Page 25 - Autonomous Mobile Robots

Visual Guidance for Autonomous Vehicles 9

in military scenarios, or there may be too many conflicting sources in a civilian
setting. At this point we also highlight a distinction between the terms “act-
ive vision” and “active sensors.” Active vision refers to techniques in which
(passive) cameras are moved so that they can fixate on particular features [4].
These have applications in robot localization, terrain mapping, and driving in
cluttered environments.

1.2.1.1 Passive imaging
From the application and performance standpoint, our primary concern
is procuring hardware that will acquire good quality data for input to
guidance algorithms; so we now highlight some important considerations when
specifying a camera for passive imaging in outdoor environments.
The image sensor (CCD or CMOS). CMOS technology offers certain
advantages over the more familiar CCDs in that it allows direct access to indi-
vidual blocks of pixels much as would be done in reading computer memory.
This enables instantaneous viewing of regions of interest (ROI) without the
integration time, clocking, and shift registers of standard CCD sensors. A key
advantage of CMOS is that additional circuitry can be built into the silicon
which leads to improved functionality and performance: direct digital out-
put, reduced blooming, increased dynamic range, and so on. Dynamic range
becomes important when viewing outdoor scenes with varying illumination:
for example, mixed scenes of open ground and shadow.
Color or monochrome. Monochrome (B&W) cameras are widely used
in lane-following systems but color systems are often needed in off-road
(or country track) environments where there is poor contrast in detecting travers-
able terrain. Once we have captured a color image there are different methods
of representing the RGB components: for example, the RGB values can be
converted into hue, saturation, and intensity (HSI) [5]. The hue component of
a surface is effectively invariant to illumination levels which can be important
when segmenting images with areas of shadow [6,7].
Infrared (IR). Figure 1.1 shows some views from our semi-urban scene test
circuit captured with an IR camera. The hot road surface is quite distinct as
are metallic features such as manhole covers and lampposts. Trees similarly
contrast well against the sky but in open country after rainfall, different types
of vegetation and ground surfaces exhibit poor contrast. The camera works on
a different transducer principle from the photosensors in CCD or CMOS chips.
Radiation from hot bodies is projected onto elements in an array that heat up,
and this temperature change is converted into an electrical signal. At present,
compared to visible light cameras, the resolution is reduced (e.g., 320 × 240
pixels) and the response is naturally slower. There are other problems to contend
with, such as calibration and drift of the sensor. IR cameras are expensive

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c001” — 2006/3/31 — 16:42 — page9—#9