A method for measuring the bending radius of a cable based on image deep learning

Abstract

The present invention discloses a method for measuring the bending radius of a cable based on image deep learning, including obtaining a depth image and an RGB image of the cable; inputting the depth image and the RGB image into a multi-dimensional semantic segmentation network to obtain a cable area mask, and performing skeletonization processing on the cable area mask to obtain a cable feature curve in the image; discretely sampling the cable feature curve, fitting the sampling points with a polynomial curve F, and solving the curvature at each point of the polynomial curve to obtain the curvature at each point of the cable feature curve.

Description

Technical Field

[0001] The present invention relates to the technical field of cable assembly detection, and particularly to a method for measuring the bending radius of a cable based on image deep learning. Background Art

[0002] Cables are mainly used for power transmission and signal transmission, and are one of the important components of industrial production and daily life, playing a crucial role in the normal operation of today's society. The bending radius of cable laying has an important impact on the service life of the cable. If the bending radius is less than the specified requirement, it will cause excessive stress at the bending position of the cable, affecting the transmission of signals inside the cable. Due to the diverse types, different thicknesses, various colors, and complex orientations of cables, the measurement of the cable bending radius generally adopts the method of using a straight ruler for measurement. After the straight ruler measurement, the fitting conversion method needs to be used to finally obtain the bending radius, with relatively low efficiency and measurement errors varying from person to person.

[0003] Methods for measuring the cable bending radius include measuring the bending radius of the cable using a bending radius detection ruler, such as Patent CN113624103A; measuring the bending radius using vision, such as Patents CN111412849B and CN113935958A;

[0004] The above specific patent comparison documents are as follows:

[0005] 1) "Cable Bending Radius Detection Ruler", Patent No. CN113624103A. This invention discloses a cable bending radius detection ruler, which includes a main ruler and two branch rulers of the same length; a chute is provided along the length direction at the front section of the main ruler, a slider is provided in the chute, and the main ruler is provided with scales; the rear ends of the two branch rulers are hinged at the same point of the slider, and the front ends of the two branch rulers are connected by a rubber strip. The length of the branch ruler is known in advance. After the main ruler is perpendicular to the straightened rubber strip, the scale value is read, and the size of the arc radius can be quickly calculated. The whole process is simple and fast, with higher accuracy. The cable bending radius of the present invention is different from the above method, and the bending radius of the cable is measured by the machine vision method, which is more feasible for measuring the bending radius of the cable in a narrow space.

[0006] 2) "Method and Measurement System for Rapid Measurement of Bending Radius of Spacecraft Cables", Patent No. CN111412849B. This invention discloses a method and a measurement system for rapidly measuring the bending radius of spacecraft cables. It obtains images of the cable to be measured and the measurement strip by taking pictures; determines the type of the cable to be measured based on the RGB values of the cable to be measured in the image and the one-to-one correspondence between the RGB values of the preset cables and the cable types; obtains the ratio of the radial dimension to the axial dimension of the annular ring where the difference between the RGB value of the measurement strip and the RGB value of the cable to be measured in the image is greater than the preset threshold to calculate the diameter of the cable; obtains the bending radius of the bending part of the cable to be measured based on the edge trajectory; and determines whether the bending radius meets the specification requirements according to the ratio of the bending radius to the diameter. Through the technical path of taking pictures, recognition, and fitting, it can quickly measure the bending radius of cables with different thicknesses and quickly compare with the standard specifications, improving the consistency and efficiency of bending radius measurement. Different from the above method, this invention uses the deep learning semantic segmentation method to segment the cable area, which has higher universality compared with determining the cable type by comparing with the preset RGB values, and the segmented cable area has higher accuracy and less interference.

[0007] 3) "Method and Device for Detecting Bending Radius of Cable", Patent No. CN113935958A. This invention relates to a method and a device for detecting the bending radius of a cable. It acquires first image information including a positioning mark image and a real cable image in a real cable space, where the positioning mark image is an image preset at a preset position in the real cable space; determines the projection matrix of the camera according to the coordinate information of the positioning mark image; performs inverse projection processing on the real cable according to the projection matrix of the camera to obtain the corresponding curve of the real cable in the cable bending plane; and determines the bending radius of the real cable according to the curve in the cable bending plane. Using this method can improve the detection accuracy of the cable bending radius. Different from the above method, this invention uses a multi-dimensional fusion semantic segmentation method to segment the cable area, then obtains the cable curve through the sampling fitting method, and then measures the bending radius of the cable by solving the curvature of the curve. Summary of the Invention

[0008] To solve the above technical problems, the object of the present invention is to provide a method for measuring the bending radius of a cable based on image deep learning.

[0009] The object of the present invention is achieved by the following technical solutions:

[0010] A method for measuring the bending radius of a cable based on image deep learning includes:

[0011] A. Obtain the depth image and RGB image of the cable;

[0012] B. Input the depth image and the RGB image into a multi - dimensional semantic segmentation network to obtain a cable area mask, and perform skeletonization processing on the cable area mask to obtain the cable feature curve C in the image. pre ;

[0013] C. Discretely sample the cable feature curve, and use the polynomial curve F to fit the sampling points (u i , v i ), and solve the curvature at each point of the polynomial curve to obtain the curvature at each point of the cable feature curve.

[0014] Compared with the prior art, one or more embodiments of the present invention may have the following advantages:

[0015] During the measurement of the cable bending radius, the depth image and the RGB image of the cable are obtained through a structured light plus binocular stereo vision imaging system, which can be applicable to the measurement of cables with different specifications and different laying scenarios; the multi - dimensional semantic segmentation technology can accurately segment the cable area from the image; the use of non - uniform iterative sampling and then fitting the sampling points with the least - squares method can effectively improve the curve fitting accuracy and effectively improve the measurement accuracy of the cable bending radius. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flowchart of a method for measuring the cable bending radius based on deep learning;

[0017] Figure 2 is a schematic diagram of a cable to be measured;

[0018] Figure 3 is a mask image generated by multi - dimensional semantic segmentation;

[0019] Figure 4 is a data scatter plot and a data fitting curve plot. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the embodiments and the accompanying drawings.

[0021] As Figure 1 shown, a method for measuring the cable bending radius based on deep learning includes the following steps:

[0022] Step 10: Obtain the depth image and the RGB image of the cable;

[0023] The depth image and RGB image of the cable are obtained by using a handheld structured light plus binocular vision measurement system; among them, the handheld structured light plus binocular stereo vision imaging system is used to obtain the depth image and RGB image of a single cable, and the interpolation algorithm is used to fill in the missing information on the depth image; corresponding image pixel coordinate systems are established in the depth image and RGB image, the vertex of the coordinate system is the lower left vertex of the image, the positive direction of the u-axis is horizontal to the right, and the positive direction of the v-axis is vertical upward.

[0024] Step 20 inputs the depth image and RGB image into a multi-dimensional semantic segmentation network to obtain a cable area mask, and the cable area mask is skeletonized to obtain the cable feature curve C in the image pre ;

[0025] The depth image is converted from one-dimensional depth information into three-dimensional HHA information, and then the HHA image and RGB image are input into the multi-dimensional semantic segmentation network NANet. The features of the HHA image and the RGB image are fused in the semantic segmentation network to accurately segment the cable area and obtain the cable area mask line-Mask (as Figures 2-3 shown);

[0026] The cable area mask line-Mask is processed by the Zhang parallel fast skeletonization algorithm to obtain a single-pixel-width cable feature curve C pre .

[0027] As Figure 4 shown, in step 30, the cable feature curve is discretely sampled, and the sampling points (u i , v i ) are fitted by the polynomial curve F, and the curvature at each point of the polynomial curve is solved to obtain the curvature at each point of the cable feature curve.

[0028] For the cable feature curve C pre First, uniform sampling with an appropriate point distance is performed to obtain sampling points The sampling points are fitted by the least squares method to obtain the polynomial curve F 1 , and the curve curvature K 1 is solved. At the curvature greater than the threshold K threshold , the sampling density is increased to sample again to obtain a new round of sampling points The sampling operation is iterated until the polynomial curve no longer changes, that is, it is considered that the cable feature curve is fitted to the polynomial curve F final , and the curvature K final of the fitted cable feature curve is solved to obtain the curvature of the cable;

[0029] The polynomial curve F is:

[0030] v = s(u; a 0 , a1 ,…a n )(n < m) (1)

[0031] where a 0 ,a 1 ,…,a n are the polynomial coefficients, and m is the number of sampling points;

[0032] Equation (1) represents that there is a unique function in the function class such that the sum of the squares of the errors δ

[0033]

[0034] at the given sampling points u i is minimized, where δ i = s(u i ) - v i (i = 0, 1, 2, …, m).

[0035] The curvature calculation formula described above is as follows:

[0036]

[0037] where F′ and F″ are the first derivative and the second derivative of the polynomial curve, respectively.

[0038] Although the embodiments disclosed in the present invention are as above, the content described is only the embodiments adopted for the convenience of understanding the present invention and is not used to limit the present invention. Any person skilled in the art within the technical field to which the present invention pertains, without departing from the spirit and scope disclosed by the present invention, can make any modifications and changes in the form of implementation and details, but the scope of patent protection of the present invention shall still be subject to the scope defined by the appended claims.

Claims

1. A method for measuring the bending radius of a cable based on image deep learning, characterized in that, it includes A. Obtain the depth image and RGB image of the cable; B. Input the depth image and the RGB image into a multi-dimensional semantic segmentation network to obtain a cable area mask, and perform skeletonization processing on the cable area mask to obtain the cable feature curve C in the image pre ; C. Discretely sample the cable characteristic curve, and use the polynomial curve F to fit the sampling points (u i , v i ), and solve the curvature at each point of the polynomial curve to obtain the curvature at each point of the cable characteristic curve; In step B, convert the depth image from one-dimensional depth information into three-dimensional HHA information, then input the HHA image and the RGB image into the multi-dimensional semantic segmentation network NANet, realize the fusion of the HHA image features and the RGB image features and the accurate segmentation of the cable area in the semantic segmentation network, and obtain the cable area mask line-Mask; After processing the cable area mask line-Mask with Zhang's parallel fast thinning algorithm, a single-pixel-wide cable feature curve C is obtained pre ; In C, for the cable characteristic curve C pre perform uniform sampling of the point spacing to obtain sampling points Use the least squares method to fit the sampling points to obtain the polynomial curve F 1 , solve the curve curvature K 1 , increase the sampling density at the curvature greater than the threshold K threshold to resample and obtain a new round of sampling points Iterate the sampling operation until the polynomial curve no longer changes, that is, it is considered that the cable characteristic curve is fitted to the polynomial curve F final , solve the curvature K of the fitted cable characteristic curve final to obtain the curvature of the cable; The polynomial curve F is: v=s(u;a 0 ,a 1 ,…a n ) n<m (1) where a 0 , a 1 , …, a n are polynomial coefficients, and m is the number of sampling points; Equation (1) indicates that there is a unique function in the function class : Minimize the sum of the squares of the error δ i at a given sampling point u i such that δ i = s(u i ), where i = 0, 1, 2, …, m.

2. The method for measuring the bending radius of a cable based on image deep learning according to claim 1, characterized in that, in step A, a hand-held structured light plus binocular vision measurement system is used to obtain the depth image and RGB image of the cable; wherein, the structured light and the binocular stereo vision measurement system adopted obtain the actual position information and color information of the cable in space through the structured light and binocular stereo vision imaging principles, and at the same time output the depth image and RGB image of the current position, and an interpolation method is used to fill the missing depth information in the depth image.

3. The method for measuring the bending radius of a cable based on image deep learning according to claim 1, characterized in that, establish a corresponding image pixel coordinate system in the depth image and the RGB image, wherein, the vertex of the coordinate system is the lower left vertex of the image, the positive direction of the u-axis is horizontal to the right, and the positive direction of the v-axis is vertical upward.

4. The method for measuring the bending radius of a cable based on image deep learning according to claim 1, characterized in that, the curvature calculation formula at each point of the cable characteristic curve is: where F′ and F″ are the first derivative and the second derivative of the polynomial curve respectively.

Images