Stereo Head
|
Foveated Vision
|
Infrared/Thermal
|
 |
 |
 |
| This
sensor
uses two
cameras separated by a known baseline and with calibrated optics. In
this case
two baselines are available, a short and a long to be used depending on
whether
the regions of interest are close or far, respectively. The algorithm
used to
generate a 3D representation applies
epipolar
search to match pixel properties, and an inverse perspective projection
to determine
coordinates in space. The 1280 by 960 frames are available at a rate of
15 fps
and are transferred via firewire interface and the algorithm runs on
the PC. |
This
sensor is composed of two similar cameras, one sitting atop the other,
fitted
with different optics. Each of these has different viewing angles and
focal
distances, one has wide angle and the other, the
foveal, has a narrow view and thus provides greater detail.
This sensor is used to test vision based attention mechanisms and thus
is mounted on a pan-tilt unit, controlled so that the image projected
onto
the fovea camera always contains the supposed region of interest, and
then
it is analyzed in detail. The supposed region of interest on the other
hand
is always determined from the image captured by the camera fitted with
the wide
angle lens, thus reducing the computational burden. The two 1032 by 776
pixel
frame rate is 30 fps and are transferred via firewire interface. |
The
infrared camera is used to allow for safer
driving during periods of the day when
lighting is insufficient for standard cameras. This sensor captures
electromagnetic waves in the infrared region and with it one can more
easily detect a
person or hot blooded pet. The 320 by 240 frame rate is 9 fps and has
been
transferred over Ethernet via real time streaming protocol in MP4
format. |
2D
|
3D
|
 |
 |
| A small laser scanner, which uses the principle of a
continuous, frequency modulated, wave is located at the front of the
vehicle. At a low level, it allows for immediate detection of obstacles
in front of the vehicle that are sectioned by the scanning plane, at a
higher level it is possible to use this sensor to determine the
presence of, to follow and predict the motion of objects which
meanwhile are classified. Scanning is performed by a rotating mirror
and the rate is 40 Hz and the field of view is 270º at a distance up to
30 m but depends greatly on the incident solar radiation. |
This is a custom sensor, based on a commercial 2D range finder,
which uses the time-of-flight principle. The 2D scanner can be
configured to provide different resolutions and fields of view. Other
than distance to the object, it can also provide the strength of the
received signal, which for a given strength of the laser pulse is a
rough indication of color. The team generally sets the field of view to
100º, with 0.25º resolution. The scanner is then rotated about an
external axis to form a 3D conical field of view. With the said
configuration, the 2D scanner is providing measurements at 18 Hz and
the full rotation, which can be considered a 3D measurement rate, is
usually set to 1.5 Hz and nominal distances up to 30 m. Though data
rate is relatively low for a final application, this sensor already
provides very interesting data which can be processed to determine
geometrical properties of the environment. |
Pedal Sensors
|
Egomotion
|
 |
  |
| The
actuation of the
pedals is sensed by the combined use of 3D low frequency MEMS
accelerometers
accelerometers .
Each pedal has an accelerometer fixed to it and there is an additional
accelerometer fixed to the car’s structure, just beneath the pedals,
thus enabling one to determine the relative angle by use of the
acceleration of gravity. This set has been designed by the team and
each board has dimensions of only 12×21.2×5.5 mm. The sensor set has
already been tested at an acquisition rate of 10 Hz, but further
developments will increase the rate to 50 Hz. |
This sensor is actually the combined use of
a MEMS IMU and a GPS receiver .
The first sensor is comprised of three orthogonally fixed
accelerometers, three magnetometers and three angular rate sensors,
allowing for dead-reckoning during absence of adequate GPS signals.
Good GPS reception allows for the correction of the accelerometer bias
which otherwise tends to induce large drift into the estimated
positions. Though determining the exact position of the vehicle is not
important in our case, egomotion compensation is necessary for correct
perception of the environment. Whilst the installed GPS receiver alone
can only provide position and speed over ground at a rate of 1 Hz, the
fused sensor set can present corrected measurements at 100 Hz. |
|
