1-1-7
views. Item names in both table are equvilient to the
address in section 2. Formulae defining the items are also
listed for reference.
Set 7° to the transformation obtained in coarse
matching. Strength of each matched line segment pair
under T° is shown in Table 4. A minus volume tells the
strength is invalid. The strength is measured when given a
threshold (=0.1 degree) as rotation step. Adjustment to
7° is the weighed average of the strengths in matched
line segment pairs (Table 5). Adjustment in each iteration
is decided by strength analysis if it can also minify the
distance measure between two views. Figure 5 shows the
change of distance measure in fine adjustment and the
method used to decide the adjustment in each iteration. It
is obvious that strength analysis is valid in most of the
case. Adjustment by strength analysis is efficient, as it
requires 250 milliseconds in each iteration, while testing
all the possible adjustment to find the best one requires
8150 milliseconds in average. Figure 6 shows the change
of strength in each matched line segment pair in fine
adjustment. From global viewpoint, the strength total is
reduced during fine adjustment.
Accuracy in Z-lmage matching is examined by comparing
the matching result with the ground truth of manually
measured data. The estimation error in shift vector is
examined using the distance between two view points,
while rotation angle is examined using the angle between
x-axis of two sensors’ coordinate system. Estimation error
in shift vector is about 1.43m by coarse match, and about
0.04m after fine adjustment. Estimation error in rotation
angle has not a lot change before and after fine
adjustment. After adjustment, estimation error in rotation
angle is about 0.1 degree. Figure 7 shows the change of
estimation error in fine adjustment process.
Figure 8 shows the result in matching ground points to
track the translation parameter along Z-axis. Accuracy in
matching has an accuracy of 0.2m. Error comes major
from inaccurate interpolation of ground points.
Improvement in accuracy has to be addressed in future
study. Figure 9 shows the final result.
Table 1. Reliability evaluation of the line segments
extracted in two views “o n ’’and “o d ” stand for the
reliability of parameter estimation of line normal and
orthogonal distance. “Var.” and “Length” are the
regression variance and length of the line segment. “Point
Number” is the number of image points used in the line
parameter estimation.
View 1
Line No.
<7„
(°)
(pixel)
Var.
(pixel)
Length
(pixel)
Point
Number
0
0.08
0.502
1.043
83
1362
1
0.066
0.452
1.441
169
899
2
0.126
0.635
1.05
78
561
3
0.168
0.86
1.7
101
453
4
0.93
1.82
0.882
16
202
5
0.471
2.723
0.253
25
20
6
2.221
6.23
2.094
43
19
7
4.097
9.875
1.338
47
43
Table 2. Major items in distance measurement after
coarse matching
Items
Sub-Items
Formulae &
Explanation
Result
/,(Z) 2 ID.)
EQ.7
105098.1
W Px)
EQ.8
73909.53
N total
Total point
number
12062
Matched
number EQ.13
11155
EQ.11
59397
I „(Pol)
EQ.9
14512
M* 2 i*.)
EQ.14
31188.54
/*(4 I M
EQ.16
28933
EQ.15
2256
Table 3. Conditional information between matched line relation pairs after coarse matching
Matched
Pair
(j,i)
angle(n), nf )
dis(d l j,df)
&jij
N‘
in
N‘
' out
rf
EQ.17
EQ.18
EQ.17
EQ.18
Matched point
number
Unmatched
point number
EQ.16
EQ.19
(2,0)
0.178
2.853
0.132
0.682
1672
142
10731
29024
(0,1)
1.07
2.685
0.102
0.675
1234
3
7799
19792
(3,2)
1.014
2.423
0.184
0.977
729
130
5442
13744
(1,3)
0.178
2.159
0.12
0.862
756
11
4774
12272
(5,5)
1.922
3.722
0.587
3.584
19
3
186
352