2 Edge line detection and fitting of crane grab
boom
2.1 Edge Line Detection of Crane Boom Based on Hough
Transform
In an image, pixels with linear features need to be detected. The
commonly used methods are divided into two categories. The first type of
method predicts the distribution of pixel points through linear
regression, with the least squares method[18] and Ransac [19]
line fitting being the most representative. The least squares method is
only applicable to a group of pixels with straight line characteristics.
Serious deviation of outliers will directly affect the accuracy of
linear regression. When used to detect straight lines in Figure 5, only
one line with serious deviation can be obtained. Although Ransac line
fitting has strong anti-interference ability for outliers, it is only
suitable for detecting a group of pixels with line features. The second
type of method can perform line detection on any pixel in an image with
line features from a global perspective. Hough transform, probabilistic
Hough transform, LSD, and other methods can detect global line features,
among which Hough transform is the most classic and can be used to
detect any shape that can be expressed using mathematical formulas. The
principle is to transform points on a specific graph into a parameter
space, and obtain a maximum solution based on the vote accumulator in
the parameter space. This solution corresponds to the parameter of the
desired geometric shape. The transformation principle is shown in Figure
9.
Figure 9. Hough transform principle diagram
According to the principle of Hough transform, the more obvious the line
features are, that is, the more pixels in an edge detection image are
located on the same line, the higher the linearity of the pixel
arrangement, and the more detectable they can be. But usually, due to
the continuous movement of the crane’s grab boom, the industrial camera
also moves, resulting in constantly changing construction backgrounds on
site and the impact of different lighting conditions. It is inevitable
to cause some small interfering pixels in the differential grayscale
image. Select edge detection images detected in harsh working
environments for explanation. As shown in Figure 10. The Hough transform
detects straight lines when the corresponding voting thresholds in
Figure 10 are 30, 40, and 60, respectively.