5cm. In
order to examine the registration accuracy, each view
point has been measured manually using GPS. It has an
accuracy of ±10cm.
4.1 Experiment of registering two views of laser range
image
Figure 4 shows two views of laser range image, and the
Z-lmages generated with a resolution of 0.25m/pixel.
Reliability evaluation of line segments in view 1 is listed in
(a) view 1
Sf.- a'v .3
1 / •- - .
w
(b) view 2
Figure 4 Two views of laser range image and their Z-
Images. Left figures are laser range images shown in 3D
space, while right figures are the Z-lmages generated.
Line segments extracted are shown in left figures.
Characters stand for the sequential numbers.
In coarse matching, given a threshold (=2°) for the
maximum difference in intersection angle, a threshold
(=2pixel) for the maximum difference in distance, there are
totally 48 candidates obtained. All candidates are tried to
match the views. The one composed by Line segment
pairs (#2,#1) of viewl and (#0,#3) of view2 which yields
minimum distance measure is selected as the optimum.
To understand the distance measure defined in this
research, Table 2 and Table 3 list the results of some
major items when using the above candidate to match two
1-1-6
