Agriculture Reference
In-Depth Information
Top position
Front position
Side position
Vehicle setup
Vehicle setup
Vehicle setup
Sample image
Sample image
Sample image
(a)
(b)
(c)
FIGURE 12.2
Sensor position and image type for 2D visual perception of a vehicle's vicinity.
Figure 12.2a sets the camera above the crop in such a way that the image plane is
approximately parallel to the ground. The advantage of this position is the lack of
perspective and vanishing points that results in a simple correspondence between
pixels and objects. However, its physical embodiment in the vehicle may need the
assemblage of bulky structures, which are not always welcomed by drivers who
often need to maneuver within narrow lanes. The
front position
of Figure 12.2b,
either mounting the camera at the nose of the tractor or centered in the cabin, yields
the preferred images for crop-following navigation. The calibration of the camera is
a determinant step to obtain accurate positions of the vehicle, and as a result, stable
navigation commands. The
lateral position
of Figure 12.2c offers a view of the inner
side of the row. This setting maintains the image plane parallel to the rows and
perpendicular to the traveling direction; therefore, image and target are parallel and
easy to correlate. Difficulties might appear, though, when rows are narrow and the
vehicle moves too close to foliage; in such a case, leaves may block the field of view
of the camera, and what is worse, branches could damage the camera or scratch its
lens. As shown in Figure 12.2, each position results in a distinct type of images; fron-
tal images such as that shown in Figure 12.2b are usually processed and discarded
once the navigation parameters have been calculated, but top and lateral images are
very often processed and integrated into a crop management map as those depicted
in Figure 12.8. Crop status and management maps may be alternatively deduced
from remote sensing images. This option, however, can only provide data from a
zenithal view, and the resolution and updating cycle of remotely sensed images can
never be compared to that of ground-borne imagery. As a result, perception engines
will preferably be affixed to land vehicles, and their performance will ideally be in
real time.
