Information Technology Reference
In-Depth Information
The inner workings of the Lane Region Finder is shown in Fig. 5b. First,
two Residual Label Growing (RLG) operations are performed using a binary
mask; one on lblCb and the other on lblCr. The two residual labeled regions are
converted to binary images by assigning each non-background label to a binary
one. The two binary images are combined using a bitwise OR operation and
assigned as yellow lane region. This region is then dilated and inverted to serve
as mask. The purpose of the mask is to avoid re-detection in areas that have
already been considered as yellow lane regions. The structuring element used in
dilation is row vector with a height equivalent to 12ft in the IPM image (6ft
on either side of the detected yellow regions). RLG is performed again but this
time on lblY using the dilated and inverted mask. The white and yellow marker
regions are merged to depict detected lane regions corresponding to a particular
template. The merged region is also inverted and dilated to serve as a mask for
the next LRF block in Fig. 5a.
Yellow color content is prevalent in the Cb and Cr channels of the YCbCr
space. Depending on the intensity of yellow, the Y channel may occasionally
contribute some information. As a result, yellow lane marker detection is per-
formed using Cb and Cr components. However, white color is intensity dependent
and contributed to only by the Y channel. Hence, white lane marker detection
is performed using only the Y components.
4.4 Ellipse Modeling
Connected components labeling is performed on the Ensemble Lane Regions
(ELR) in Fig. 5a to find a collection of distinct objects. The horizontal lengths
of each object is measured for classification. Based on the FHA specifications, if
the measured length of an object exceeds the equivalent of 10ft in the IPM image,
then it is categorized as a full or solid line. Otherwise, the object is categorized as
a broken line. In Fig. 6, the solid line is shown in red and broken lines are shown
in Cyan. Each broken line is modeled as an ellipse whose major axis shown in
brown is projected towards the front of the vehicle on the left. For solid lines, the
leading 10ft of pixels shown in yellow are used in modeling the ellipse. The major
axis of the ellipse allows to estimate the traversing direction of the markers from
one frame to the next. The Hough transform was initially used to approximate
0ft
30ft
d
Ideal left lane location
Y
X
Ideal right lane location
Lane markers
on a curve
Fig. 6.
Ellipse modeling and major axis projection
Search WWH ::
Custom Search