Method and system for measuring double-sided contour of stainless steel medium-thickness plate

Abstract

The invention belongs to the technical field of plate shape monitoring and visual analysis, and discloses a stainless steel medium-thickness plate double-face contour measurement method and system, and the method comprises the steps: S1, completing the calibration of a double-line-scan digital camera, and obtaining the upper and lower surface images of a medium-thickness plate; s2, carrying out binocular stereoscopic vision three-dimensional reconstruction to generate a point cloud picture, and converting the point cloud picture into a point cloud picture under the same coordinate system for display; s3, eliminating transverse rotation and deviation according to point cloud center line fitting and plane fitting of the upper surface and the lower surface of the medium-thickness plate; s4, comparing the geometrical relationship between the point clouds of the upper and lower surfaces to quantify local vibration and thickness change, correcting vibration errors, and obtaining the thickness of the medium-thickness plate; and S5, extracting an edge point set from the point cloud image to obtain the overall contour of the medium-thickness plate, and obtaining the position of a shear line based on the width and edge linearity detection of the medium-thickness plate. By the adoption of the technical scheme, the overall contour, overall thickness distribution and accurate shearing position of the medium-thickness plate can be obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of strip shape monitoring and visual analysis, and particularly relates to a method and system for measuring the double-sided profile of medium and heavy stainless steel plates. Background Art

[0002] Currently, during the production process of medium and heavy plates, the rough billet undergoes multiple repeated rollings, and temperature fluctuations, strip length, rolling force, etc. will all affect the strip shape, and will also cause the thickness of the medium and heavy plate to be uneven at different positions. Due to the existence of these objective factors, after rolling, the medium and heavy plate generally has strip shape defects. Such as head and tail deformations in the length direction and sickle bend deformations in the width direction. After rolling to the standard thickness, according to the standard size width of the medium and heavy plate, side shearing is performed on the medium and heavy plate, and then operations such as cutting the head, cutting the tail, and sizing are carried out, which are important processes for producing finished medium and heavy plates that meet the specification requirements.

[0003] At present, after the medium and heavy plate completes side shearing, the irregular shapes of the head and tail still need to be visually judged by artificial eyes to determine the cutting positions of the head and tail, which is likely to cause phenomena such as under-shearing and over-shearing. The thickness of the medium and heavy plate is usually measured by a laser thickness gauge. Although this instrument has high measurement accuracy and can measure the thickness of the steel plate in real time, the laser thickness gauge has poor adaptability to medium and heavy plates with different materials and surface characteristics. In addition, the high-temperature environment will also cause scattering or refraction of the laser beam, affecting the measurement accuracy, and a single laser thickness gauge cannot comprehensively obtain the overall thickness distribution of the medium and heavy plate.

[0004] In terms of visual measurement, due to the complex production site environment, there are adverse factors such as high temperature, dust, and noise, which affect the stability and measurement accuracy of the optical sensor for collecting images and the measurement system. At the same time, the friction between the medium and heavy plate and the roller table during transportation may cause unpredictable rotation and offset, accompanied by up and down vibrations, further increasing the uncertainty of the movement trajectory of the medium and heavy plate, and it is impossible to accurately obtain the contour information of the medium and heavy plate, affecting the overall thickness statistics and shear position determination of the medium and heavy plate. Therefore, in order to obtain the overall contour, overall thickness distribution of the medium and heavy plate, and determine the shear position, it is of important application value to invent a detection method for medium and heavy plates with high measurement accuracy and strong adaptability. Summary of the Invention

[0005] To solve the problems existing in the prior art, the present invention provides a method and system for measuring the double-sided profile of medium and heavy stainless steel plates, which uses a multi-camera group to simultaneously collect images on the upper and lower sides of the medium and heavy plate, performs binocular stereo vision three-dimensional reconstruction, obtains the three-dimensional data of the upper and lower surface contours of the medium and heavy plate, corrects the lateral offset and up and down vibration errors of the medium and heavy plate, and obtains the accurate information of the surface contour of the medium and heavy plate and the overall thickness distribution of the medium and heavy plate. Based on the detection of the width consistency and edge linearity of the medium and heavy plate, the shear positions of different-shaped plate heads are accurately located, so as to quickly determine the shear position and shear length.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] A method for measuring the double-sided profile of medium-thick stainless steel plates, the method comprising:

[0008] S1. Complete the calibration of the double-line array camera to obtain the images of the upper and lower surfaces of the medium-thick stainless steel plate;

[0009] S2. Perform binocular stereo vision three-dimensional reconstruction on the images of the upper and lower surfaces of the medium-thick stainless steel plate to generate a point cloud map, and convert it to be displayed in the same coordinate system;

[0010] S3. Eliminate the lateral rotation and offset according to the center line fitting and plane fitting of the point clouds on the upper and lower surfaces of the medium-thick plate;

[0011] S4. Compare the geometric relationships between the point clouds on the upper and lower surfaces to quantify the local vibration and thickness change, correct the vibration error, and obtain the thickness of the medium-thick plate;

[0012] S5. Extract the edge point set from the point cloud map to obtain the overall contour of the medium-thick plate, and detect the shearing line position based on the width and edge linearity of the medium-thick plate.

[0013] Preferably, in the S1, the calibration of the double-line array camera includes:

[0014] Select a transparent calibration plate, and through the stable movement of the calibration plate, scan and splice row by row to generate a complete two-dimensional image;

[0015] Extract the feature points in the two-dimensional image, and combine with the known physical size of the calibration plate to optimize and calculate the internal and external parameters of the camera through the calibration algorithm;

[0016] Calibrate the geometric relationship of the double-line array camera.

[0017] Preferably, in the S2, the binocular stereo vision three-dimensional reconstruction of the images of the upper and lower surfaces of the medium-thick stainless steel plate to generate a point cloud map includes:

[0018] After obtaining the images of the upper and lower surfaces of the medium-thick stainless steel plate, perform stereo correction on the images according to the calibration parameters, and map the images to a coplanar plane to eliminate the vertical parallax;

[0019] Use the stereo matching algorithm to calculate the disparity map, and generate depth information through the position difference of the corresponding pixel points in the two images;

[0020] According to the disparity map and the internal and external parameters of the camera, project the pixel points into the three-dimensional space to generate the point cloud data of the surface of the medium-thick plate.

[0021] Preferably, in the S3, eliminating the lateral rotation and offset according to the center line fitting and plane fitting of the point clouds on the upper and lower surfaces of the medium-thick plate includes:

[0022] Detect low-density points or abnormal distance points through statistical characteristic analysis method, combine local curvature calculation method and normal vector consistency analysis, eliminate interference points from the point cloud image after 3D reconstruction, and use nearest neighbor interpolation to repair the vacant area;

[0023] Fit the point cloud planes of the upper and lower surfaces, calculate the center lines respectively, and judge the offset and rotation amounts through the relationship between the center lines, and construct a rigid body transformation matrix for correction;

[0024] Among them, the fitting of the point cloud planes of the upper and lower surfaces includes:

[0025] Use the RANSAC algorithm to fit the upper and lower surfaces of the medium-thick plate, and the fitting formula is:

[0026] ;

[0027] Among them, is the fitting parameter, is the coordinate of the point in the 3D point cloud;

[0028] Calculating the center line includes:

[0029] Select parallel cross-section point clouds with a fixed interval on the upper and lower surfaces respectively, and calculate the geometric center of each cross-section:

[0030] ;

[0031] Among them, The geometric center point of the cross-section is the coordinate average value (i.e., the centroid) of all the point clouds within a certain slice cross-section, The number of point clouds within the current cross-section, The index variable indicates that the current point being accumulated is the th point, and the value range is from 1 to , is the point cloud coordinate on the cross-section;

[0032] Connect the center points of all cross-sections to obtain the center lines of the upper and lower surfaces;

[0033] Judging the offset and rotation amounts through the relationship between the center lines includes:

[0034] ;

[0035] Among them, and are the center point coordinates of the upper and lower surfaces respectively;

[0036] Constructing a rigid body transformation matrix for correction includes:

[0037] According to the offset and rotation amounts calculated above, construct a rigid body transformation matrix:

[0038] ;

[0039] Among them, is the rotation matrix;

[0040] Perform a rigid body transformation on the point clouds of the upper and lower surfaces:

[0041] ;

[0042] Among them, is the original point cloud, is the corrected point cloud.

[0043] Preferably, in S4, compare the geometric relationship between the point clouds of the upper and lower surfaces to quantify local vibrations and thickness changes, correct the vibration error, and obtain the thickness of the medium plate, including:

[0044] Use frequency domain analysis to extract the vibration error component, calculate the offset vector, and adjust it along the normal vector direction;

[0045] After eliminating the vibration error, the point clouds of the upper and lower surfaces of the medium plate are divided into several small regions, and the average thickness of the points in each region is calculated, and finally the thickness distribution and overall thickness statistics of the medium plate are obtained.

[0046] Preferably, using frequency domain analysis to extract the vibration error component, calculate the offset vector, and adjust it along the normal vector direction includes:

[0047] Adopt nearest neighbor point matching to make each point in the point clouds of the upper and lower surfaces find the corresponding relationship, and calculate the vertical distance between each pair of matching points on the upper and lower surfaces:

[0048] ;

[0049] Among them, is the projection distance of the point cloud on the normal vector, and are the corresponding points on the upper and lower surfaces respectively, is the global normal vector;

[0050] For Perform a fast Fourier transform to extract the main frequency distribution:

[0051] ;

[0052] Among them, is the frequency component of the vibration signal obtained by the fast Fourier transform, is the index of the discrete data points, is the total number of data points in the input signal, is the projection distance of the point cloud on the normal vector, is the complex exponential basis function of Fourier transform;

[0053] According to the main frequency distribution, the vibration frequency range is set as , and a band-pass filter is used to extract the vibration signal:

[0054] ;

[0055] Among them, is the vibration error signal after frequency-domain filtering, are the upper and lower limits of the vibration frequency range, is the frequency component of the vibration signal, is the complex exponential basis function of inverse Fourier transform;

[0056] Vibration error components are extracted using frequency-domain analysis , which represents the vertical vibration between the upper and lower surfaces of the medium-thick plate, and the offset vector is calculated and adjusted along the normal vector direction;

[0057] ;

[0058] For the original coordinates of each pair of point clouds, the point clouds are adjusted according to the vibration components:

[0059] ;

[0060] Among them, are the corrected point coordinates, is the original point coordinate of the upper surface before correction, is the corresponding original point coordinate of the lower surface before correction, is the offset vector;

[0061] By correcting the error between the corresponding point clouds on the upper and lower surfaces of the medium-thick plate, the influence of the medium-thick plate vibration is eliminated.

[0062] Preferably, after eliminating the vibration error, the point clouds on the upper and lower surfaces of the medium-thick plate are divided into several small regions, and the average thickness of the points in each region is calculated. Finally, the thickness distribution and overall thickness statistics of the medium-thick plate are obtained, including:

[0063] For each grid region, calculate the average thickness of all point pairs within the region:

[0064] ;

[0065] Among them, is the number of points within the grid;

[0066] The point clouds on the upper and lower surfaces are segmented into small regions using the grid-averaged thickness method, and the average thickness is calculated for each grid, and finally the regional thickness distribution map of the medium-thick plate is generated;

[0067] According to the regional thickness distribution map of the medium plate, the locally varying thickness of the medium plate is identified, and finally the overall thickness of the medium plate is obtained.

[0068] Preferably, in S5, extracting the edge point set from the point cloud map to obtain the overall contour of the medium plate, and detecting the shearing line position based on the width and edge linearity of the medium plate includes:

[0069] Obtaining the width of the medium plate through the medium plate specification and model parameters ;

[0070] According to the width of the medium plate , calculating the corresponding width point by point for the head and tail regions of the medium plate contour, constructing a width sequence and preliminarily determining the cutting position according to the change of the width, and the corresponding width sequence is ;

[0071] Judging whether the width sequences in the regions are consistent. If the width consistency condition is met, that is, the width fluctuation is within the specified threshold, it is considered that the region is suitable for cutting, and the continuous sub-regions of the width sequence are detected. Let the starting index be . If the preset conditions are met, it is considered that the width of the region is consistent; among them, the preset conditions are:

[0072] ;

[0073] Among them, is the length of the detection range, is the width consistency threshold;

[0074] In the region where the width consistency is met, the left and right edge points of each row are further extracted, and the linear fitting is used to calculate the fitting residual to judge whether the edge meets the preset linearity requirement, that is, by comparing the maximum value of the fitting residual with the set threshold, accurately judging whether the left and right edges are straight, so as to further lock the optimal cutting position and ensure that the cut edge is flat and smooth.

[0075] The present invention also provides a double-sided contour measurement system for stainless steel medium plates. The system is used for any one of the methods described above. The system includes: an image acquisition unit, a three-dimensional reconstruction unit, a motion correction unit, a thickness measurement and error correction unit, and a contour detection and shearing positioning unit;

[0076] The image acquisition unit is used to complete the calibration of the double linear array camera to obtain the upper and lower surface images of the stainless steel medium plate;

[0077] The three-dimensional reconstruction unit is used to perform binocular stereo vision three-dimensional reconstruction on the upper and lower surface images of the stainless steel medium plate to generate a point cloud map and display it in the same coordinate system;

[0078] The motion correction unit is used to eliminate lateral rotation and offset according to the fitting of the center line of the point cloud on the upper and lower surfaces of the medium-thick plate and plane fitting.

[0079] The thickness measurement and error correction unit is used to compare the geometric relationship between the point clouds on the upper and lower surfaces to quantify local vibration and thickness change, correct the vibration error, and obtain the thickness of the medium-thick plate.

[0080] The contour detection and shearing positioning unit is used to extract the edge point set from the point cloud map to obtain the overall contour of the medium-thick plate, and detect the shearing line position based on the width of the medium-thick plate and the linearity of the edge.

[0081] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0082] The present invention performs stereo vision three-dimensional reconstruction on the collected images of the upper and lower surfaces of the medium-thick plate, eliminates interference points, eliminates lateral rotation and offset by fitting the center line of the point clouds on the upper and lower surfaces and plane fitting, compares the geometric relationship between the point clouds on the upper and lower surfaces to quantify local vibration and thickness change, corrects the vibration error, obtains the overall thickness of the medium-thick plate, extracts the edge point set from the point cloud map to obtain the edge contour of the medium-thick plate, and detects the shearing line position based on the width of the medium-thick plate and the linearity of the edge. The present invention can obtain the overall contour, overall thickness distribution and accurate shearing position of the medium-thick plate, provide comprehensive and accurate data for the production quality of medium-thick plates in the metallurgical industry, promote the optimization of the production process and the improvement of the production quality of medium-thick plates, and has important theoretical significance and great practical application value. Brief Description of the Drawings

[0083] In order to more clearly illustrate the technical solutions of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0084] Figure 1 It is a schematic flow chart of the double-sided contour measurement method for stainless steel medium-thick plates in the embodiments of the present invention;

[0085] Figure 2 It is another schematic flow chart of the double-sided contour measurement method for stainless steel medium-thick plates in the embodiments of the present invention;

[0086] Figure 3 It is a schematic diagram of the actual measurement of medium-thick plates on the production line in the embodiments of the present invention;

[0087] Figure 4 It is a top view schematic diagram of the actual measurement of medium-thick plates on the production line.

[0088] In the figure: 1. Profiler, 2. Metal protective box, 3. Linear light source, 4. Automatic cleaning device, 5. Synchronous trigger, 6. Medium-thick plate to be measured, 7. Roller table. Specific embodiments

[0089] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0090] To make the above objects, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

[0091] Embodiment 1

[0092] As Figures 1-4 shown, the double-sided contour shape measurement method and system for stainless steel medium-thick plates mainly include dual linear array camera calibration, binocular stereo vision three-dimensional reconstruction, eliminating the motion deviation of the medium-thick plate, and adaptively determining the shearing position. This process includes the following steps:

[0093] The structured light scanning device is composed of 4 high-precision industrial linear array cameras. Two linear array cameras are integrated and placed side by side to form a contour detector. One contour detector is installed above the roller table 7 and arranged parallel to the roller table 7. The total field of view of the two cameras completely covers the transverse width of the medium-thick plate. At the same time, it is ensured that there is a partial overlapping area in the images collected by the two cameras. A dust-proof cover is added to the contour detector to avoid dust accumulation. Another contour detector is installed below the roller table 7 and symmetrically arranged with the upper surface camera, and the symmetry center is the surface of the roller table 7. A closed metal protective box 2 is used, and a transparent high-temperature resistant glass window is installed inside for dust prevention, shock prevention, and oil stain prevention, and an automatic cleaning device 4 is equipped. A uniformly scattered linear light source 3 is used on the upper surface of the roller table 7, and a high-power linear light source 3 is used on the lower surface to ensure clear image acquisition.

[0094] A synchronous trigger 5 is installed at the position of the roller table 7 corresponding to the contour detector to record the real-time speed and position of the medium-thick plate movement and provide a synchronous trigger signal. Ensure that the upper and lower cameras collect images of the same area simultaneously.

[0095] The present invention discloses a double-sided contour measurement method for stainless steel medium-thick plates, and the method includes:

[0096] S1. Complete the calibration of the dual linear array cameras to obtain the upper and lower surface images of the stainless steel medium-thick plate;

[0097] S2. Perform binocular stereo vision three-dimensional reconstruction on the upper and lower surface images of the medium-thick stainless steel plate to generate a point cloud map, and convert it to be displayed in the same coordinate system;

[0098] S3. Eliminate lateral rotation and offset according to the fitting of the center line and plane fitting of the point clouds on the upper and lower surfaces of the medium-thick plate;

[0099] S4. Compare the geometric relationships between the point clouds on the upper and lower surfaces to quantify local vibrations and thickness changes, correct the vibration errors, and obtain the thickness of the medium-thick plate;

[0100] S5. Extract the edge point set from the point cloud map to obtain the overall contour of the medium-thick plate, and obtain the shearing line position based on the width and edge linearity detection of the medium-thick plate.

[0101] S1: The calibration of the double linear array cameras in the upper and lower surface profilers obtains two-dimensional images through relative motion with the object. First, select a transparent calibration plate (high-contrast black and white checkerboard grid). Through the stable movement of the calibration plate, scan and splice line by line to generate a complete two-dimensional image. Then extract the checkerboard corner points in the image. Combining the known physical dimensions of the calibration plate, calculate the internal parameters (focal length, principal point, distortion coefficient) and external parameters (rotation matrix and translation vector) of the camera through the calibration method. The imaging model of the linear array camera forms a two-dimensional image through motion scanning, and its projection relationship is:

[0102] ;

[0103] Among them, is the pixel coordinate of the image plane, is the internal parameter matrix of the camera, including focal length, principal point, distortion coefficient, is the rotation matrix between the camera coordinate system and the world coordinate system, is the translation vector of the camera, is the three-dimensional point in the world coordinate system.

[0104] For the relative external parameter calibration of the double cameras, obtain the relative position relationship between the two camera coordinate systems. Let the external parameters of the left camera ( ) and the right camera ( ) be and respectively. The relative external parameter of the right camera coordinate system relative to the left camera coordinate system is:

[0105] ;

[0106] ;

[0107] Among them, is the rotation matrix of the right camera coordinate system relative to the left camera coordinate system, is the translation vector of the right camera coordinate system relative to the left camera coordinate system.

[0108] After calibration, the accuracy of the results is evaluated through reprojection error, and the internal and external parameters are saved.

[0109] S2: After the upper and lower surface profilers obtain the medium plate images, use Bouguet stereo calibration. According to the calibration parameters obtained in S1, perform stereo calibration on the images, map the images to a coplanar plane to eliminate vertical parallax, and further calculate the calibration rotation matrix to achieve row alignment. First, define the baseline direction, construct an orthogonal basis, and build the calibration rotation matrix. Then, apply the average rotation to the left and right cameras, with each camera rotating by half to make the dual-camera image planes coplanar, and calibrate the projection matrix to achieve image row alignment. The implementation process of calculating the calibration rotation matrix is as follows:

[0110] ;

[0111] ;

[0112] ;

[0113] ;

[0114] ;

[0115] Among them, are respectively the components of the translation vector of the right camera coordinate system relative to the left camera coordinate system in the direction, is the pole in the same direction as the translation vector, is the vector in the direction of the image plane, is the vector perpendicular to the plane,

[0116] Apply the rotation matrix, make the dual-camera planes coplanar, calculate the calibrated projection matrix, convert the original image into the calibrated image, eliminate distortion and perspective differences, and the implementation process of achieving image row alignment is as follows:

[0117] ;

[0118] ;

[0119] ;

[0120] ;

[0121] ;

[0122] Among them, is the calibration rotation matrix of the left camera, is the calibration rotation matrix of the right camera, is the unified internal parameter matrix after calibration, is the projection matrix of the left camera, is the projection matrix of the right camera.

[0123] For the pixel points in the left camera image, find their matching points in the right camera image. Calculate the pixel difference between the corresponding points in the two images to obtain the disparity, that is, calculate the horizontal displacement of the same object in the two images. According to the relationship between the disparity and the depth, convert the disparity map into a depth map, and calculate the corresponding depth value according to the disparity. The depth map provides the spatial position of each pixel point and can be used for further three-dimensional analysis and applications. The implementation methods for calculating the disparity map and the depth map are as follows:

[0124] ;

[0125] ;

[0126] Where: is the pixel difference between the pixel points in the left camera image and the pixel points in the right image, and respectively represent the horizontal coordinates of the corresponding points in the two images, is the depth corresponding to this pixel, is the camera focal length, is the baseline between the two cameras (the horizontal distance between the two cameras).

[0127] Based on the ideal imaging model of the binocular camera, under the conditions of image row correction and coplanarity, using the disparity map and the pixel coordinate differences, combined with the internal and external parameters of the camera, calculate the coordinates of each pixel point in the three-dimensional space to generate the point cloud data on the surface of the medium plate. The pixel coordinates of the left camera are , and the pixel coordinates of the right camera are . According to the imaging geometric relationship of the binocular camera, the coordinates in the three-dimensional space are calculated as follows:

[0128] ;

[0129] ;

[0130] ;

[0131] Where: is the baseline between the two cameras (the horizontal distance between the two cameras). and are the horizontal pixel coordinates in the left and right camera images, and is the vertical pixel coordinate in the left and right camera images, is the pixel difference between the pixel points in the left camera image and the pixel points in the right image, is the camera focal length.

[0132] Specifically, feature extraction is performed on the obtained medium-thick plate image, stereo matching is carried out, a disparity map is generated, and three-dimensional coordinate information is calculated through disparity. The double cameras on both sides of the medium-thick plate are symmetrically placed with respect to the roller table surface. According to the geometric relationship, the coordinate systems of the point cloud images on the upper and lower surfaces of the medium-thick plate are transformed into the same coordinate system.

[0133] S3: Due to problems such as noise, environmental dust, uneven illumination, motion errors, and projection distortion, interference points are generated in the point cloud images on the upper and lower surfaces of the medium-thick plate after three-dimensional reconstruction. Low-density points or abnormal distance points are detected through statistical characteristic analysis methods. Low-density points refer to points in a certain local area of the point cloud where the point distribution density is significantly lower than the surrounding area (such as data loss caused by dust occlusion or sensor failure), and abnormal distance points refer to points in the point cloud where the vertical distance from some points to the surface of the medium-thick plate significantly deviates from the normal range (such as outliers caused by motion errors or projection distortion). Combining the local curvature calculation method and the normal vector consistency analysis, the point clouds that do not conform to the plane characteristics of the medium-thick plate are removed. After removing the interference points, the nearest neighbor interpolation is used to repair the vacant areas. For the holes formed by low-density or removed points, the mean value of the nearest neighbor coordinates of the surrounding valid points is taken as the interpolation to ensure the integrity and consistency of the point cloud data.

[0134] Specifically, low-density points or abnormal distance points are detected through statistical characteristic analysis methods. Combining the local curvature calculation method and the normal vector consistency analysis, interference points are removed from the point cloud image after three-dimensional reconstruction, and the nearest neighbor interpolation is used to repair the vacant areas. Plane fitting of the point clouds on the upper and lower surfaces is performed. The centerlines are calculated respectively, and the offset and rotation amounts are judged through the relationship between the centerlines. A rigid body transformation matrix is constructed for correction.

[0135] S4: Convert the point cloud coordinate system generated by the lower profiler to the point cloud coordinate system of the upper profiler. The upper and lower profilers are symmetrically placed with respect to the roller table surface, and the corresponding simplified rotation matrix and translation vector are:

[0136] ;

[0137] where, and respectively represent the distances from the upper and lower profilers to the roller table surface.

[0138] Through the rotation matrix and translation vector, the point cloud image coordinate system generated by the lower profiler is transformed into the point cloud image coordinate system generated by the upper profiler.

[0139] S5: During the movement of the medium-thick plate, due to the friction with the roller path 7 or the action of lateral force, the medium-thick plate may rotate around its central axis (around the Z-axis), or tilt around the transverse and longitudinal axes (around the X and Y axes). Perform point cloud plane fitting on the upper and lower surfaces. Calculate the center lines respectively, and judge the offset and rotation amount through the relationship between the center lines. Use the RANSAC algorithm to perform plane fitting on the upper and lower surfaces of the medium-thick plate, and the fitting formula is:

[0140] ;

[0141] where, is the fitting parameter, is the coordinate of the point in the three-dimensional point cloud.

[0142] Select the parallel cross-section point clouds with a fixed interval on the upper and lower surfaces respectively (take a section of point cloud at a certain distance along the length direction of the medium-thick plate), and calculate the geometric center of each cross-section:

[0143] ;

[0144] where, the geometric center point of the cross-section is the coordinate average value (centroid) of all the point clouds in a certain slice cross-section, the number of point clouds in the current cross-section, the index variable, indicating that the current being accumulated is the th point, and the value range is from 1 to , is the point cloud coordinate on the cross-section.

[0145] Connect all the cross-section center points to obtain the center lines of the upper and lower surfaces. The center lines of the upper and lower surfaces should theoretically coincide, but due to the deformation of the medium-thick plate and measurement errors in actual measurement, there is an offset, and the offset vector is:

[0146] ;

[0147] where, and are the center point coordinates of the upper and lower surfaces respectively.

[0148] Use the angle between the normal vectors of the upper and lower surfaces to directly construct the rotation matrix through the rotation formula:

[0149] ;

[0150] where, and are the normal vectors of the upper and lower surfaces respectively.

[0151] According to the offset and rotation amount calculated above, construct the rigid body transformation matrix:

[0152] ;

[0153] Perform a rigid body transformation on the point clouds of the upper and lower surfaces:

[0154] ;

[0155] wherein, is the original point cloud, is the corrected point cloud.

[0156] By fitting the center lines of the point clouds on the upper and lower surfaces of the medium-thick plate and analyzing the plane normal vectors, the offset and rotation angles are accurately calculated. After correction using the rigid body transformation matrix, the normal vectors of the corrected upper and lower surface point clouds are parallel, and the included angle of the fitted plane approaches zero.

[0157] S6: After eliminating the offset and rotation, eliminate the vibration influence through the point cloud data of the upper and lower surfaces. The vibration of the medium-thick plate usually shows small periodic or random fluctuations in the point cloud data, which are manifested as fluctuations in the vertical distance between the local undulating center lines in the point cloud image, or abnormal local curvatures of the point cloud. The deviation of the upper and lower surface point clouds has been corrected in step 5, and the local vibration and thickness change can be quantified by directly comparing the geometric relationship between the upper and lower surface point clouds.

[0158] Specifically, use frequency domain analysis to extract the vibration error components, calculate the offset vector, and adjust along the normal vector direction. After eliminating the vibration error, the point clouds on the upper and lower surfaces of the medium-thick plate are divided into several small regions, and the average thickness of the point clouds in each region is calculated, and finally the thickness distribution and overall thickness statistics of the medium-thick plate are obtained.

[0159] Adopt the nearest neighbor point matching to make each point in the upper and lower surface point clouds find the corresponding relationship, and calculate the vertical distance between each pair of matching points on the upper and lower surfaces.

[0160] ;

[0161] wherein, is the projection distance of the point cloud on the normal vector, and are the corresponding points on the upper and lower surfaces respectively, is the global normal vector.

[0162] Perform a fast Fourier transform on to extract the main frequency distribution:

[0163] ;

[0164] wherein, is the frequency component of the vibration signal obtained by the fast Fourier transform, is the index of the discrete data points, is the total number of data points in the input signal, is the projection distance of the point cloud on the normal vector, is the complex exponential basis function of the Fourier transform.

[0165] The low-frequency component is the overall medium-thick plate, and the high-frequency component may be the error caused by vibration. Set the vibration frequency range to , and use a band-pass filter to extract the vibration signal:

[0166] ;

[0167] Among them, is the vibration error signal after frequency-domain filtering, are the upper and lower limits of the vibration frequency range, is the frequency component of the vibration signal, is the complex exponential basis function of the inverse Fourier transform.

[0168] Use frequency-domain analysis to extract the vibration error component , which represents the vertical vibration between the upper and lower surfaces of the medium-thick plate, and calculate the offset vector , and adjust it along the normal vector direction.

[0169] ;

[0170] For the original coordinates of each pair of point clouds, adjust the point clouds according to the vibration component:

[0171] ;

[0172] Among them, are the corrected point coordinates, is the original point coordinate of the upper surface before correction, is the corresponding original point coordinate of the lower surface before correction, is the offset vector.

[0173] By correcting the error between the corresponding point clouds on the upper and lower surfaces of the medium-thick plate, the influence of the vibration of the medium-thick plate is eliminated.

[0174] S7: Process the point cloud data of the upper and lower surfaces that have been corrected by vibration elimination and error compensation, and remove the isolated points and noise points in the point cloud. By dividing the point cloud of the upper and lower surfaces of the medium-thick plate into several small regions (grids), calculate the average thickness of the points in each grid, and finally obtain the thickness distribution of the medium-thick plate. Remove the grid regions with abnormalities (locally thinner regions or regions with excessive thickness), and calculate the overall thickness of the medium-thick plate.

[0175] ;

[0176] ;

[0177] Among them, is the number of divided grids, is the average thickness within the th grid area, is the number of points within the th grid, is the thickness between the th point cloud and the corresponding point within the grid, is the overall average thickness of the medium - thick plate, is the total number of grid areas after removing abnormal grid areas, is the average thickness of the grid area after removing abnormal grid areas.

[0178] The grid - based average thickness method divides the upper and lower surface point clouds into small areas and calculates the average thickness grid - by - grid , and the average thickness of each grid area corresponds to a color value. Using color gradient to map the thickness value to color, the thicker areas of the medium - thick plate are shown as darker colors, and the relatively thinner areas are shown as lighter colors. Finally, a regional thickness distribution map of the medium - thick plate is generated, which can identify the local variable thickness of the medium - thick plate, facilitate quality assessment and defect location, and simplifies the complexity of global point matching by means of grid division.

[0179] S8: After processing the upper and lower surface point clouds of the medium - thick plate, the point cloud data has been aligned and noise removed, and has high precision and consistency. On this basis, project the point cloud onto a two - dimensional plane and extract edge points through an edge detection algorithm. Use the moving average method to remove local noise from the extracted edge point set to obtain a smooth edge point set. The moving average formula is:

[0180] ;

[0181] Among them, is the moving window size, is the index of the points in the moving window, from to , indicating that the coordinates of all points within the window range will participate in the calculation. is the index of the currently calculated point, is the unsmoothed edge point, is the smoothed edge point.

[0182] Perform sparse region detection on the smoothed edge point set. If the number of points in some regions is small or there are obvious blanks, triangulation can be used to fill in the missing regions. For the complete edge point set, spline interpolation is used to smooth these edge points to obtain a smoother boundary curve. The cubic spline interpolation formula is:

[0183] ;

[0184] where is the value of the interpolation curve at , is the independent variable of the interpolation, is the index of the current interpolation segment, are two adjacent known edge points, is the abscissa of the previous edge point of the current interpolation point, is the abscissa of the next edge point of the current interpolation point, are the spline interpolation coefficients, which are obtained by solving a system of linear equations to ensure that the curve is continuous and smooth at each interpolation point.

[0185] Through the above steps, a smooth, closed and complete medium-thick plate contour line is finally generated, which can accurately reflect the shape characteristics and dimensional information of the medium-thick plate. This contour line not only has high precision in geometric shape, but also has good continuity and smoothness, and can be comprehensively used for subsequent shearing of the irregular shapes at the head and tail of the plate. At the same time, this high-precision contour line also provides reliable basic data for process analysis, optimizes the processing technology, controls the production precision, and further improves the overall quality and efficiency of the medium-thick plate manufacturing process.

[0186] S9: Obtain the width of the medium-thick plate through the medium-thick plate specification model parameters . According to the width of the medium-thick plate, calculate the corresponding width point by point for the head and tail regions of the medium-thick plate contour, construct the width sequence and preliminarily determine the possible cutting positions according to the significant changes in the width. The corresponding width sequence is . To ensure the accuracy of the cutting line, it is necessary to further accurately determine the rough positioning result, and a precise positioning method for the cutting line based on width consistency and edge linearity is proposed. Judge whether the width sequence in the region is consistent. If the width consistency condition (width fluctuation within the specified threshold) is met, it is considered that the region is suitable for cutting. Detect the continuous sub-regions of the width sequence , and set the starting index as . If the following conditions are met, it is considered that the width of the region is consistent:

[0187] ;

[0188] where is the detection range length, is the width consistency threshold. Width consistency detection ensures that the width fluctuation within the cropping area is small, thus avoiding cropping errors caused by local abnormal widths.

[0189] Within the area satisfying width consistency, the left and right edge points of each row are further extracted, and the fitting residuals are calculated through linear fitting to determine whether the edges have good linearity. By comparing the maximum value of the fitting residuals with the set threshold, it is accurately determined whether the left and right edges are straight, thereby further locking the optimal cropping position to ensure that the cropped edges are flat and smooth. This method realizes multiple screenings from width determination to edge shape detection, providing a reliable basis for the final precise positioning of the shearing line and ensuring the cropping quality.

[0190] Embodiment 2

[0191] The present invention discloses a double-sided contour measurement system for medium-thick stainless steel plates. The system is used to implement any one of the methods described above. The system includes: an image acquisition unit, a three-dimensional reconstruction unit, a motion correction unit, a thickness measurement and error correction unit, and a contour detection and shearing positioning unit;

[0192] The image acquisition unit is used to complete the calibration of the double-line array camera to obtain the upper and lower surface images of the medium-thick stainless steel plate;

[0193] The three-dimensional reconstruction unit is used to perform binocular stereo vision three-dimensional reconstruction on the upper and lower surface images of the medium-thick stainless steel plate to generate a point cloud map and convert it for display in the same coordinate system;

[0194] The motion correction unit is used to eliminate lateral rotation and offset according to the fitting of the center lines of the point clouds on the upper and lower surfaces of the medium-thick plate and plane fitting;

[0195] The thickness measurement and error correction unit is used to compare the geometric relationships between the point clouds on the upper and lower surfaces to quantify local vibrations and thickness changes, correct the vibration errors, and obtain the thickness of the medium-thick plate;

[0196] The contour detection and shearing positioning unit is used to extract the edge point set from the point cloud map to obtain the overall contour of the medium-thick plate, and obtain the shearing line position based on the width and edge linearity detection of the medium-thick plate.

[0197] In this embodiment, the rolled medium plate is smoothly laid on the roller table 7. The roller table 7 serves as a transportation and support device to ensure the stable position and flat surface of the medium plate during movement. The surface of the roller table 7 is arranged opposite to the upper and lower surface profilers to ensure a good relative position relationship between the surface of the medium plate and the field of view of the profilers during the scanning process. Each of the upper and lower surface profilers consists of two high-precision linear array cameras. Among them, the two cameras of each profiler 1 are arranged side by side, and the total field of view completely covers the entire transverse width of the medium plate. The fields of view of the two cameras are designed with a partially overlapping area for subsequent binocular stereo vision three-dimensional reconstruction to ensure measurement accuracy and data integrity. When the medium plate moves to a specific position on the roller table 7, synchronous acquisition is achieved through the sensor triggering device on the roller table 7. The upper profiler acquires images of the upper surface of the medium plate through the two cameras above the roller table 7, and the lower profiler acquires images of the lower surface of the medium plate through the two cameras below the roller table 7. To ensure image quality, a uniformly scattering linear light source 3 is equipped on the upper surface, and a high-power linear light source 3 is equipped on the lower surface to provide uniform and sufficient illumination for the upper and lower profilers respectively, avoiding the influence of insufficient light or reflection on image clarity. To reduce the interference of dust, vibration, and oil stains, the profilers 1 are all installed in a closed metal protection box 2. A transparent high-temperature resistant glass window is set inside the protection box, and an automatic cleaning device 4 is equipped to ensure the stability of the long-term operation of the equipment.

[0198] Through the above arrangement and design, the profile detection system can efficiently acquire the upper and lower surface images of the medium plate during the movement of the medium plate. By using an accurate trigger synchronization mechanism and high-quality image acquisition conditions, reliable high-precision original images are provided, which provides solid data support for subsequent three-dimensional reconstruction, point cloud processing, and analysis work such as obtaining the overall profile, thickness distribution, and shear line positioning of the medium plate.

[0199] In this embodiment, the image acquisition unit mainly consists of a profiler 1 integrating two linear array cameras, a synchronous trigger 5, a linear light source 3, etc. When the medium plate moves to a specific position on the roller table 7, synchronous acquisition is achieved through the sensor or trigger device on the roller table 7. The upper profiler acquires images of the upper surface of the medium plate through the two cameras above the roller table 7, and the lower profiler acquires images of the lower surface of the medium plate through the two cameras below the roller table 7.

[0200] In this embodiment, binocular stereo vision is used for three-dimensional reconstruction to generate a point cloud map and convert it for display in the same coordinate system. Through the fitting of the centerlines and planes of the point clouds on the upper and lower surfaces of the medium-thick plate, lateral rotation and offset are eliminated. By comparing the geometric relationships between the point clouds on the upper and lower surfaces, local vibrations and thickness changes are quantified, and the vibration errors are corrected to obtain the thickness of the medium-thick plate. Edge point sets are extracted from the point cloud map to obtain the overall contour of the medium-thick plate, and based on the detection of the width and edge linearity of the medium-thick plate, the position of the shear line is accurately located. Each functional unit works in cooperation to achieve high-precision measurement and positioning of the overall contour, thickness distribution, and shear position of the medium-thick plate.

[0201] The embodiments described above are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for measuring the double-sided profile of a medium and thick stainless steel plate, characterized in that: The method comprises: S1. Complete the calibration of the dual-line array camera to obtain the upper and lower surface images of the stainless steel medium and thick plate; S2, perform binocular stereoscopic vision 3D reconstruction on the upper and lower surface images of the stainless steel medium and thick plate to generate a point cloud map, and convert it into the same coordinate system for display; S3, eliminating lateral rotation and offset according to the centerline fitting and plane fitting of the point cloud of the upper and lower surfaces of the medium and thick plate; S4, compare the geometric relationship between the upper and lower surface point clouds to quantify the local vibration and thickness change, correct the vibration error, and obtain the thickness of the medium and thick plate; S5. Extract edge point sets from the point cloud image to obtain the overall contour of the medium and thick plate, and obtain the shear line position based on the medium and thick plate width and edge linearity detection.

2. The method for measuring the double-sided profile of a medium and thick stainless steel plate according to claim 1, characterized in that: In S1, the dual-line array camera calibration includes: Use a transparent calibration plate, and scan and splice line by line to generate a complete two-dimensional image through the smooth movement of the calibration plate; Extract feature points from the two-dimensional image, combine them with the known physical dimensions of the calibration plate, and use the calibration algorithm to optimize the calculation of the camera's intrinsic and extrinsic parameters; Calibrate the geometric relationship of a dual-line array camera.

3. The method for measuring the double-sided profile of a medium and thick stainless steel plate according to claim 1, characterized in that: In S2, performing binocular stereoscopic three-dimensional reconstruction of the upper and lower surface images of the stainless steel medium and thick plate to generate a point cloud image includes: After acquiring the upper and lower surface images of the stainless steel plate, the images are stereo rectified according to the calibration parameters, and the images are mapped to the same plane to eliminate vertical parallax. The disparity map is calculated using a stereo matching algorithm, and depth information is generated by the position difference of corresponding pixels in the two images; According to the disparity map combined with the internal and external parameters of the camera, the pixel points are projected into the three-dimensional space to generate point cloud data of the medium and thick plate surface.

4. The method for measuring the double-sided profile of a medium and thick stainless steel plate according to claim 1, characterized in that: In S3, eliminating lateral rotation and offset according to centerline fitting and plane fitting of upper and lower surface point clouds of the medium and thick plate includes: Low-density points or abnormal distance points are detected by statistical characteristic analysis, and interference points are removed from the 3D reconstructed point cloud image by combining local curvature calculation method and normal vector consistency analysis, and the vacant areas are repaired by nearest neighbor interpolation. Fit the point cloud planes of the upper and lower surfaces, calculate the center lines respectively, and determine the offset and rotation amount through the relationship between the center lines, and construct the rigid body transformation matrix for correction; Among them, the point cloud plane fitting of the upper and lower surfaces includes: The RANSAC algorithm is used to perform plane fitting on the upper and lower surfaces of the medium and thick plate. The fitting formula is: ； in, is the fitting parameter, is the coordinate of the point in the 3D point cloud; Calculating the centerline includes: Select parallel cross-section point clouds with fixed intervals on the upper and lower surfaces respectively, and calculate the geometric center of each cross section: ； in, The geometric center point of the section is the coordinate average of all point clouds in a certain slice section, i.e., the centroid. The number of point clouds in the current section, Index variable, indicating that the number currently being accumulated is points, ranging from 1 to , is the point cloud coordinate on the cross section; Connect all the cross-section center points to obtain the center lines of the upper and lower surfaces; Determining the offset and rotation between center lines includes: ； in, and are the center point coordinates of the upper and lower surfaces respectively; Constructing the rigid body transformation matrix for correction includes: Based on the offset and rotation calculated above, construct the rigid body transformation matrix: ； in, is the rotation matrix; Perform rigid body transformation on the point clouds of the upper and lower surfaces: ； in, is the original point cloud, To correct the point cloud.

5. The method for measuring double-sided profile of medium and thick stainless steel plates according to claim 1, characterized in that: In S4, the geometric relationship between the upper and lower surface point clouds is compared to quantify the local vibration and thickness change, correct the vibration error, and obtain the thickness of the medium and thick plate, including: Use frequency domain analysis to extract vibration error components, calculate the offset vector, and adjust along the normal vector direction; After eliminating vibration errors, the point clouds of the upper and lower surfaces of the medium and thick plates are divided into several small areas. The average thickness of the point clouds in each area is calculated, and finally the thickness distribution and overall thickness statistics of the medium and thick plates are obtained.

6. The method for measuring double-sided profile of medium and thick stainless steel plates according to claim 5, characterized in that: Use frequency domain analysis to extract vibration error components, calculate the offset vector, and adjust along the normal vector direction including: Use nearest neighbor matching to find the corresponding relationship between each point in the upper and lower surface point clouds, and calculate the vertical distance between each pair of matching points on the upper and lower surfaces: ； in, is the projection distance of the point cloud on the normal vector, and are the corresponding points on the upper and lower surfaces, is the global normal vector; right Perform a fast Fourier transform to extract the main frequency distribution: ； in, is the frequency component of the vibration signal obtained by fast Fourier transform, is the index of the discrete data point, is the total number of data points in the input signal, is the projection distance of the point cloud on the normal vector, is the complex exponential basis function of Fourier transform; According to the main frequency distribution, the vibration frequency range is set to , use a bandpass filter to extract the vibration signal: ； in, is the vibration error signal after frequency domain filtering, are the upper and lower limits of the vibration frequency range, is the frequency component of the vibration signal, is the complex exponential basis function of the inverse Fourier transform; Extracting vibration error components using frequency domain analysis , represents the vertical vibration between the upper and lower surfaces of the medium-thick plate, and calculates the offset vector , adjust along the direction of the normal vector; ； For each pair of original coordinates of the point cloud, adjust the point cloud according to the vibration component: ； in, is the corrected point coordinate, is the coordinate of the original point on the upper surface before correction, is the original point coordinate of the lower surface before correction, is the offset vector; The vibration effect of the medium-thick plate can be eliminated by correcting the error between the corresponding point clouds on the upper and lower surfaces of the medium-thick plate.

7. The method for measuring double-sided profile of a medium and thick stainless steel plate according to claim 6, characterized in that: After eliminating vibration errors, the point cloud of the upper and lower surfaces of the medium and thick plate is divided into several small areas. The average thickness of the point cloud in each area is calculated, and the thickness distribution and overall thickness statistics of the medium and thick plate are finally obtained, including: For each mesh region, calculate the average thickness of all pairs of points in the region: ； in, is the number of points in the grid; The gridded average thickness method is used to divide the upper and lower surface point clouds into small areas, and the average thickness is calculated grid by grid, and finally the regional thickness distribution map of the medium and thick plate is generated; According to the regional thickness distribution diagram of the medium and thick plate, the local thickness variation of the medium and thick plate is identified, and finally the overall thickness of the medium and thick plate is obtained.

8. The method for measuring double-sided profile of medium and thick stainless steel plates according to claim 1, characterized in that: In S5, extracting edge point sets from the point cloud image to obtain the overall contour of the medium and thick plate, and obtaining the shear line position based on the medium and thick plate width and edge linearity detection includes: Obtain the width of the medium and thick plate through the specification model parameters of the medium and thick plate ; According to the width of the plate , calculate the corresponding width of the head and tail areas of the medium and thick plate contour point by point, and construct a width sequence And according to the change of width, the cropping position is preliminarily determined, and the corresponding width sequence is ; Determine whether the width sequence in the region is consistent. If the width consistency condition is met, that is, the width fluctuation is within the specified threshold, the region is considered suitable for cropping. Detect the continuous sub-regions of , if the preset conditions are met, the width of the region is considered to be consistent; the preset conditions are: ； in, is the detection range length, is the width consistency threshold; In the area that meets the width consistency, the left and right edge points of each row are further extracted, and the fitting residual is calculated by linear fitting to determine whether the edge meets the preset linearity requirements. That is, by comparing the maximum value of the fitting residual with the set threshold, it is accurately determined whether the left and right edges are straight, thereby further locking the optimal cropping position to ensure that the cropped edge is flat and smooth.

9. A double-sided profile measurement system for medium and thick stainless steel plates, the system being used to implement the double-sided profile measurement method for medium and thick stainless steel plates according to any one of claims 1 to 8, characterized in that: The system comprises: an image acquisition unit, a three-dimensional reconstruction unit, a motion correction unit, a thickness measurement and error correction unit, and a contour detection and shear positioning unit; The image acquisition unit is used to complete the calibration of the dual-line array camera to acquire the upper and lower surface images of the stainless steel medium and thick plate; The three-dimensional reconstruction unit is used to perform binocular stereoscopic three-dimensional reconstruction of the upper and lower surface images of the stainless steel medium and thick plate to generate a point cloud image, and convert it into the same coordinate system for display; The motion correction unit is used to eliminate lateral rotation and offset according to the centerline fitting and plane fitting of the point cloud of the upper and lower surfaces of the medium-thick plate; The thickness measurement and error correction unit is used to compare the geometric relationship between the upper and lower surface point clouds to quantify local vibration and thickness changes, correct vibration errors, and obtain the thickness of the medium and thick plate; The contour detection and shear positioning unit is used to extract edge point sets from the point cloud image to obtain the overall contour of the medium and thick plate, and to obtain the shear line position based on the medium and thick plate width and edge linearity detection.

Images