1 IMAGE PREPROCESSING
1.1 Color Space Selection
The main goal of crane grab boom extraction is to separate the crane
grab boom from the background in a complex environment. The position of
the industrial camera constantly changes with the movement of the crane
grab boom, and its background mainly includes foreign objects such as
segment beam lifting holes, segment beam planes, stones on the road
surface, local buildings on the construction site, and other mechanical
equipment. Due to significant differences in color between the crane’s
grab boom and the surrounding environmental characteristics. Therefore,
this article uses color features as the basis for image segmentation.
However, considering that the on-site construction environment is
outdoor and the working environment is complex, for example, it is
necessary to work under lighting conditions such as rainy, cloudy, and
night, which are very unstable. The resulting uneven lighting, shadows,
weak lighting, and strong lighting can make image segmentation unstable.
Using the HSV color model, although it can easily segment the target
color, it requires high stability of lighting and cannot meet the
requirements of outdoor construction. This paper adopts RGB color space,
which will make a specific color different under different lighting
conditions. The RGB color space contains 256 levels of red, green, and
blue, which can represent very subtle color difference, thus ensuring
that the edge details of the target will not be lost during image
segmentation. Secondly, after separating the three channels in the RGB
color space, the values of the image matrix can be mapped to the 0-255
range, and the calculation between each channel can be simplified by
adding, subtracting, multiplying, etc.
1.2 Construct differential grayscale
images
This article adopts the method of constructing differential grayscale
images to prepare for image segmentation. In the original grayscale
image, the grayscale image is constructed based on the different weights
of each pixel’s R, G, and B, which cannot reflect the differences
between each component, as shown in Figure 1.