Information Technology Reference
In-Depth Information
(a)
(b)
Fig. 18.1
(a) fixed frame
W
, camera frame
C
, and relative coordinates (ξ
1
,
ξ
2
,
ξ
3
) and
(ρ
,
φ
,
β); and (b) fixed frame
W
, camera frame
C
and relative feature coordinates
The localization method thus presented is based on punctual estimation and does
not take into account the known robot motion model. To better exploit the available
information, an EKF is adopted, given for granted the association between current
and desired point features. However, should feature outliers occur in the process,
more robust filtering should be used in place of simple EKF, such as
e.g.
those
described in [32] or [22].
The EKF localization state is
S
=[
c
z
n
]
T
. Estimated state
initial guess is computed using the previously presented least mean squares static
localization. The initial model covariance matrix
P
0
is block diagonal.
P
0
∈
R
c
x
1
,...,
c
x
n
,
ξ
1
,
ξ
2
,
ξ
3
,
3
×
3
,
Search WWH ::
Custom Search