Information Technology Reference
In-Depth Information
o
I
o
in (19.5). To simplify notation, we write
ξ
ξ
ξ
where
is defined by replacing
with
J
,
J
+
,
J
I
and
J
+
I
instead of
J
(
q
o
),
J
+
(
q
o
),
J
I
(
q
o
) and
J
+
(
q
o
), respectively, in this
I
section.
We next obtain an estimate of
ξ
ξ
I
from
. Substituting (19.18) into (19.15),
we have
ˆ
o
I
o
=
JJ
+
I
ξ
(
ξ
I
−
ξ
)+
ξ
.
(19.19)
ˆ
ξ
The vector
implies an estimate of all the image features from correctly extracted
image features, if
I
is a set of correctly extracted image features. We here define
ˆ
16
=
ˆ
1
2
ˆ
ˆ
ξ
,
(19.20)
ξ
ξ
...
ξ
ˆ
2
ξ
i
∈
R
,
∈{
,
,...,
}.
i
1
2
16
(19.21)
19.5.2
Image Feature Selection
Similar to (19.19), we can estimate
ξ
I
from itself by
˜
o
I
o
I
.
ξ
I
=
J
I
J
+
ξ
I
−
ξ
ξ
(
)+
(19.22)
I
We h e r e s e t
˜
=
˜
σ
1
σ
2
σ
|
I
|
˜
˜
ξ
ξ
...
ξ
ξ
I
,
(19.23)
˜
2
ξ
σ
j
∈
R
,
σ
j
∈
I
,
j
∈{
1
,
2
,...,|
I
|}.
(19.24)
˜
.If
˜
The vector
ξ
σ
j
implies an estimate of
ξ
σ
j
from
ξ
I
ξ
σ
j
is far from
ξ
σ
j
,then
we may determine that the
σ
j
-th image feature is not extracted correctly due to
occlusion. We therefore obtain a better candidate for a set of correctly extracted
image features by the following procedure:
˜
= arg max
σ
j
∈
I
ξ
σ
j
−
ξ
σ
j
,
(19.25)
I
←
I
−.
(19.26)
˜
If
ξ
σ
j
−
ξ
σ
j
is within a given tolerance for all
σ
j
∈
I
,then
I
implies a set of
correctly extracted image features
19.6
Visual Servo Control with Occlusion Handling
This section shows a visual servo control algorithm implemented in our control
system.
Several notations are defined to describe details of the implemented algorithm.
The vector of the generalized coordinates at time
k
is denoted by
q
(
k
), and the vector
Search WWH ::
Custom Search