B2. Istanbul 2004
naximum achievable
th CCDs whereas it
; (Blanc, 2001).
m temporal noise
ensors. FPN is time-
natch due to process
ctronics can cancel
N correction
correction
iced by thermally
e, thermal noise and
that of CMOS due
n output amplifiers
r to larger noise.
high dark currents,
imes. However, this
context of mobile
'e to adjacent pixels
bright light blooms
4OS architecture is
eover, smear that is
illumination 1s non-
5 CAMERA TO
e vision system that
isition and digital
nterpreted results or
to host computers
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV. Part B2. Istanbul 2004
through a 10-Mbit Ethernet connection. À serial RS232
interface and three optoisolated I/O lines allow synchronized
image acquisition. Consequently, the connection to the
computer hosting the mobile mapping software is simple and
without a need a frame-grabber.
Figure 10. The Ethercam
The Ethercam embeds a Linux operating system, which
supports high level programming languages and thus a wide
range of vision libraries. Keeping the computation tasks within
the camera helps its host computer to better focus upon time-
critical tasks such as the synchronisation of the GPS-RTK
position data with the captured frames.
The imaging sensor mounted on the Ethercam is a
monochromatic matrix of VGA size, i.c. 640x480 pixels. It
presents a dynamic range of six decades (120 dB) as a
consequence of the logarithmic response of pixels to light
intensity (Fossum, 1997). Such a response implies that relative
variations of light intensity (A) are perceived with constant
sensitivity over the entire range. This property is particularly
useful for the analysis of outdoor scenes where light intensity
varies substantially from high sunny conditions (100 000 Lux)
to dark shadows (10 Lux).
4.1 First tests with the Ethercam
The previous surveys of the road with the CCD sensors showed
that most automatic algorithms of centreline detection are
deceived by varying light conditions (Gilliéron et al., 2002). In
fact, using fast low-level filtration techniques, such as
binarization, reject under-exposed pixels of shadowed areas or
promote over-saturated pixels under the direct sunlight.
Due to its logarithmic response to illumination and unlike a
CCD camera, the Ethercam CMOS sensor allows the
reproduction of the outdoor scenes without any imperfections
such as blooming, smearing, or time lag (Figure 11).
Figure 11. Light and shadow on the same frame, as seen by
CCD (left) and CMOS (right) cameras.
Hence, CMOS reproduction of the reality simplifies the
methodology for extracting the pixel coordinates of the road
centreline. This can be easily achieved by the application of a
Sobel filter. Such a filter consists of two kernels that detect
horizontal (6) and vertical (7) changes in an image.
TT 0
$054. 9,0. (6)
bar d
das Oed
S, 7|-2 9 3 (7)
4H
When both filters are applied to a frame, the results can be used
to compute the magnitude and direction of the edges in the
frame. The application of the Sobel kernels results in two
images which are stored in the arrays Ghi, (veigni1)[0..(width-1)] and
GYo..cheight-1)][0.(width-11- Consquently, the magnitude of the edge
passing through the pixel x, y is given by (8), whereas the
direction is given by (9).
Mas P]]-J[{]bT +] ©
(9)
-1
Psobel BH E pn
First experiments with the camera-embedded implementation of
the Sobel filter give very promising results of contour detection
in real time, as illustrated in Figure 12.
Figure 12. Real-time contour detection using the Ethercam OS
5. CONCLUSIONS AND PERSPECTIVES
We explored two promising solutions to map the road
centreline in real-time: Internet-based RTK and a logarithmic
CMOS camera. The presented enhancements due to the CMOS-
based Ethercam relieved the workstation hosting the mobile
mapping software from image grabbing and processing.
Consequently, most of the computer time can be dedicated to
more critical tasks and the processor can be given further help
