When performed in succession, these operations results in the
“closing” of the image. The dilation operation performed
makes lane and vehicle regions predominant by enhancing the
shape of these areas. The enhanced shape can later be detected
using the eccentricity and the number of pixels required to
make up the region. This is performed to effectively eliminate
the unwanted regions.
Fig. 2 shows an image of a closed
video frame overlayed on the original unprocessed video
frame. The regions within the red box indicate the lane regions
beginning to be enhanced in the video but do not yet have
enough pixels to be classified as a lane marking. The region
within the blue box shows how the eccentricity and the
number of pixels are enough to classify it as a lane region.
These results indicate the success of these operations in terms
of detecting road boundaries. Although the unconnected lane
markers are detected only when they appear close to the vehicle, the information gathered in prior frames will help us
track the lane markers. A road model is constructed as a
template for detecting any road, lane and vehicle regions in the
frames that follow. This would reduce a large amount of
processing required to perform morphological filtering in
every frame.