Agriculture Reference
In-Depth Information
70
60
50
40
30
20
f
= 8 mm
10
0
0
10
20
30
40 50
Baseline (cm)
60
70
80
90
100
FIGURE 12.3
FF-based selection of fundamental parameters for stereo cameras.
be used to generate alternative graphics when the baseline is fixed due to a physical
constriction in the vehicle, or the expected ranges are well known and do not vary
excessively.
10
⋅
B
(
mm
)
⋅
f
(
mm
)
FF
=
(12.1)
R
(
mm
)
The best combination of camera parameters does not guarantee a 3-D perception
free of errors; as a matter of fact, poor calibrations, lack of texture in the targeted
objects, or the use of disadvantageous images typically result in miscorrelated pix-
els, which in turn lead to wrong 3-D positions. The nature of each application will
determine the degree of tolerance to noisy mismatches, but regardless of the par-
ticular objective pursued with the stereo system, filtering routines must always be
incorporated in 3-D perception systems. Different approaches can be followed to
reduce the impact of noise in ground-based stereo images (Rovira-Más et al. 2010;
Rovira-Más 2011), but two elemental steps have resulted effective: the execution of
mismatching filters typically embedded in correlation algorithms, and the limitation
of the calculated coordinates to a likely 3-D space around the vehicle (
validity box
),
discarding those positions that are obviously impossible such as underground per-
ception or objects too high above the vehicle (often caused by miscorrelated clouds).
In general, though, agricultural scenes and off-road environments are rich in texture
and illumination, and therefore wrongly placed objects are normally insignificant in
the complete visualization of a vehicle's surroundings. Nevertheless, noisy patches in
the disparity images of correlated stereo pairs must be avoided whenever possible to
avoid the wrong actuation of automated vehicles, especially if stereo is performing
obstacle avoidance tasks.
Local perception is necessarily tied to vehicle-fixed coordinate systems; without
a well-defined coordinate system, sensing of the environment is meaningless. The
