Agriculture Reference
In-Depth Information
described in the previous section and is illustrated in Figure 12.5. The latter follows
the philosophy of the procedure described above to generate global maps of point
clouds but adapted to environments represented by regular grids (Rovira-Más 2012),
as shown by the conceptual scheme of Figure 12.7. Notice that counting on accurate
estimates for both global positioning and vehicle altitude is also essential to build
global grids. In the explanatory representation of Figure 12.7, the perception engine
of a vehicle following trajectory
Γ
and measuring heading angles
ϕ
has merged two
local vehicle-fixed grids of resolution
n
x
×
n
y
and cell size
L
L
into a global grid of
resolution
n
V
n
H
and cell dimension
L
U
.
As shown in Figure 12.7, the content that fills the cells of global grids is exactly
that of the original local grids used in the building process, that is, significant per-
ception information acquired around a vehicle. Global maps can, therefore, carry an
unlimited variety of parameters according to the nature of the application developed.
Figure 12.8a, for instance, depicts a global map specifically created for navigation
where filled cells represent obstacles sensed with a stereoscopic camera, and the
intensity of the color filling the cells indicates the 3-D density. The rows formed by
the cells marked by high 3-D density correspond to grape vines detected when a
tractor endowed with a stereo system navigated along the lanes. The same vineyard
was used to generate another global grid, represented in Figure 12.8b, but this time
the content of the cells corresponds to the vegetative vigor of grape vines, estimated
by a monocular camera set to capture reflectance in the NIR, and mounted on a
tractor cabin, looking down over the vine rows (Figure 12.2a). The objective of this
global grid was to predict yield variations from spatial variations in plant vigor. In
this particular application, the result of analyzing each NIR-filtered image captured
with a monocular camera was a numerical value (percentage of vegetation cover)
rather than a local grid of potential obstacles, and consequently, the assemblage of
×
North
L
U
L
U
φ
L
L
12
11
10
9
8
7
6
5
4
3
2
1
L
L
φ
Γ
East
123456789
n
H
10
11
12
13
14
15
16
FIGURE 12.7
Construction procedure for global grids.
