Figure 2. R channel grayscale map, G channel grayscale map, and
differential grayscale map
(a) (b) (c)
Figure 3. R Channel 3D Gray Distribution Map, G Channel 3D Gray
Distribution Map, Difference 3D Gray Distribution Map
Compared with Figure 3b, Figure 3a shows that the distribution of
background pixel values in the R and G channels is roughly the same, but
there are small differences. For example, the hanging hole red paint in
Figure 3b is significantly darker in brightness than in 3a. When the G
channel is subtracted from the R channel, the result in this area is
negative. To ensure that image details are not lost, the absolute value
of the difference is usually taken as the result and stored in the
corresponding pixel coordinates. However, based on prior information, it
is known that in the RGB color model, there are multiple mixed colors in
an image. The R component in the red, brown, and yellow series colors is
usually greater than the G component. This does not highlight the gray
level difference in the area to be segmented, which can cause the gray
level peak of the interfering pixel area to be similar to the target
hanger, and even the average gray level value of the interfering pixel
area to be higher than the target hanger, Although retaining more edge
information in the image, it also introduces a large amount of
interference.
The reason why is not used in this article is that we do not want to
construct the grayscale difference too clearly. When there are dark blue
or green areas in the background, the constructed grayscale difference
formula enlarges the background clutter, resulting in the problem of
image oversegmentation.
This article adopts the following formula to construct a differential
grayscale imageļ¼
1.3 Image segmentation and morphological
processing
This article uses the Otsu algorithm to perform threshold segmentation
on the constructed differential grayscale image, completing the
separation of crane grab boom and background. The Otsu algorithm solves
a binary classification problem, which satisfies the requirement of
maximum inter class variance and minimum intra class variance. The Otsu
algorithm is more suitable for grayscale histograms with bimodal
features. Namely, construct a grayscale difference map to meet the
bimodal features as much as possible. If the absolute value of the
difference is taken to construct a grayscale image, as shown in Figure
4a.