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.