Long material cutting method based on machine vision

Abstract

This invention relates to a machine vision-based long material cutting method, belonging to the field of cutting methods, and includes the following steps: S1: Set up a camera that can capture images of the long material and the cutting machine; S2: In the video frames at time t-1 and time t, perform instance segmentation on the images to obtain the long material instance with the largest area mask; S3: Compare the coordinates of the long material instance with the largest mask at time t-1 and time t to determine whether the long material has moved; if the long material has not moved on the roller bed, proceed to S4; otherwise, proceed to S2; S4: The long material is virtually cut by the cutting line of the cutting machine; S5: The cutting line virtually cuts the long material into segments A and B, and identifies the segment to be cut; S6: The smaller of the heights P1 and P2 in S5 is recorded as W, and W is compared with a set threshold; if W is less than or equal to the threshold, a warning is issued to the operator. This invention can identify the segment to be cut and calculate its height. By comparing it with the threshold, if the height does not meet the requirements, a warning is issued to the worker.

Description

Technical Field

[0001] The present invention relates to the field of long material cutting methods, and particularly to a long material cutting method based on machine vision. Background Art

[0002] During the production of long H-shaped steel in a steel mill, it is necessary to perform segmentation cutting, head cutting, tail cutting, and sampling cutting on the long material. The significance of sampling cutting is to perform component analysis on the long material steel to avoid inevitable losses caused by defective products flowing into the market. All the above-mentioned cutting operations only divide a small section at either one end or both ends of the long material. The size of the steel section after cutting is very small. If the size of the steel section after cutting is too small, the steel section is likely to get stuck in the gap of the roller table, seriously affecting the production rhythm and also unable to guarantee the safety of the workers cleaning the steel section. Summary of the Invention

[0003] In order to solve the problem of how to control the cutting size of the steel section and avoid the steel section being too small.

[0004] To achieve the above object, the long material cutting method based on machine vision of the present invention includes the following steps: S1: Install a camera that can capture the long material and the cutting machine; S2: In the video frame at time t - 1, perform instance segmentation on the image to obtain the long material instance with the largest area mask; in the video frame at time t, perform instance segmentation on the image to obtain the long material instance with the largest area mask; S3: Compare the coordinates of the largest mask long material instances at time t - 1 and time t to determine whether the long material moves; if the long material does not move on the roller table, go to S4; if the long material moves on the roller table, go to S2; S4: In the video frame at either time t - 1 or time t, perform virtual cutting on the long material by the cutting line of the cutting machine; S5: The cutting line virtually cuts the long material into section A and section B. Respectively, find the heights of the circumscribed rectangles P1 of section A of the long material and the circumscribed rectangle P2 of section B of the long material; the heights of P1 and P2 are the lengths of the long material along the feeding direction; record and compare the sizes of the heights of the circumscribed rectangles P1 and P2; denote them as W1 and W2 respectively; if W1 < W2, then section A is the to-be-cut section; otherwise, section B is the to-be-cut section; S6: Denote the smaller height among the heights of P1 and P2 in S5 as W, and compare W with the set threshold; if W is less than or equal to the threshold, issue a warning to remind the operator; if W is greater than the threshold, allow cutting.

[0005] Further, in step S1, the camera looks down at the long material and the cutting machine; the camera uses a flat-angle lens.

[0006] Wide-angle lenses have a short focal length and a wide angle of view, allowing for the capture of a large area of ​​scenery within a relatively short shooting distance. However, their drawback is the distortion of the captured scene. In contrast, this method uses a flat-angle lens, which minimizes distortion of long materials and the cutting machine in the image.

[0007] Furthermore, the center point coordinates of the maximum mask long piece instance at time t-1 and time t are compared. If the coordinate error is less than 1%, then the long piece has not moved on the roller bed.

[0008] A 1% accuracy is sufficient to effectively identify whether a long material has moved, and it is also accurate enough for image recognition. Any smaller error would increase the burden on image recognition and computation.

[0009] Furthermore, if we consider the cutting machine as a point mass and draw a perpendicular line to the feed direction of the long material, that is the cutting line.

[0010] The image is cut by making a perpendicular line to the direction of the long material feed. Compared with other cutting lines that are set to be inclined to the feed direction, the shape of the long material cut by the perpendicular line is more regular, which can also reduce the amount of calculation in the later calculation.

[0011] Furthermore, the time interval from time t-1 to time t is 1 minute.

[0012] The time interval should not be too short, as this can easily lead to misjudgments about whether the long material has moved. This is because the movement of the roller conveyor is not obvious in a short time, which also increases the difficulty of image recognition and calculation. Using a 1-minute time interval can ensure that the long material has not moved on the roller conveyor.

[0013] Further, in step S1, 4000 images of long materials under different motion states and 5000 images of cutting machines under different motion states are collected; labelme is used to label all long material images and labelImg is used to label all cutting machine images; yolvo8 is used to train all long materials and cutting machines to identify models of long materials and cutting machines respectively.

[0014] By using a large number of samples, the cutting machine and long material annotation training are improved to enhance the accuracy of long material identification.

[0015] Beneficial effects:

[0016] This method uses machine vision to identify long materials and cutting machines in an image, and then uses the cutting line of the cutting machine to virtually cut the long materials, determine the length of the segment to be cut, and compare the length of the segment to be cut with a set threshold. If the segment to be cut is not within the set threshold range, a warning is issued to remind the cutting worker. Attached Figure Description

[0017] Figure 1This is one of the states of long material cutting;

[0018] Figure 2 This is the second state of long material cutting. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] A machine vision-based method for cutting long materials includes the following steps:

[0021] S1: Set up a camera that can capture images of long materials and the cutting machine;

[0022] In this embodiment, the camera uses a top-down view to capture images of the long material and the cutting machine, and employs a level-angle lens. 4000 images of the long material and 5000 images of the cutting machine under different motion states are acquired. LabelMe is used to label all the long material images, and LabelImg is used to label all the cutting machine images. Yolvo8 is used to train models for recognizing both the long material and the cutting machine.

[0023] S2: In the video frame at time t-1, perform instance segmentation on the image to obtain the longest mask instance with the largest area; in the video frame at time t, perform instance segmentation on the image to obtain the longest mask instance with the largest area.

[0024] Specifically, a mask long material instance is the area occupied by the long material identified in the image. The reason for selecting the largest area mask long material instance is that the video frame captured on the rolling mill contains two mask long material instances: a complete long material and a portion of another long material. Actual segmentation, head-removal, tail-removal, and sampling cutting target the complete long material, so a portion of the mask long material instances needs to be discarded. By comparing the areas of the two mask long material instances, the mask long material instance with the largest area can be selected, which is the long material to be cut. This is because, in practice, the area of ​​a complete mask long material instance is much larger than the area of ​​a portion of the mask long material instance.

[0025] The time interval between t-1 and t can be set; in this embodiment, it can be set to 1 second. A longer time interval can effectively determine that the long material is indeed in a stationary state.

[0026] S3: Compare the coordinates of the maximum mask long material instances at time t-1 and time t to determine whether the long material has moved; if the long material has not moved on the roller table, go to S4; if the long material has moved on the roller table, go to S2;

[0027] Specifically, obtain the center point coordinates of the maximum mask long material instance at time t-1; obtain the center coordinates of the maximum mask long material instance at time t; compare the coordinates of the two aforementioned points. If the coordinate positions have not changed, it is considered that from time t-1 to time t, the maximum mask long material instance has not moved in the image of the video frame, that is, the long material has not moved on the roller table. Otherwise, the long material has moved on the roller table, and at this time, go to S2;

[0028] S4: In the video frame at either time t-1 or time t, virtually cut the long material with the cutting line of the cutting machine, and virtually cut the long material into section A and section B;

[0029] Specifically, since the long material is in a stopped state at time t-1 and time t, the image information in the video frames at both times is the same. Taking the cutting machine as a mass point, draw a perpendicular line perpendicular to the feeding direction of the long material, which is the cutting line. The cutting line virtually divides the long material into section A and section B; if the cutting line of the cutting machine cannot divide the long material, the worker will manually adjust the position of the cutting line along the feeding direction until the cutting line of the cutting machine can divide the long material.

[0030] S5: Respectively find the circumscribed rectangles P1 of section A of the long material and the circumscribed rectangle P2 of section B of the long material; compare the heights of the circumscribed rectangles P1 and P2; determine the length of the section to be cut;

[0031] Specifically, the widths of the circumscribed rectangle P1 and the circumscribed rectangle P2 are the lengths of the long material perpendicular to the feeding direction, and the widths of the circumscribed rectangle P1 and the circumscribed rectangle P2 are equal. The heights of the circumscribed rectangle P1 and the circumscribed rectangle P2 are the lengths of the long material in the feeding direction, denoted as W1 and W2 respectively. Compare the sizes of W1 and W2, see Figure 1 and Figure 2 , if W1 < W2, it indicates that the length of section A of the long material is short, which also indicates that section A of the long material is the section to be cut; otherwise, section B of the long material is the section to be cut. The section to be cut is the steel section after being cut from the long material, and it is the part of the long material that needs to be pushed off the roller table for analysis.

[0032] In actual operation, W1 and W2 cannot be equal because splitting cuts, head cuts, tail cuts, and sampling cuts only intercept a small section of the long material. That is to say, the numerical difference between W1 and W2 is very large.

[0033] S6: Compare W1 and W2 in S5 and take the smaller value, denoted as W. Compare W with a set threshold. If W is less than or equal to the threshold, it indicates that the cut steel segment is too small and may get stuck in the roller bed, and a warning is issued to the operator. If W is greater than the threshold, it indicates that the cut steel segment meets the requirements and cutting is allowed. The threshold can be set; in this embodiment, it is set to 45cm (this value is the value mapped from the image to the actual long material).

[0034] Based on the above-described preferred embodiments of the present invention, and through the foregoing description, those skilled in the art can make various changes and modifications without departing from the inventive concept. The technical scope of this invention is not limited to the contents of the specification, but must be determined according to the scope of the claims.

Claims

1. A long material cutting method based on machine vision, characterized in that, including the following steps, S1: Set up a camera that can capture long materials and a cutting machine; S2: In the video frame at time t - 1, perform instance segmentation on the image to obtain the long material instance with the largest area mask; in the video frame at time t, perform instance segmentation on the image to obtain the long material instance with the largest area mask; S3: Compare the coordinates of the long material instances with the largest masks at time t - 1 and time t to determine whether the long material has moved; if the long material has not moved on the roller table, go to S4; if the long material has moved on the roller table, go to S2; S4: In the video frame at either time t - 1 or time t, virtually cut the long material with the cutting line of the cutting machine; S5: The cutting line virtually cuts the long material into two segments, A and B. Respectively, find the heights of the circumscribed rectangles P1 of segment A of the long material and P2 of segment B of the long material; the heights of P1 and P2 are the lengths of the long material along the feeding direction; record and compare the magnitudes of the heights of the circumscribed rectangles P1 and P2; denote them as W1 and W2 respectively; if W1 < W2, then segment A is the segment to be cut; otherwise, segment B is the segment to be cut; S6: Denote the smaller of the heights of P1 and P2 in S5 as W, and compare W with a set threshold; If W is less than or equal to the threshold, issue a warning to alert the operator; If W is greater than the threshold, allow cutting.

2. The long material cutting method based on machine vision according to claim 1, characterized in that, In step S1, the camera looks down at the long material and the cutting machine; the camera uses a flat - angle lens.

3. The long material cutting method based on machine vision according to claim 1, characterized in that, In step S3, compare the center - point coordinates of the long material instances with the largest masks at time t - 1 and time t. If the coordinate error is less than 1%, then the long material has not moved on the roller table.

4. The long material cutting method based on machine vision according to claim 1, characterized in that, In steps S4 and S5, consider the cutting machine as a particle and draw a perpendicular line perpendicular to the feeding direction of the long material, which is the cutting line.

5. The long material cutting method based on machine vision according to claim 1, characterized in that, In step S2, the time interval from time t - 1 to time t is 1 minute.

6. The long material cutting method based on machine vision according to claim 1, characterized in that, In step S1, collect 4000 pictures of long materials and 5000 pictures of cutting machines in different motion states; use labelme to label all the long - material pictures and use labelImg to label all the cutting - machine pictures; use yolvo8 to train all the long materials and cutting machines respectively to obtain the models for identifying long materials and cutting machines.

Images