X. Li et al.: An Interactive Visual Navigation Method Using a Hand-drawn-Route-Map in an Unknown Dynamic Environment
313
that is, there is a mapping relation
3
between L
and
Landmarks(.).
P
includes the jumping-off point S of
route, the terminal D and the route, along which mobile
robot doesn't really and entirely run. This path is sto-
chastic, flexible, and imprecise, which only guides mo-
bile robot to run according to its approximate tendency,
this is because there is no precise size and scale about
the real environment.
represents the initial approxi-
mate pose of mobile robot. There is a relaxing mapping
relation between it and the real one. In this paper, we set
out from the tendency of sketched route. The original
route is divided into several segments. every segment is
composed of two key boot points and curve. In order to
control mobile robot well and avoid accumulative errors
for frequent turns, we adopt the way of linear motion
between two key boot points. How to pick up key boot
points relies on the principle of lesser deviation. That is,
it not only reflects the original tendency of robot motion,
but only reduce the turn frequency of robot, so that we
can try to reduce the number of the key boot points. How
to pick up key boot points is shown in Figure 1, where
curve represents the original route sketched, the mini-
mum circle represents the process of discretization, the
bigger circle represents the candidate of key boot points,
and the biggest circle represents the key boot points
picked up. The key steps are listed as follows:
1) Pick up the candidate of key boot points. Start from
the jumping-off point, detect in turn every discrete point,
and set the threshold M of angle change, and set the
minimum and maximum distance threshold
min
D
and
max
D
. The particular flowchart is shown in Figure 2.
According to experience, we set M
20
o
=
,
min
1Dcm
,
and
max
8Dcm=
.
Figure 1. Draw a path and pick up its key boot points.
2) Pick up the key boot points. The key flowchart of
picking up the key boot points is shown in Figure 3.
i
T
represents the
i key boot point. T is the number of the
candidates of key boot points.
b
i and
e
i represent re-
spectively the beginning and ending detection points. H
3
The relative position relation among every landmarks in sketched
map is consistent to that in real environment.
represents the maximum number of candidates between
two key boot points.
be
ii
L represents the line segment
between the
b
i candidate and the
e
i one.
Max(Dist(p,
be
ii
L )) represents the maximum distance
among all discrete points p on original curve between
b
i and
e
i to the line
be
ii
L . D is the pixel distance be-
tween any two neighboring discrete points.
repre-
sents the threshold to decide whether to be a key boot
point. Here H=4 and
1.0
.
Figure 2. The flowchart of selecting some candidate points.
3. Rough localization of robot based on vision
In order to roughly localize robot by applying the vis-
ual information in the process of navigation, here two
necessary hypothesises are proposed as follows: 1) the
initial approximate distance of shooting picture is known
and there are abundant of environmental features. 2) The
angles between these landmarks and floor are kept no
greater change when mobile robot is running in the en-
vironment.
A. Acquisition of approximate distance
According to the model of pinhole, any one point in
space is transformed from camera coordinates to image
2
2