49
5. The next step is to remove the objects present on the plane and to extract only the 3D
objects in the image that represents the obstacles. This
is done by presenting two
thresholded images of this sequence to CNN along with a reference or target image.
Figure 5-6 shows the initial condition, the input and the target images applied to the CNN
processor. A target image is constructed by removing all the plane objects and by leaving
only those above the plane. The targeting is achieved just after 170 iterations and thereby
producing the following CNN parameters:
(5-5)
Figure 5-
7 shows the output of the concept’s simulator. It can
be seen that all the plane
textures are removed and only the objects above the plane remain. The final output image
does not contain the lower edges of the objects since there is no way to discriminate the
lower edges of the objects and the edge pixel of texture on the plane.
