Page 35 - Autonomous Mobile Robots
P. 35
18 Autonomous Mobile Robots
d
h
w
FIGURE 1.3 The ability of a sensor to image a negative obstacle is affected by the
sensor’s height, resolution, and the size of the obstacle. It is very difficult to detect holes
until the vehicle is within 10 m.
It can be observed from Table 1.2 that:
1. The higher the driving speed, the further the camera lookahead dis-
tanceshouldbetogivesufficienttimeforevasiveaction. Forexample,
if the system computation time is 0.2 sec and the mechanical latency
is 0.5 sec, a rough guideline is that at least 50 m warning is required
when driving at 60 kph.
2. At longer lookahead distances, there are fewer obstacle pixels in the
image — we would like to see at least ten pixels to be confident
of detecting the obstacle. A narrower FOV is required so that the
obstacle can be seen.
A more difficult problem is posed by the concept of a negative obstacle: a
hole, trench, or water hazard. It is clear from simple geometry and Figure 1.3
that detection of trenches from imaging or range sensing is difficult. A trench
is detected as a discontinuity in range data or the disparity map. In effect we
only view the projection of a small section of the rear wall of the trench: that is,
the zone bounded by the rays incident with the forward and rear edges.
We conclude from Table 1.3 that with a typical camera mounting height of
2.5 m, a trench of width 1 m will not be reliably detected at a distance of 15 m,
assuming a minimum of 10 pixels are required for negative obstacle detection.
This distance is barely enough for a vehicle to drive safely at 20 kph. The
situation is improved by raising the camera; at a height of 4 m, the ditch will
be detected at a distance of 15 m. Alternatively, we can select a narrow FOV
◦
lens. For example, a stereo vision system with FOV (15 H × 10 V) is able to
◦
© 2006 by Taylor & Francis Group, LLC
FRANKL: “dk6033_c001” — 2006/3/31 — 16:42 — page 18 — #18
