Page 27 - Autonomous Mobile Robots

Visual Guidance for Autonomous Vehicles 11

available in 2D and 3D versions but the principles are essentially similar: a
laser beam is scanned within a certain region; if it reflects back to the sensor
off an obstacle, the time-of-flight (TOF) is measured.

2D scanning. The majority of devices used on mobile robots scan (pan)
◦
◦
through 180 in about 13 msec at an angular resolution of 1 . Higher resolution
◦
is obtained by slowing the scan, so at 0.25 resolution, the scan will take about
52 msec. The sensor thus measures both range and bearing {r, θ} of obstacles
in the half plane in front of it. On a moving vehicle the device can be inclined
at an angle to the direction of travel so that the plane sweeps out a volume as
the vehicle moves. It is common to use two devices: one pointing ahead to
detect obstacles at a distance (max. range ∼80 m); and one inclined downward
to gather 3D data from the road, kerb, and nearby obstacles. Such devices are
popular because they work in most conditions and the information is easy to
process. The data is relatively sparse over a wide area and so is suitable for
applications such as localization and mapping (Section 1.4.2). A complication,
in off-road applications, is caused by pitching of the vehicle on rough terrain:
this creates spurious data points as the sensor plane intersects the ground plane.
Outdoor feature extraction is still regarded as a very difficult task with 2D ladar
as the scan data does not have sufficient resolution, field-of-view (FOV), and
data rates [10].
3D scanning. To measure 3D data, the beam must be steered though
an additional axis (tilt) to capture spherical coordinates {r, θ, φ: range, pan,
tilt}. There are many variations on how this can be achieved as an opto-
electromechanical system: rotating prisms, polygonal mirrors, or galvono-
metric scanners are common. Another consideration is the order of scan; one
option is to scan vertically and after each scan to increment the pan angle
to the next vertical column. As commercial 3D systems are very expensive,
many researchers augment commercial 2D devices with an extra axis, either by
deflecting the beam with an external mirror or by rotating the complete sensor
housing [11].
It is clear that whatever be the scanning method, it will take a protracted
length of time to acquire a dense 3D point cloud. High-resolution scans used
in construction and surveying can take between 20 and 90 min to complete a
single frame, compared to the 10 Hz required for a real-time navigation system
[12]. There is an inevitable compromise to be made between resolution and
frame rate with scanning devices. The next generation of ladars will incorporate
flash technology, in which a complete frame is acquired simultaneously on a
focal plane array (FPA). This requires that individual sensing elements on the
array incorporate timing circuitry. The current limitation of FLASH/FPA is the
number of pixels in the array, which means that the FOV is still small, but this
can be improved by panning and tilting of the sensor between subframes, and
then creating a composite image.

© 2006 by Taylor & Francis Group, LLC

FRANKL: “dk6033_c001” — 2006/3/31 — 16:42 — page 11 — #11