Agriculture Reference
In-Depth Information
man-set structures. These structures typically correspond to one of the following
situations: (1) swaths of cut cereal or forage plants waiting to be picked; (2) vehicle-
wide lanes delimited by rows of orchard trees, as 5-m separated lines of trees in an
orange grove; (3) rows of bulk crops equally spaced such as 30-in spaced corn rows;
and (4) man-made supporting structures such as wooden posts to hold hops or trellis
to guide apple trees. Greenhouses and nurseries still represent a more heavily struc-
tured environment although perceiving in controlled atmospheres is less problematic
than moving outdoors where light is constantly changing and unexpected obstacles
are prone to appear. This diversity of environments can be simplified if field struc-
tures are labeled as
traversable
and
non-traversable
(Rovira-Más 2009). Obviously,
this classification based on
traversability
not only depends on the structures them-
selves but also on the characteristics of the vehicle used. The same structure, say a
1.5-m-high vineyard, will not be traversable by a small tractor but will necessarily be
passable by the harvester collecting the grapes. In addition to the fundamental task
of avoiding crashes, local perception is very useful for the execution of those fine
adjustments that place the vehicle within the safest course. Very often, trajectories
have been determined by global positioning devices, but the reality encountered by
the vehicle at a given time might have changed from the route planned beforehand,
and only local perception can provide the updated scene with the detail and accuracy
that secure operations require.
The technology devised to compose the perception engine of a vehicle can be
generally divided into
visual
and
nonvisual
, and each system or subsystem will nor-
mally encompass sensors and processing units. The four-core subsystem architecture
proposed by Rovira-Más (2010) for intelligent agricultural vehicles assigns one of
the four main subsystems of the vehicle to
sensors for local perception and vicinity
monitoring
, given their importance in the constitution of automated farm equipment.
Visual
sensors are electronic devices that use any kind of image to capture the sur-
rounding scenario, such as CCD cameras, thermographic scanners, and stereo vision
rigs.
Nonvisual
sensors for surrounding awareness typically include ultrasonic and
laser (lidar) rangefinders.
12.2 TWO-DIMENSIONAL PERCEPTION FOR
NAVIGATION AND MONITORING
The reality around agricultural vehicles, the one to be sensed by the onboard per-
ception engine, is always 3-D. However, the intricacies of 3-D perception may be
simplified by considering partial views of the scene in a 2-D plane that holds the key
information for the commanded task. Even though some specific tasks require the
acquisition of frontal or side views of the sensed scene with respect to the vehicle,
the plane where results are usually represented is, most of the times, that coincident
with the ground and providing a top view or floor plan of the scene. For this situation,
two
nonvisual sensors
have been successfully implemented in automated vehicles:
ranging sonars and lidar (light detection and ranging) heads. Although especially
effective indoors, the potential of ultrasonic devices as safety components of farm
vehicles has been demonstrated (Guo et al. 2002). Nevertheless, their use in outdoor
applications is somewhat limited because of the narrowness of the cone angle and
