A High-Precision Visual Measurement Method for Bending Angle of Metal Shells Based on Point Cloud Registration

By establishing a continuous bending field based on point cloud registration and conducting multiple rounds of candidate parameter searches, the problems of consistency and insufficient accuracy in the measurement of bending angle of metal shells are solved, and high-precision visual measurement results are achieved.

CN122289183BActive Publication Date: 2026-11-10ZHEJIANG ZHENYAO PURIFICATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202610380922.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-03-26
Publication Date
2026-11-10
Estimated Expiration
2046-03-26

AI Technical Summary

Technical Problem

Existing technologies for measuring the bending angle of metal shells suffer from problems such as insufficient consistency of measurement results, susceptibility to operational influences, difficulty in adapting to different workpieces and working conditions, lack of overall point cloud registration constraints in local geometric fitting, and failure to fully utilize height distribution information, resulting in insufficient accuracy.

Method used

Based on the point cloud registration method, a measurement coordinate system with reference marker points as anchors is established. By combining the registration residuals of the predicted point cloud and the measured point cloud with the compact residuals of the bending zone, the bending line position, bending angle and bending zone width are solved simultaneously to form a continuous bending field and perform multiple rounds of candidate parameter search to generate the final predicted point cloud.

Benefits of technology

It achieves high-precision visual measurement of the bending angle of metal shells, improves the stability and consistency of measurement results, reduces sensitivity to local defects and noise, and enhances the adaptability to complex shell shapes and the reliability of measurement.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122289183B_ABST
    Figure CN122289183B_ABST
Patent Text Reader

Abstract

This invention relates to the field of machine vision measurement technology and discloses a high-precision visual measurement method for the bending angle of metal shells based on point cloud registration. A measurement coordinate system is established using reference marker points, and the point cloud after bending is transformed. A reference point set is constructed by filtering boundaries and calculating the reference sampling step size. The average bending height and bending strength are statistically analyzed by width to obtain the initial position of the bending line and the initial bending band width. A continuous bending field is constructed within the candidate range to generate a predicted point cloud, and the registration residual is obtained by comparing the nearest point with the measured point cloud. The compact residual and column height residual are obtained by combining the projection boundary, out-of-band diffusion, and column height comparison. The total residual is constructed and searched multiple times to obtain the optimal bending line, bending angle, and bending band width. The predicted point cloud, registration error, and bending angle are output, reducing segmentation fitting bias, suppressing multi-solution drift, and improving stability and consistency under noise and missing conditions.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of machine vision measurement technology, specifically to a high-precision visual measurement method for the bending angle of metal shells based on point cloud registration. Background Technology

[0002] In the manufacturing process of metal shells, the angular accuracy of a single bend of a rectangular metal plate directly affects subsequent assembly clearances, shell flatness, edge alignment accuracy, and overall structural stability. Therefore, rapid, non-contact, and high-precision measurement of the bending angle is required. Existing technologies commonly employ contact angle gauge measurement, angle measurement based on 2D image edge extraction, and local plane fitting measurement based on 3D point clouds. While contact measurement is simple to implement, it is easily affected by the operating method, probe positioning, and workpiece springback, and is difficult to adapt to online inspection requirements. 2D vision methods typically calculate the included angle by extracting the bending contour, edge lines, or intersection lines. They are sensitive to lighting, reflections, occlusion, and edge burrs, and exhibit poor measurement stability when the bending area has a smooth transition, non-ideal sharp corners, or local deformation. While 3D point cloud methods can obtain richer surface geometric information, existing solutions often involve selecting segmented regions, fitting the two side planes separately, and then inferring the bending angle from the plane normal angle. This typically requires a clearly defined bending area location and a fixed or manually adjustable transition area width.

[0003] However, in actual bending production, the position of the bending line, the width of the bending transition zone, and the distribution of the point cloud on the surface after bending all change with the plate thickness, mold condition, pressing conditions, and material springback, resulting in the following shortcomings of existing technologies: First, relying on manual selection of the two planar areas or preset bending zone width makes it difficult to accurately adapt to different workpieces and working conditions, resulting in insufficient consistency of measurement results; Second, when only local geometric fitting is used without overall point cloud registration constraints, it is easily affected by local outliers, uneven sampling, missing columns, and surface reflection noise, causing deviations in angle calculation; Third, existing methods usually treat the bending area as an abrupt boundary, failing to fully characterize the continuous transition characteristics inside the bending zone, making it difficult to simultaneously consider bending line positioning, bending zone width identification, and angle calculation; Fourth, for single-bend rectangular metal plates, the height distribution information in the width direction after bending is not fully utilized, resulting in insufficient accuracy of existing solutions in high-precision visual measurement scenarios.

[0004] Therefore, this proposal aims to propose a high-precision visual measurement method for the bending angle of metal shells based on point cloud registration. This scheme establishes a measurement coordinate system with reference marker points as anchors based on the point cloud data after the metal shell is bent, and models the bending deformation as a bending field that changes continuously along the width direction. Then, by combining the registration residual between the predicted point cloud and the measured point cloud, the bending band compaction residual and the column height residual, the bending line position, bending angle and bending band width are solved simultaneously, and the final bending angle and registration error are output. Summary of the Invention

[0005] This invention provides a high-precision visual measurement method for the bending angle of metal shells based on point cloud registration, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a high-precision visual measurement method for the bending angle of metal shells based on point cloud registration, comprising: Establish a measurement coordinate system, and write the theoretical length of the rectangular metal plate, the theoretical width of the rectangular metal plate, the point cloud of the measured surface after bending, the origin reference marker, the reference marker in the length direction, and the reference marker in the width direction. Complete the point cloud coordinate transformation, boundary filtering, reference sampling step size calculation, and reference point set construction. Divide the measurement width after bending into columns along the direction of the measurement width after bending, assign each measured point to the corresponding measurement width after bending column, obtain the measured average folding height and bending strength of each measurement width after bending column, and obtain the initial position of the bending line and the initial bending band width. Write the range of parameters to be determined, including candidate bend line position, candidate bend angle, and candidate bend band width. Construct a continuous local bend angle distribution based on the positional difference between the width coordinate before bending and the candidate bend line position to form a continuous bend field. Perform a continuous bending transformation on the reference point set to generate a predicted point cloud; Establish the nearest point correspondence between the predicted points and the measured points to obtain the point cloud registration residual; The compactness residual and column height residual of the bending strip are obtained by taking the left and right projection boundaries of the bending strip in the measurement coordinates after bending, the outward diffusion distance of the measurement width column after bending, the predicted average folding height of the measurement width column after bending, and the measured average folding height of the measurement width column after bending. The total residual is constructed based on the point cloud registration residual, the bending band compaction residual, and the column height residual. The final optimal bending line position, the final optimal bending angle, and the final optimal bending band width are obtained through multiple rounds of candidate parameter search. Substituting the final optimal bend line position, final optimal bend angle, and final optimal bend band width into the continuous bend transformation and point cloud registration residual calculation process, we obtain the final predicted point cloud, final registration error, and final bend angle.

[0007] Optionally, the establishment of the measurement coordinate system, including the input of the theoretical length and width of the rectangular metal plate, the point cloud of the measured surface after bending, the origin reference marker, the length direction reference marker, and the width direction reference marker, and the completion of point cloud coordinate transformation, boundary filtering, reference sampling step size calculation, and reference point set construction, specifically includes: The center of the measurement surface of the rectangular metal plate before bending is set as the origin of the coordinate system. The length direction of the rectangular metal plate is set as the positive direction of the length direction of the coordinate system. The width direction of the rectangular metal plate is set as the positive direction of the width direction of the coordinate system. The normal of the plate surface is set as the positive normal of the coordinate system. The measurement surface area before bending is limited to the rectangular area enclosed by half of the theoretical length of the rectangular metal plate in the length direction and half of the theoretical width of the rectangular metal plate in the width direction. Write the factory-fixed 3D point cloud device coordinate system of the vision system into the original coordinate system of the vision system, and write the point cloud of the bent surface obtained by the vision system. At the same time, write the three-dimensional coordinates of the origin reference mark point, the length direction reference mark point, and the width direction reference mark point in the original coordinate system of the vision system onto the fixture. Construct a unit vector in the length direction according to the direction from the origin reference point to the length reference point. Construct a unit vector in the width direction by removing the projection of the width reference point onto the unit vector in the length direction. Construct a unit vector in the normal direction according to the cross product direction of the unit vector in the length direction and the unit vector in the width direction. When the origin reference mark coincides with the length direction reference mark, or when the length of the width direction auxiliary vector after removing the projection is zero, the final optimal bending angle, final bending angle, final optimal bending line position, final optimal bending band width, and final registration error are marked as invalid, and the measurement calculation is terminated. For each origin point, calculate its projection relative to the origin reference point on the unit vectors of the length direction, width direction, and normal direction, and obtain the length direction coordinates, width direction coordinates, and normal coordinates of each origin point in the measurement coordinate system. The transformed points whose length direction coordinates are within ±1 / 2 of the theoretical length of the rectangular metal plate and whose width direction coordinates are within ±3 / 2 of the theoretical width of the rectangular metal plate are retained. The retained points are renumbered and a measured point cloud is formed. Under the condition that the total number of measured points after retention is not less than one, take the square root of the product of the theoretical length and the theoretical width of the rectangular metal plate divided by the total number of measured points after retention, and compare the square root result with half of the theoretical width of the rectangular metal plate, and take the smaller value as the reference sampling step size. Divide the theoretical length of the rectangular metal plate by the reference sampling step size and the theoretical width of the rectangular metal plate by the reference sampling step size, round up the results, and increment them by one to obtain the total number of sampling points in the length direction and the total number of sampling points in the width direction before bending. Then, divide the theoretical length of the rectangular metal plate by the total number of sampling points in the length direction minus one, and divide the theoretical width of the rectangular metal plate by the total number of sampling points in the width direction before bending minus one to obtain the sampling interval in the length direction and the sampling interval in the width direction before bending. Write these intervals into the reference point set according to the sampling sequence number in the length direction and the sampling sequence number in the width direction.

[0008] Optionally, the step of dividing the measured width into columns along the direction of the measured width after bending, assigning each measured point to the corresponding column of measured width after bending, obtaining the measured average folding height and bending strength of each column of measured width after bending, and obtaining the initial position of the bending line and the initial bending band width, specifically includes: The result of dividing three times the theoretical width of the rectangular metal plate by the reference sampling step size is rounded up and then incremented by one to obtain the total number of discrete columns in the direction of the measured width after bending. The spacing between the measured width columns after bending is obtained by dividing three times the theoretical width of the rectangular metal plate by the total number of discrete columns minus one. The starting width coordinate is taken as three-half of the theoretical width of the rectangular metal plate, and written into the center of each measured width column after bending according to the spacing between the measured width columns after bending. For each measured point, compare its width direction coordinate with the center of all the measured width columns after bending. Write the measured width column after bending with the smallest distance as the corresponding nearest measured width column after bending. When multiple measured width columns after bending reach the minimum distance at the same time, write the measured width column after bending with the smaller column number. For each column of measured width after bending, if there are measured points that should be included in the current column, the absolute values ​​of the normal coordinates of all measured points in the current column are averaged to obtain the average folded height of the current column. If there are no measured points that should be included in the current column, the average folded height of the non-empty column closest to the center of the current column is written into the current column. If only the left side has the nearest non-empty column, the average folded height of the left nearest non-empty column is written. If only the right side has the nearest non-empty column, the average folded height of the right nearest non-empty column is written. If the distances from the nearest non-empty columns on both sides to the center of the current column are the same, the arithmetic mean of the average folded heights of the nearest non-empty columns on both sides is written. Calculate the bending strength for all measured width columns after bending. The bending strength of the first measured width column after bending is written as the absolute value of the difference between the measured average folded height of the second measured width column after bending and the measured average folded height of the first measured width column after bending. The bending strength of the last measured width column after bending is written as the absolute value of the difference between the measured average folded height of the last measured width column after bending and the measured average folded height of the second-to-last measured width column after bending. The bending strength of the measured width columns after bending in the middle is written as the absolute value of the difference between the measured average folded height of the next measured width column after bending and the measured average folded height of the previous measured width column after bending. Take the maximum bending strength from the bending width measurement columns excluding the first and last bending width measurement columns to obtain the maximum bending strength value; when the maximum bending strength value is zero, mark the final optimal bending angle, final bending angle, final optimal bending line position, final optimal bending band width, and final registration error as invalid, and terminate the measurement calculation. Write the column corresponding to the maximum bending strength after bending as the peak column; when the bending strength of multiple columns of measured widths after bending reaches the maximum value at the same time, prioritize writing the column of measured widths after bending with the smaller center coordinates. If the center coordinates of the columns of measured widths after bending are still the same, write the column of measured widths after bending with the smaller column number; then restrict the center of the column of measured widths after bending corresponding to the peak column to within ±1 / 2 of the theoretical width of the rectangular metal plate to obtain the initial position of the bending line; Half of the bending strength of the peak column is written as the half-peak threshold. Continuously search to the left from the peak column for the smallest consecutive bend width measurement column number for all bend strengths not lower than the half-peak threshold. Continuously search to the right from the peak column for the largest consecutive bend width measurement column number for all bend strengths not lower than the half-peak threshold. Multiply the difference between the largest and smallest consecutive bend width measurement column numbers by the bend width measurement column spacing to obtain the initial bending band width. When the initial bending band width is less than the reference sampling step size, it is written as the reference sampling step size; when the initial bending band width is greater than the theoretical width of the rectangular metal plate, it is written as the theoretical width of the rectangular metal plate.

[0009] Optionally, the step of writing the range of parameters to be determined, including the candidate bend line position, candidate bend angle, and candidate bend band width, and constructing a continuous local bend angle distribution based on the positional difference between the width coordinate before bending and the candidate bend line position to form a continuous bend field, specifically includes: Write the candidate bend line position, candidate bend angle and candidate bend strip width as parameters to be determined, and limit the candidate bend angle between zero and pi, and limit the candidate bend strip width between the reference sampling step size and the theoretical width of the rectangular metal plate. For each pre-bend width coordinate, calculate the position difference between the pre-bend width coordinate and the candidate bend line position to obtain the signed distance from each pre-bend width coordinate to the candidate bend line position; According to the candidate bend width, a continuous local bend angle distribution is written in the width direction before bending. Specifically, for each signed distance, a corresponding continuous local bend angle is written. When the signed distance is outside the left boundary of the bend defined by the candidate bend line position and the candidate bend width, a zero angle is written. When the signed distance is inside the bend defined by the candidate bend line position and the candidate bend width, a continuous local bend angle is written in a linearly increasing manner from the left boundary of the bend defined by the candidate bend line position and the candidate bend width to the right boundary of the bend defined by the candidate bend line position and the candidate bend width. When the signed distance is outside the right boundary of the bend defined by the candidate bend line position and the candidate bend width, a candidate bend angle is written.

[0010] Optionally, performing a continuous bending transformation on the reference point set to generate a predicted point cloud specifically includes: For each reference point, perform continuous local bend angle rotation around a rotating straight line whose width direction coordinate is equal to the candidate bend line position and whose normal coordinate is equal to zero. Directly write the length direction coordinate of the reference point as the length direction coordinate of the predicted point. Add the product of the signed distance and the cosine value of the corresponding continuous local bend angle to the candidate bend line position and write it as the width direction coordinate of the predicted point. Write the product of the signed distance and the sine value of the corresponding continuous local bend angle as the normal coordinate of the predicted point to generate the corresponding predicted point. A continuous bending transformation is performed on all reference points to form a predicted point cloud.

[0011] Optionally, establishing the nearest point correspondence between the predicted points and the measured points to obtain the point cloud registration residual specifically includes: For each predicted point, search for the measured point cloud that is closest to the predicted point and write the measured point with the smallest distance as the unique nearest corresponding point of the current predicted point; when the distance between multiple measured points and the same predicted point reaches the minimum value at the same time, write the measured point with the smaller index. The point cloud registration residual is obtained by summing the squared distances between all predicted points and their unique nearest corresponding points, and then averaging the sum using the number of reference points in the reference point set.

[0012] Optionally, the step of obtaining the compactness residual and column height residual of the bent strip based on the left and right projected boundaries of the bent strip in the post-bending measurement coordinates, the outward diffusion distance of the post-bending measurement width column, the predicted average folding height of the post-bending measurement width column, and the measured average folding height of the post-bending measurement width column specifically includes: For each signed distance located inside the bend, the projection width coordinates of the bend in the measurement coordinates after bending are calculated point by point according to the combination relationship between the candidate bend line position, the signed distance and the cosine value of the continuous local bend angle corresponding to the signed distance. The minimum value is written as the left projection boundary and the maximum value is written as the right projection boundary. For each column center of the measured width after bending, compare the positional relationship between the current column center and the left and right projection boundaries. When the current column center falls between the left and right projection boundaries, write the out-of-band diffusion distance as zero; when the current column center is to the left of the left projection boundary, write the difference between the left projection boundary and the current column center as the out-of-band diffusion distance; when the current column center is to the right of the right projection boundary, write the difference between the current column center and the right projection boundary as the out-of-band diffusion distance. The product of the square of the out-of-band diffusion distance of each of the measured width columns after bending and the square of the bending strength of the same measured width column after bending is accumulated, and then normalized by the total number of discrete columns in the measured width direction after bending and the square of the theoretical width of the rectangular metal plate to obtain the compact residual of the bending band. For each prediction point, compare its width direction coordinate with the center of all the bend measurement width columns, and write the bend measurement width column with the smallest distance as the nearest bend measurement width column of the current prediction point; when multiple bend measurement width columns reach the minimum distance at the same time, write the bend measurement width column with the smaller column number. For each column of post-bend measurement width, if there is a predicted point that should be included in the current post-bend measurement width column, the absolute values ​​of the normal coordinates of all predicted points in the current post-bend measurement width column are averaged to obtain the predicted average folding height of the current post-bend measurement width column. If there is no predicted point that should be included in the current post-bend measurement width column, the predicted average folding height of the non-empty post-bend measurement width column closest to the center of the current post-bend measurement width column is written into the current post-bend measurement width column. If there is only a nearest non-empty post-bend measurement width column on the left, the predicted average folding height of the nearest non-empty post-bend measurement width column on the left is written. If there is only a nearest non-empty post-bend measurement width column on the right, the predicted average folding height of the nearest non-empty post-bend measurement width column on the right is written. If the distances from the nearest non-empty post-bend measurement width columns on both the left and right sides to the center of the current post-bend measurement width column are the same, the arithmetic mean of the predicted average folding heights of the nearest non-empty post-bend measurement width columns on both the left and right sides is written. The average of the squares of the differences between the measured average folded height and the predicted average folded height of each of the measured widths after bending in all the bends is summed to obtain the column height residual.

[0013] Optionally, the total residual is constructed based on the point cloud registration residual, the bend band compaction residual, and the column height residual. The final optimal bend line position, the final optimal bend angle, and the final optimal bend band width are obtained through multiple rounds of candidate parameter search. Specifically, this includes: The total residual is obtained by adding the point cloud registration residual, the bending band compaction residual, and the column height residual. One-quarter of the reference sampling step length is written as the search interval for the first round bend line position and the search interval for the first round bend band width, respectively. The result of dividing pi by 180 is written as the search interval for the first round bend angle. Based on the distance from the initial position of the bending line to the negative half-width position of the rectangular metal plate and the distance from the positive half-width position of the rectangular metal plate to the initial position of the bending line, divide each distance by the maximum integer number obtained after the first round of bending line position search interval. Generate candidate values ​​for the first round of bending line positions by offsetting the initial position of the bending line by integer steps. Limit candidate values ​​that exceed the range of the positive and negative half-width of the rectangular metal plate to the range of the positive and negative half-width of the rectangular metal plate. Then delete duplicate values ​​to form the first round of bending line position candidate set. Based on the distance from the initial bending strip width to the reference sampling step length and the distance from the theoretical width of the rectangular metal plate to the initial bending strip width, respectively, divide by the maximum integer number of steps obtained after the first round of bending strip width search interval, generate candidate values ​​for the first round of bending strip width by offsetting by integer steps around the initial bending strip width, and restrict candidate values ​​that exceed the range from the reference sampling step length to the theoretical width of the rectangular metal plate to the range from the reference sampling step length to the theoretical width of the rectangular metal plate, and then delete duplicate values ​​to form the candidate set for the first round of bending strip width; Take values ​​between zero and pi according to the first round of bending angle search interval, delete zero and pi, and form the first round of bending angle candidate set; The candidate sets for the first-round bend line position, the first-round bend band width, and the first-round bend angle are combined using Cartesian products to form a first-round candidate parameter set. The total residual is calculated for each first-round candidate parameter set. The first-round candidate parameter set with the smallest total residual is selected as the center bend line position, center bend angle, and center bend band width obtained in the first round search. When the total residuals of multiple first-round candidate parameter sets simultaneously reach their minimum values, the first-round candidate parameter set with the smaller point cloud registration residual is selected. When the point cloud registration residuals are the same, the first-round candidate parameter set with the smaller bend band compaction residual is selected. When both the point cloud registration residual and the bend band compaction residual are the same, the first-round candidate parameter set with the smaller column height residual is selected. If they are still tied, the first-round candidate parameter set with the smaller bend band width is selected. The search intervals for the first round of bend line position, bend band width, and bend angle are halved to obtain the search intervals for the second round of bend line position, bend band width, and bend angle. Candidate values ​​for the second round of bend line positions are generated by offsetting the center bend line position obtained in the first round by integer steps of -2, -1, 0, +1, and +2. Candidate values ​​exceeding ±1 / 2 of the theoretical width of the rectangular metal plate are restricted to within ±1 / 2 of the theoretical width of the rectangular metal plate. Duplicate values ​​are removed to form the candidate set for the second round of bend line positions. The width of the center bend obtained in the first round of search is offset by integer steps of -2, -1, zero, +1, and +2 to generate candidate values ​​for the width of the second round of bend. Candidate values ​​that exceed the range of the reference sampling step size to the theoretical width of the rectangular metal plate are restricted to the range of the reference sampling step size to the theoretical width of the rectangular metal plate. After deleting duplicate values, the second round of candidate values ​​for the width of the bend is formed. The center bend angle obtained in the first round of search is offset by integer steps of -2, -1, zero, +1, and +2 to generate candidate values ​​for the second round of bend. Candidate values ​​that do not belong to zero and pi are deleted to form the second round of candidate values ​​for the bend. The candidate sets for the second round of bending line positions, bending band widths, and bending angles are combined using Cartesian products to form a second-round candidate parameter set. The total residual is calculated for each candidate parameter set in this set. The candidate parameter set with the smallest total residual is selected as the center bending line position, center bending angle, and center bending band width obtained in the second round of search. When the total residuals of multiple candidate parameter sets simultaneously reach their minimum values, the candidate parameter set with the smaller point cloud registration residual is selected. When the point cloud registration residuals are the same, the candidate parameter set with the smaller bending band compaction residual is selected. When both the point cloud registration residual and the bending band compaction residual are the same, the candidate parameter set with the smaller column height residual is selected. If they are still parallel, the candidate parameter set with the smaller bending band width is selected. The search intervals for the second round of bend line position, bend strip width, and bend angle are halved to obtain the search intervals for the third round of bend line position, bend strip width, and bend angle. Candidate values ​​for the third round of bend line position are generated by offsetting the center bend line position obtained in the second round of search by integer steps of -2, -1, 0, +1, and +2. Candidate values ​​exceeding the theoretical width plus or minus half of the rectangular metal plate are restricted to the range of the theoretical width plus or minus half of the rectangular metal plate. After deleting duplicate values, the candidate set for the third round of bend line position is formed. The third round of candidate values ​​for the bend width is generated by offsetting the center bend width obtained in the second round of search by integer steps of -2, -1, zero, +1, and +2. Candidate values ​​that exceed the range of the reference sampling step size to the theoretical width of the rectangular metal plate are restricted to the range of the reference sampling step size to the theoretical width of the rectangular metal plate. After deleting duplicate values, the third round of candidate values ​​for the bend width is formed. The third round of candidate values ​​for the bend angle is generated by offsetting the center bend angle obtained in the second round of search by integer steps of -2, -1, zero, +1, and +2. Candidate values ​​that do not belong to the range between zero and pi are deleted, forming the third round of candidate values ​​for the bend angle. The candidate sets for the third-round bend line position, the third-round bend band width, and the third-round bend angle are combined using Cartesian products to form a third-round candidate parameter set. The total residual is calculated for each third-round candidate parameter set. The third-round candidate parameter set with the smallest total residual is selected as the final optimal bend line position, final optimal bend angle, and final optimal bend band width. When the total residuals of multiple third-round candidate parameter sets simultaneously reach their minimum values, the third-round candidate parameter set with the smaller point cloud registration residual is selected. When the point cloud registration residuals are the same, the third-round candidate parameter set with the smaller bend band compaction residual is selected. When both the point cloud registration residual and the bend band compaction residual are the same, the third-round candidate parameter set with the smaller column height residual is selected. If they are still tied, the third-round candidate parameter set with the smaller bend band width is selected.

[0014] Optionally, the step of substituting the final optimal bend line position, final optimal bend angle, and final optimal bend band width into the continuous bend transformation and point cloud registration residual calculation process to obtain the final predicted point cloud, final registration error, and final bend angle specifically includes: Substitute the final optimal bend line position, the final optimal bend angle, and the final optimal bend strip width into the continuous bend transformation, and re-perform the transformation on the reference point set to generate the final predicted point cloud. Substituting the final optimal bend line position, final optimal bend angle, and final optimal bend band width into the point cloud registration residual calculation process, we obtain the final registration error. The final optimal bending angle is converted to degrees according to the conversion relationship between radians and degrees, and the final bending angle is output.

[0015] The present invention has the following beneficial effects: 1. A measurement coordinate system is constructed using the origin reference point, length direction reference point, and width direction reference point. The point cloud is transformed from equipment coordinates to a coordinate expression consistent with the workpiece geometry. Boundary filtering and reference sampling step size calculation are performed simultaneously to form a reference point set. The physical implementation of the measurement reference is coupled with the unified coordinate reference for point cloud processing in advance, so that the solution of bending line position, bending band width, and bending angle are completed in a closed loop within the same coordinate frame. At the same time, an invalidation judgment for coordinate construction anomalies is set to avoid drift of the entire process due to reference point anomalies.

[0016] 2. Establish discrete width columns along the measured width direction after bending, assign the measured points to the nearest width column, calculate the average folding height of each column, and obtain the bending strength based on the changes between columns, thereby obtaining the initial position of the bending line and the initial bending band width. Using column statistics to replace a single profile or a small number of feature points, and utilizing the overall trend in the width direction to locate the bending line and estimate the bending band width, can effectively resist local distortions caused by local defects, holes, burrs, or obstructions; at the same time, height compensation is applied to empty columns using the nearest non-empty column, making the bending strength sequence more continuous.

[0017] 3. Within the given range of candidate bend line positions, candidate bend angles, and candidate bend band widths, a continuous local bend angle distribution is constructed using the signed distance from the pre-bend width coordinates to the candidate bend line position as an index, forming a continuous bend field. The bend process is expressed as a continuous field model where the angle gradually changes within the bend band and remains constant outside the band. This gives the generation of the predicted point cloud a clear physical meaning, representing the transition zone within the bend band while avoiding simplification of the bend as an ideal acute-angled broken line.

[0018] 4. Perform continuous bending transformation on the reference point set to generate a predicted point cloud corresponding to the candidate parameters. Using a structured reference point set as the theoretical geometric carrier before bending, the positions of candidate bending lines, bending angles, and bending band widths are uniformly mapped to the predicted point cloud, thus transforming the parameter solving problem into a consistency evaluation between the predicted and measured point clouds. This scheme avoids segmentation decisions that strongly depend on the measured point cloud, reducing structural errors caused by segmentation threshold selection, noise distribution, and missing regions. Simultaneously, the predicted point cloud, as a holistic assumption of a complete surface shape, allows subsequent residuals to cover the entire region, improving the global consistency and interpretability of the angle solution.

[0019] 5. Establish the nearest-point correspondence between predicted and measured points to form the point cloud registration residual. Define the registration residual as the overall matching error across all reference points, and provide deterministic rules for the parallel correspondence to ensure reproducibility. This residual incorporates overall surface consistency into the evaluation, providing a stable gradient direction when the bending line position, bending angle, and bending band width are coupled, reducing situations where the angle appears reasonable but the overall surface shape is offset. It is less sensitive to point cloud density unevenness, local missing points, and outliers caused by reflections, and the residual form naturally matches the continuous bending field, providing reliable criteria for subsequent multi-round candidate searches.

[0020] 6. On the one hand, by measuring the left and right projection boundaries of the bending strip in the coordinate system after bending, the out-of-band diffusion distance of the center of each width column is calculated, forming a compact residual for the bending strip. On the other hand, the predicted average folding height of each width column is calculated and compared with the measured average folding height to form a column height residual. Introducing the geometric constraint of out-of-band diffusion suppresses excessively wide or drifting projections of the bending strip caused by candidate parameters. Simultaneously, the column height consistency constraint ensures that the predicted height distribution is consistent with the measured height in a statistical sense, resulting in a more robust evaluation against noise and missing values. This joint residual incorporates both the rationality of the positional boundary and the consistency of the height distribution, making the solution for the bending line position and the bending strip width more identifiable and reducing multiple solutions and misunderstandings. In practical applications, it can significantly reduce angular drift caused by misregistration of the bending transition zone, improving adaptability to complex shell shapes.

[0021] 7. The point cloud registration residual, the bending strip compaction residual, and the column height residual are combined into a total residual. Multiple rounds of candidate parameter search are then used to obtain the final optimal bending line position, the final optimal bending angle, and the final optimal bending strip width. A multi-index joint, round-by-round convergence solution strategy is adopted: first, a coarser granularity is used to cover the parameter space to avoid getting trapped in local optima; then, the search interval is gradually reduced round by round to refine the result near the center, ensuring both global accuracy and precision. Simultaneously, residuals are compared in a fixed order when they are parallel, forming a stable decision rule. This strategy is less dependent on initial values, more robust to noise and missing values, and will not select parameters that are generally unreasonable due to the occasional small error of a single term. It reduces the difficulty of parameter tuning, improves the consistency of the solution under different workpieces and different acquisition conditions, and is particularly suitable for the stability and repeatability requirements of production line environments.

[0022] 8. Substitute the final optimal bending line position, final optimal bending angle, and final optimal bending band width into the continuous bending transformation and registration residual calculation to obtain the final predicted point cloud, final registration error, and final bending angle. Substitute the resulting parameters back into the prediction and evaluation chain to form a verifiable final predicted point cloud and quantified registration error. This ensures that the output not only provides angle values ​​but also quality indicators consistent with the measured point cloud, facilitating online judgment, traceability, and process closure. This solution provides production-oriented result reliability, can be used to screen out unreliable measurements caused by acquisition anomalies, benchmark anomalies, or insufficient point cloud quality, and supports subsequent visual comparison between the predicted and measured point clouds to help locate process and acquisition problems, thereby improving the usability and maintainability of the entire measurement solution in engineering implementation. Attached Figure Description

[0023] Figure 1 This is a schematic diagram of the process of the present invention.

[0024] Figure 2 This is a schematic diagram of the measurement coordinate system of the present invention.

[0025] Figure 3 This is a schematic diagram of the original coordinate system of the vision system of the present invention.

[0026] Figure 4 This is a schematic diagram of the process for constructing the bending strength sequence and obtaining initial parameters according to the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments 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, and 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.

[0028] Example, refer to Figure 1 A high-precision visual measurement method for the bending angle of metal shells based on point cloud registration includes: Establish a measurement coordinate system, and write the theoretical length of the rectangular metal plate, the theoretical width of the rectangular metal plate, the point cloud of the measured surface after bending, the origin reference marker, the reference marker in the length direction, and the reference marker in the width direction. Complete the point cloud coordinate transformation, boundary filtering, reference sampling step size calculation, and reference point set construction. Divide the measurement width after bending into columns along the direction of the measurement width after bending, assign each measured point to the corresponding measurement width after bending column, obtain the measured average folding height and bending strength of each measurement width after bending column, and obtain the initial position of the bending line and the initial bending band width. Write the range of parameters to be determined, including candidate bend line position, candidate bend angle, and candidate bend band width. Construct a continuous local bend angle distribution based on the positional difference between the width coordinate before bending and the candidate bend line position to form a continuous bend field. Perform a continuous bending transformation on the reference point set to generate a predicted point cloud; Establish the nearest point correspondence between the predicted points and the measured points to obtain the point cloud registration residual; The compactness residual and column height residual of the bending strip are obtained by taking the left and right projection boundaries of the bending strip in the measurement coordinates after bending, the outward diffusion distance of the measurement width column after bending, the predicted average folding height of the measurement width column after bending, and the measured average folding height of the measurement width column after bending. The total residual is constructed based on the point cloud registration residual, the bending band compaction residual, and the column height residual. The final optimal bending line position, the final optimal bending angle, and the final optimal bending band width are obtained through multiple rounds of candidate parameter search. Substituting the final optimal bend line position, final optimal bend angle, and final optimal bend band width into the continuous bend transformation and point cloud registration residual calculation process, we obtain the final predicted point cloud, final registration error, and final bend angle.

[0029] By first establishing a measurement coordinate system and uniformly transforming the point cloud of the measured surface after bending to a coordinate expression consistent with the workpiece geometry, and then dividing the width along the measured width direction after bending and statistically analyzing the average folding height and bending strength of each column, the initial position of the bending line and the initial bending band width are obtained. This solves the problems of unstable bending line position and sensitivity to occlusion and uneven sampling caused by relying only on local cross-sections or a few feature points. By constructing a continuous local bending angle distribution within the candidate bending line position, candidate bending angle, and candidate bending band width range, and performing continuous bending transformation on the reference point set to generate a predicted point cloud, and then establishing the nearest point correspondence between the predicted points and the measured points to achieve registration. The residual method addresses the common problems in existing technologies, such as the difficulty in unifying segmentation thresholds and the introduction of fitting bias by noise points in the "segmentation-then-fitting" process. Furthermore, it constructs compact residuals for the bending zone and column height residuals by comparing the projection boundary of the bending zone, the out-of-band diffusion distance, and column height. These residuals, together with the registration residuals, form the total residuals for multi-round candidate searches. This solves the problem of multiple solutions and drift caused by the coupling of bending angle, bending line position, and bending zone width. The output not only provides the final bending angle but also the final predicted point cloud and the final registration error, facilitating quantitative verification of measurement reliability. This results in more stable and consistent results under complex acquisition conditions such as noise, missing data, and reflection.

[0030] The establishment of the measurement coordinate system involves inputting the theoretical length and width of the rectangular metal plate, the point cloud of the measured surface after bending, the origin reference marker, the length direction reference marker, and the width direction reference marker. This includes point cloud coordinate transformation, boundary filtering, reference sampling step size calculation, and reference point set construction. Specifically, this includes: The center of the measurement surface of the rectangular metal plate before bending is set as the origin of the coordinate system. The length direction of the rectangular metal plate is set as the positive direction of the length direction of the coordinate system. The width direction of the rectangular metal plate is set as the positive direction of the width direction of the coordinate system. The normal direction of the plate surface is set as the positive normal direction of the coordinate system. The measurement surface area before bending is limited to the rectangular area enclosed by half of the theoretical length of the rectangular metal plate in the length direction and half of the theoretical width of the rectangular metal plate in the width direction. Write the factory-fixed 3D point cloud device coordinate system of the vision system into the original coordinate system of the vision system, and write the point cloud of the bent surface obtained by the vision system. At the same time, write the three-dimensional coordinates of the origin reference mark point, the length direction reference mark point, and the width direction reference mark point in the original coordinate system of the vision system onto the fixture. Construct a unit vector in the length direction according to the direction from the origin reference point to the length reference point. Construct a unit vector in the width direction by removing the projection of the width reference point onto the unit vector in the length direction. Construct a unit vector in the normal direction according to the cross product direction of the unit vector in the length direction and the unit vector in the width direction. When the origin reference mark coincides with the length direction reference mark, or when the length of the width direction auxiliary vector after removing the projection is zero, the final optimal bending angle, final bending angle, final optimal bending line position, final optimal bending band width, and final registration error are marked as invalid, and the measurement calculation is terminated. For each origin point, calculate its projection relative to the origin reference point on the unit vectors of the length direction, width direction, and normal direction, and obtain the length direction coordinates, width direction coordinates, and normal coordinates of each origin point in the measurement coordinate system. The transformed points whose length direction coordinates are within ±1 / 2 of the theoretical length of the rectangular metal plate and whose width direction coordinates are within ±3 / 2 of the theoretical width of the rectangular metal plate are retained. The retained points are renumbered and a measured point cloud is formed. Under the condition that the total number of measured points after retention is not less than one, take the square root of the product of the theoretical length and the theoretical width of the rectangular metal plate divided by the total number of measured points after retention, and compare the square root result with half of the theoretical width of the rectangular metal plate, and take the smaller value as the reference sampling step size. Divide the theoretical length of the rectangular metal plate by the reference sampling step size and the theoretical width of the rectangular metal plate by the reference sampling step size, round up the results, and increment them by one to obtain the total number of sampling points in the length direction and the total number of sampling points in the width direction before bending. Then, divide the theoretical length of the rectangular metal plate by the total number of sampling points in the length direction minus one, and divide the theoretical width of the rectangular metal plate by the total number of sampling points in the width direction before bending minus one to obtain the sampling interval in the length direction and the sampling interval in the width direction before bending. Write these intervals into the reference point set according to the sampling sequence number in the length direction and the sampling sequence number in the width direction.

[0031] Reference Figure 2 and Figure 3 The center of the measuring surface of the rectangular metal plate before bending is set as the origin of the coordinate system. Set the length direction of the rectangular plate to In the positive axis direction, set the width direction of the rectangular plate as... In the positive direction of the axis, set the normal of the plate surface to The positive axis direction forms the measurement coordinate system; The length of the rectangular plate is denoted as . The width is denoted as The surface area measured before bending is written as: ;in, The surface area to be measured before bending; The factory-defined 3D point cloud device coordinate system of the vision system is denoted as the original coordinate system of the vision system. Among them, the origin Find the position of the principal optical center of the imaging lens of the vision system in the three-dimensional reconstruction coordinates; The positive direction of the lens principal optical axis is taken from the direction of the vision system pointing towards the metal plate being measured. The positive axis is taken as the horizontal direction to the right of the imaging plane; The positive direction of the axis is determined by the right-hand rule, and the vertical direction of the imaging plane is taken to be upward; The point cloud of the bent measured surface in the original coordinate system of the vision system, acquired and identified by the vision system, is written as: Set the fixture to correspond to the origin of the measurement coordinate system. Origin reference point , To the reference marker point and To the reference marker point All three points are written in the original coordinate system of the visual system; among them, This is the original point cloud set in the original coordinate system of the visual system; For the original point cloud collection The Middle One original point; The original point number; , , They represent the first Original point The three coordinate components in the original coordinate system of the visual system; For the original point cloud collection Total number of original points in the middle; The three-dimensional coordinates of the origin reference mark point in the fixture in the original coordinate system of the vision system; For the fixture The three-dimensional coordinates of the reference marker point in the original coordinate system of the vision system; For the fixture The three-dimensional coordinates of the reference marker point in the original coordinate system of the vision system; First calculate: , , , ;in, For measuring coordinate system The unit direction vector of the axis; To construct a measurement coordinate system Auxiliary vector along the axis; For measuring coordinate system The unit direction vector of the axis; For measuring coordinate system The unit direction vector of the axis; like: or If the angle result, bend line position result, bend strip width result, and registration error result are all marked as invalid, the measurement calculation will be terminated. For each origin point Calculate the measured coordinates, specifically as follows: , , ;in, , , They represent the first The length coordinates, width coordinates, and normal coordinates of an original point after coordinate transformation. Retaining the following conditions: , The transformed points are then renumbered as follows: ;in, This is a set of measured point clouds after coordinate transformation and boundary filtering. For the preserved set of measured point clouds The Middle One measured point; The retained measured point number; , , They represent the measured points respectively. The length direction coordinates, width direction coordinates, and normal coordinates; This represents the total number of measured points after retention. exist Under these conditions, a reference sampling step size is constructed using the area of ​​the rectangular plate, the width of the rectangular plate, and the number of measured points. Specifically: ; First, calculate the number of samples in the length and width directions respectively: , ;in, This represents the total number of sampling points along the length direction. This represents the total number of sampling points in the width direction before bending. The sampling intervals in the length direction and the width direction before bending are then calculated as follows: , ;in, The sampling interval is along the length direction; The sampling interval in the width direction before bending; Then write the reference point set as follows: , , , , ;in, This is the set of reference points before bending; For reference point set The Middle Line number List reference points; For the length direction, the first Each sampling coordinate; The sampling sequence number is in the length direction; The width direction before bending Each sampling coordinate; This is the sampling sequence number in the width direction.

[0032] Reference Figure 4 The process of dividing the measured width along the direction of the bent section into columns, assigning each measured point to the corresponding column, obtaining the measured average folding height and bending strength of each column, and obtaining the initial position of the bending line and the initial bending band width, specifically includes: The result of dividing three times the theoretical width of the rectangular metal plate by the reference sampling step size is rounded up and then incremented by one to obtain the total number of discrete columns in the direction of the measured width after bending. The spacing between the measured width columns after bending is obtained by dividing three times the theoretical width of the rectangular metal plate by the total number of discrete columns minus one. The starting width coordinate is taken as three-half of the theoretical width of the rectangular metal plate, and written into the center of each measured width column after bending according to the spacing between the measured width columns after bending. For each measured point, compare its width direction coordinate with the center of all the measured width columns after bending. Write the measured width column after bending with the smallest distance as the corresponding nearest measured width column after bending. When multiple measured width columns after bending reach the minimum distance at the same time, write the measured width column after bending with the smaller column number. For each column of measured width after bending, if there are measured points that should be included in the current column, the absolute values ​​of the normal coordinates of all measured points in the current column are averaged to obtain the average folded height of the current column. If there are no measured points that should be included in the current column, the average folded height of the non-empty column closest to the center of the current column is written into the current column. If only the left side has the nearest non-empty column, the average folded height of the left nearest non-empty column is written. If only the right side has the nearest non-empty column, the average folded height of the right nearest non-empty column is written. If the distances from the nearest non-empty columns on both sides to the center of the current column are the same, the arithmetic mean of the average folded heights of the nearest non-empty columns on both sides is written. Calculate the bending strength for all measured width columns after bending. The bending strength of the first measured width column after bending is written as the absolute value of the difference between the measured average folded height of the second measured width column after bending and the measured average folded height of the first measured width column after bending. The bending strength of the last measured width column after bending is written as the absolute value of the difference between the measured average folded height of the last measured width column after bending and the measured average folded height of the second-to-last measured width column after bending. The bending strength of the measured width columns after bending in the middle is written as the absolute value of the difference between the measured average folded height of the next measured width column after bending and the measured average folded height of the previous measured width column after bending. Take the maximum bending strength from the bending width measurement columns excluding the first and last bending width measurement columns to obtain the maximum bending strength value; when the maximum bending strength value is zero, mark the final optimal bending angle, final bending angle, final optimal bending line position, final optimal bending band width, and final registration error as invalid, and terminate the measurement calculation. Write the column corresponding to the maximum bending strength after bending as the peak column; when the bending strength of multiple columns of measured widths after bending reaches the maximum value at the same time, prioritize writing the column of measured widths after bending with the smaller center coordinates. If the center coordinates of the columns of measured widths after bending are still the same, write the column of measured widths after bending with the smaller column number; then restrict the center of the column of measured widths after bending corresponding to the peak column to within ±1 / 2 of the theoretical width of the rectangular metal plate to obtain the initial position of the bending line; Half of the bending strength of the peak column is written as the half-peak threshold. Continuously search to the left from the peak column for the smallest consecutive bend width measurement column number for all bend strengths not lower than the half-peak threshold. Continuously search to the right from the peak column for the largest consecutive bend width measurement column number for all bend strengths not lower than the half-peak threshold. Multiply the difference between the largest and smallest consecutive bend width measurement column numbers by the bend width measurement column spacing to obtain the initial bending band width. When the initial bending band width is less than the reference sampling step size, it is written as the reference sampling step size; when the initial bending band width is greater than the theoretical width of the rectangular metal plate, it is written as the theoretical width of the rectangular metal plate.

[0033] First, calculate the total number of discrete columns in the width direction after bending: ; The column spacing after bending is then calculated as follows: ; Then write the center of the width column as: , ;in, For the first After each bend, measure the width coordinates of the center of the width column; The column number is used to determine the width after bending; For each measured point To calculate the nearest width column index, specifically: If multiple width columns are the same as the measured points If the distances to both columns reach their minimum values, then the column with the smaller column number is selected; among them, For actual measurement points The corresponding nearest width column number; If the first The column exists that satisfies The measured point is then recorded as the first... The number of columns is And calculate the first The average folded height of the column is: ;in, For the first The number of measured points in each width column; For the first Average fold height of each width column; If the first If no measured points exist in the column, then... The value is assigned to the distance from the center of the width column. Height of the nearest non-empty column; If only the leftmost non-empty column exists, take the height of the leftmost non-empty column. If only the rightmost non-empty column exists, take the height of the rightmost non-empty column. The nearest non-empty column on the left and right sides and the center of the width column When the distances are the same, The value is assigned to the arithmetic mean of the heights of the two nearest non-empty columns on the left and right sides; Calculate the bending strength for all width columns as follows: ;in, For the first Bending strength of each width column; The maximum bending strength of the inner width column is calculated as follows: ; like If the angle result, bend line position result, bend strip width result, and registration error result are all marked as invalid, the measurement calculation will be terminated. Take the column number corresponding to the maximum bending strength and write it to the initial position of the bending line, specifically: , ;in, The column number corresponding to the maximum bending strength; This is the initial position of the bend line; If multiple If both reach their maximum values, then take... The smallest column number; like If they are still the same, then take the one with the smaller column number; Take the half-peak threshold as ; From column number Search continuously to the left, and find the one that satisfies the condition. And for any column number All The smallest column number is ;in, This is the column index of the width corresponding to the left boundary of the semi-peak continuous interval; This is the column number of the middle width used when determining the continuity of the half-peak continuous interval; From column number Search continuously to the right, and find the one that satisfies the condition. And for any column number All The largest column number is ;in, This is the column index of the width corresponding to the right boundary of the half-peak continuous interval; The initial estimated bending bandwidth is: .

[0034] The parameters to be determined, including the candidate bend line position, candidate bend angle, and candidate bend band width, are written into a range. A continuous local bend angle distribution is constructed based on the positional difference between the pre-bend width coordinate and the candidate bend line position, forming a continuous bend field. Specifically, this includes: Write the candidate bend line position, candidate bend angle and candidate bend strip width as parameters to be determined, and limit the candidate bend angle between zero and pi, and limit the candidate bend strip width between the reference sampling step size and the theoretical width of the rectangular metal plate. For each pre-bend width coordinate, calculate the position difference between the pre-bend width coordinate and the candidate bend line position to obtain the signed distance from each pre-bend width coordinate to the candidate bend line position; According to the candidate bend width, a continuous local bend angle distribution is written in the width direction before bending. Specifically, for each signed distance, a corresponding continuous local bend angle is written. When the signed distance is outside the left boundary of the bend defined by the candidate bend line position and the candidate bend width, a zero angle is written. When the signed distance is inside the bend defined by the candidate bend line position and the candidate bend width, a continuous local bend angle is written in a linearly increasing manner from the left boundary of the bend defined by the candidate bend line position and the candidate bend width to the right boundary of the bend defined by the candidate bend line position and the candidate bend width. When the signed distance is outside the right boundary of the bend defined by the candidate bend line position and the candidate bend width, a candidate bend angle is written.

[0035] The location of the actual bending line to be determined is denoted as... The actual bending angle to be determined is denoted as Let the actual bending strip width be denoted as And limited to: , ; For each bend, the width coordinate Calculate the width coordinates before bending The signed distance to the bend line is as follows: ;in, For the first Width sampling coordinates before bending To the bend line position The signed distance; According to the width of the bending strip Before bending, a continuous local corner distribution is constructed in the width direction, specifically as follows: ;in, For the first Width sampling coordinates before bending The distribution of local bending angles at the location.

[0036] The step of performing a continuous bending transformation on the reference point set to generate a predicted point cloud specifically includes: For each reference point, perform continuous local bend angle rotation around a rotating straight line whose width direction coordinate is equal to the candidate bend line position and whose normal coordinate is equal to zero. Directly write the length direction coordinate of the reference point as the length direction coordinate of the predicted point. Add the product of the signed distance and the cosine value of the corresponding continuous local bend angle to the candidate bend line position and write it as the width direction coordinate of the predicted point. Write the product of the signed distance and the sine value of the corresponding continuous local bend angle as the normal coordinate of the predicted point to generate the corresponding predicted point. A continuous bending transformation is performed on all reference points to form a predicted point cloud.

[0037] For each reference point , around a straight line , Perform local cornering The rotation yields the predicted point as: , , , ;in, For reference point In parameters , , The predicted point corresponding to the continuous bending field determined by the situation; , , These represent the prediction points. The length direction coordinates, width direction coordinates, and normal coordinates; A continuous bending transformation is performed on all reference points to form a predicted point cloud, specifically: ;in, For all reference points Transformation function through continuous bending The calculated set of predicted point clouds.

[0038] The process of establishing the nearest point correspondence between predicted points and measured points to obtain point cloud registration residuals specifically includes: For each predicted point, search for the measured point cloud that is closest to the predicted point and write the measured point with the smallest distance as the unique nearest corresponding point of the current predicted point; when the distance between multiple measured points and the same predicted point reaches the minimum value at the same time, write the measured point with the smaller index. The point cloud registration residual is obtained by summing the squared distances between all predicted points and their unique nearest corresponding points, and then averaging the sum using the number of reference points in the reference point set.

[0039] For each prediction point In the measured point cloud Find the unique nearest point in the array, and denote the index of the nearest point as . Specifically: If the distances between multiple measured points and the same predicted point simultaneously reach their minimum values, the measured point with the smaller index is selected; among them, For prediction points The corresponding most recent measured point number; Constructing point cloud registration residual function Specifically: .

[0040] The bending tape compactness residual and column height residual are obtained by using the left and right projected boundaries of the bending tape in the post-bending measurement coordinates, the outward diffusion distance of the post-bending measurement width column, the predicted average folding height of the post-bending measurement width column, and the measured average folding height of the post-bending measurement width column. Specifically, this includes: For each signed distance located inside the bend, the projection width coordinates of the bend in the measurement coordinates after bending are calculated point by point according to the combination relationship between the candidate bend line position, the signed distance and the cosine value of the continuous local bend angle corresponding to the signed distance. The minimum value is written as the left projection boundary and the maximum value is written as the right projection boundary. For each column center of the measured width after bending, compare the positional relationship between the current column center and the left and right projection boundaries. When the current column center falls between the left and right projection boundaries, write the out-of-band diffusion distance as zero; when the current column center is to the left of the left projection boundary, write the difference between the left projection boundary and the current column center as the out-of-band diffusion distance; when the current column center is to the right of the right projection boundary, write the difference between the current column center and the right projection boundary as the out-of-band diffusion distance. The product of the square of the out-of-band diffusion distance of each of the measured width columns after bending and the square of the bending strength of the same measured width column after bending is accumulated, and then normalized by the total number of discrete columns in the measured width direction after bending and the square of the theoretical width of the rectangular metal plate to obtain the compact residual of the bending band. For each prediction point, compare its width direction coordinate with the center of all the bend measurement width columns, and write the bend measurement width column with the smallest distance as the nearest bend measurement width column of the current prediction point; when multiple bend measurement width columns reach the minimum distance at the same time, write the bend measurement width column with the smaller column number. For each column of post-bend measurement width, if there is a predicted point that should be included in the current post-bend measurement width column, the absolute values ​​of the normal coordinates of all predicted points in the current post-bend measurement width column are averaged to obtain the predicted average folding height of the current post-bend measurement width column. If there is no predicted point that should be included in the current post-bend measurement width column, the predicted average folding height of the non-empty post-bend measurement width column closest to the center of the current post-bend measurement width column is written into the current post-bend measurement width column. If there is only a nearest non-empty post-bend measurement width column on the left, the predicted average folding height of the nearest non-empty post-bend measurement width column on the left is written. If there is only a nearest non-empty post-bend measurement width column on the right, the predicted average folding height of the nearest non-empty post-bend measurement width column on the right is written. If the distances from the nearest non-empty post-bend measurement width columns on both the left and right sides to the center of the current post-bend measurement width column are the same, the arithmetic mean of the predicted average folding heights of the nearest non-empty post-bend measurement width columns on both the left and right sides is written. The average of the squares of the differences between the measured average folded height and the predicted average folded height of each of the measured widths after bending in all the bends is summed to obtain the column height residual.

[0041] First, calculate the left and right projected boundaries of the bent strip in the coordinate system after bending: , ;in, These are the position coordinates of the inner part of the bend band relative to the bend line; This represents the left projection boundary of the bent strip in the coordinate system after bending. This represents the right projection boundary of the bent strip in the coordinate system after bending. Measure the width of the column center after each bend. Calculate the center of the width column. Relative projection range The out-of-band diffusion distance is as follows: ;in, For the first After each bend, measure the out-of-band diffusion distance from the center of the width column to the projection area of ​​the bend zone; Constructing a compact residual function for the bending strip Specifically: ; For each prediction point The nearest width column index is calculated as follows: If multiple width columns are simultaneously at their minimum distance from the predicted point, the column with the smaller column number is selected; among them, For prediction points The nearest width column index; If the first There exist theoretical width columns that satisfy... If the predicted points are such that the number of corresponding predicted points is denoted as , then the number of predicted points is denoted as And calculate the first The average folded height of each theoretical width column is: ;in, For the first Number of predicted points in a column of theoretical width; For the first Average fold height of each theoretical width column; like Then The value is assigned to the distance from the center of the width column. The most recent non-empty theoretical width column height; If only the leftmost non-empty theoretical width column exists, take the height of the leftmost non-empty theoretical width column. If only the rightmost non-empty theoretical width column exists, take the height of the rightmost non-empty theoretical width column. The nearest non-empty theoretical width column and the center of the width column on both the left and right sides When the distances are the same, The value is assigned to the arithmetic mean of the heights of the nearest non-empty theoretical width columns on the left and right sides; Construct column height residual function Specifically: .

[0042] The total residual is constructed from the point cloud registration residual, the bend band compaction residual, and the column height residual. The final optimal bend line position, the final optimal bend angle, and the final optimal bend band width are obtained through multiple rounds of candidate parameter search. Specifically, this includes: The total residual is obtained by adding the point cloud registration residual, the bending band compaction residual, and the column height residual. One-quarter of the reference sampling step length is written as the search interval for the first round bend line position and the search interval for the first round bend band width, respectively. The result of dividing pi by 180 is written as the search interval for the first round bend angle. Based on the distance from the initial position of the bending line to the negative half-width position of the rectangular metal plate and the distance from the positive half-width position of the rectangular metal plate to the initial position of the bending line, divide each distance by the maximum integer number obtained after the first round of bending line position search interval. Generate candidate values ​​for the first round of bending line positions by offsetting the initial position of the bending line by integer steps. Limit candidate values ​​that exceed the range of the positive and negative half-width of the rectangular metal plate to the range of the positive and negative half-width of the rectangular metal plate. Then delete duplicate values ​​to form the first round of bending line position candidate set. Based on the distance from the initial bending strip width to the reference sampling step length and the distance from the theoretical width of the rectangular metal plate to the initial bending strip width, respectively, divide by the maximum integer number of steps obtained after the first round of bending strip width search interval, generate candidate values ​​for the first round of bending strip width by offsetting by integer steps around the initial bending strip width, and restrict candidate values ​​that exceed the range from the reference sampling step length to the theoretical width of the rectangular metal plate to the range from the reference sampling step length to the theoretical width of the rectangular metal plate, and then delete duplicate values ​​to form the candidate set for the first round of bending strip width; Take values ​​between zero and pi according to the first round of bending angle search interval, delete zero and pi, and form the first round of bending angle candidate set; The candidate sets for the first-round bend line position, the first-round bend band width, and the first-round bend angle are combined using Cartesian products to form a first-round candidate parameter set. The total residual is calculated for each first-round candidate parameter set. The first-round candidate parameter set with the smallest total residual is selected as the center bend line position, center bend angle, and center bend band width obtained in the first round search. When the total residuals of multiple first-round candidate parameter sets simultaneously reach their minimum values, the first-round candidate parameter set with the smaller point cloud registration residual is selected. When the point cloud registration residuals are the same, the first-round candidate parameter set with the smaller bend band compaction residual is selected. When both the point cloud registration residual and the bend band compaction residual are the same, the first-round candidate parameter set with the smaller column height residual is selected. If they are still tied, the first-round candidate parameter set with the smaller bend band width is selected. The search intervals for the first round of bend line position, bend band width, and bend angle are halved to obtain the search intervals for the second round of bend line position, bend band width, and bend angle. Candidate values ​​for the second round of bend line positions are generated by offsetting the center bend line position obtained in the first round by integer steps of -2, -1, 0, +1, and +2. Candidate values ​​exceeding ±1 / 2 of the theoretical width of the rectangular metal plate are restricted to within ±1 / 2 of the theoretical width of the rectangular metal plate. Duplicate values ​​are removed to form the candidate set for the second round of bend line positions. The width of the center bend obtained in the first round of search is offset by integer steps of -2, -1, zero, +1, and +2 to generate candidate values ​​for the width of the second round of bend. Candidate values ​​that exceed the range of the reference sampling step size to the theoretical width of the rectangular metal plate are restricted to the range of the reference sampling step size to the theoretical width of the rectangular metal plate. After deleting duplicate values, the second round of candidate values ​​for the width of the bend is formed. The center bend angle obtained in the first round of search is offset by integer steps of -2, -1, zero, +1, and +2 to generate candidate values ​​for the second round of bend. Candidate values ​​that do not belong to zero and pi are deleted to form the second round of candidate values ​​for the bend. The candidate sets for the second round of bending line positions, bending band widths, and bending angles are combined using Cartesian products to form a second-round candidate parameter set. The total residual is calculated for each candidate parameter set in this set. The candidate parameter set with the smallest total residual is selected as the center bending line position, center bending angle, and center bending band width obtained in the second round of search. When the total residuals of multiple candidate parameter sets simultaneously reach their minimum values, the candidate parameter set with the smaller point cloud registration residual is selected. When the point cloud registration residuals are the same, the candidate parameter set with the smaller bending band compaction residual is selected. When both the point cloud registration residual and the bending band compaction residual are the same, the candidate parameter set with the smaller column height residual is selected. If they are still parallel, the candidate parameter set with the smaller bending band width is selected. The search intervals for the second round of bend line position, bend strip width, and bend angle are halved to obtain the search intervals for the third round of bend line position, bend strip width, and bend angle. Candidate values ​​for the third round of bend line position are generated by offsetting the center bend line position obtained in the second round of search by integer steps of -2, -1, 0, +1, and +2. Candidate values ​​exceeding the theoretical width plus or minus half of the rectangular metal plate are restricted to the range of the theoretical width plus or minus half of the rectangular metal plate. After deleting duplicate values, the candidate set for the third round of bend line position is formed. The third round of candidate values ​​for the bend width is generated by offsetting the center bend width obtained in the second round of search by integer steps of -2, -1, zero, +1, and +2. Candidate values ​​that exceed the range of the reference sampling step size to the theoretical width of the rectangular metal plate are restricted to the range of the reference sampling step size to the theoretical width of the rectangular metal plate. After deleting duplicate values, the third round of candidate values ​​for the bend width is formed. The third round of candidate values ​​for the bend angle is generated by offsetting the center bend angle obtained in the second round of search by integer steps of -2, -1, zero, +1, and +2. Candidate values ​​that do not belong to the range between zero and pi are deleted, forming the third round of candidate values ​​for the bend angle. The candidate sets for the third-round bend line position, the third-round bend band width, and the third-round bend angle are combined using Cartesian products to form a third-round candidate parameter set. The total residual is calculated for each third-round candidate parameter set. The third-round candidate parameter set with the smallest total residual is selected as the final optimal bend line position, final optimal bend angle, and final optimal bend band width. When the total residuals of multiple third-round candidate parameter sets simultaneously reach their minimum values, the third-round candidate parameter set with the smaller point cloud registration residual is selected. When the point cloud registration residuals are the same, the third-round candidate parameter set with the smaller bend band compaction residual is selected. When both the point cloud registration residual and the bend band compaction residual are the same, the third-round candidate parameter set with the smaller column height residual is selected. If they are still tied, the third-round candidate parameter set with the smaller bend band width is selected.

[0043] Constructing the total residual function Specifically: ; Take the first round interval: , , ;in, This is the interval for the first round of location search; This is the first round of bandwidth search interval; This is the interval for the first round of angle search; make: , ;in, For the first round of candidate positions The maximum integer number of steps required to cover the physical boundary to the left; For the first round of candidate positions The maximum integer number of steps required to cover the physical boundary to the right; The first round of candidate position values ​​are: , ;in, The integer offset step number of the first round of candidate position values; The same value in the first round of candidate positions is retained only once to form the first round of candidate positions set; make: , ;in, For the first round of bandwidth candidate set relative The maximum integer number of steps required to cover down to the physical boundary; For the first round of bandwidth candidate set relative The maximum integer number of steps required to cover upwards to the physical boundary; The first round of bandwidth candidate values ​​are: , ;in, This is the integer offset step number of the first round of bandwidth candidate values; The same value in the first round of bandwidth candidate values ​​is retained only once to form the first round of bandwidth candidate set; make: ;in, The largest integer index of the first round of angle candidate values; The first round of candidate angle values ​​are: , ;in, Integer index of the first round of angle candidate values; The first round of candidate parameter sets is formed by combining the first round of position candidate sets, the first round of bandwidth candidate sets, and the first round of angle candidate values ​​using the Cartesian product. The total residual is calculated for each first round of candidate parameter sets. ; The parameter set with the smallest total residual in the first round is selected as follows. ;in, , , These represent the position of the center bend line, the center bend angle, and the width of the center bend strip obtained in the first round of search, respectively. If the total residuals of multiple parameter groups reach their minimum values ​​simultaneously, then compare them sequentially. , , Take the smaller one; if they are still tied, then take the smaller one. The smaller one; Take the second round interval: , , ;in, This is the interval for the second round of position search; This is the second round bandwidth search interval; This is the interval for the second round of angle search; The second round of candidate position values ​​are: , ;in, The integer offset step number of the second round of position candidate values; The same value in the second round of position candidate values ​​is retained only once to form the second round of position candidate set; The second round of bandwidth candidate values ​​is: , ;in, The integer offset step number of the second round of bandwidth candidate values; The same value in the second round of bandwidth candidate values ​​is retained only once to form the second round of bandwidth candidate set; The second round of candidate angle values ​​is: , And delete those that do not belong to the interval. The candidate values; where, The integer offset step number for the second round of angle candidate values; The second round of candidate parameter sets is formed by combining the second round of position candidate sets, the second round of bandwidth candidate sets, and the second round of angle candidate values ​​using the Cartesian product. The total residual is calculated for each second round of candidate parameter sets. ; Take the parameter set with the smallest total residual in the second round as ;in, , , These represent the position of the center bend line, the center bend angle, and the width of the center bend strip obtained in the second round of search, respectively. If the total residuals of multiple parameter groups reach their minimum values ​​simultaneously, then compare them sequentially. , , Take the smaller one; if they are still tied, then take the smaller one. The smaller one; Take the third round interval: , , ;in, This is the interval for the third round of location search; This is the interval for the third round of bandwidth search; This is the interval for the third round of angle search; The third round of candidate position values ​​is: , ;in, The integer offset step number of the third round of position candidate values; The same value in the third round of position candidate values ​​is retained only once to form the third round of position candidate set; The third round of bandwidth candidate values ​​is: , ;in, The integer offset step number of the third round of bandwidth candidate values; The same value in the third round of bandwidth candidate values ​​is retained only once to form the third round of bandwidth candidate set; The third round of candidate angle values ​​is: , And delete those that do not belong to the interval. The candidate values; where, The integer offset step number for the third round of angle candidate values; The third round of candidate parameter sets is formed by combining the third round of position candidate sets, the third round of bandwidth candidate sets, and the third round of angle candidate values ​​using the Cartesian product. The total residual is calculated for each third round of candidate parameter sets. ; Construct the parameter set with the minimum total residual from the third round of candidate parameter sets, specifically as follows: ;in, This represents the final optimal bend line position; The final optimal bending angle; This represents the final optimal bending strip width.

[0044] The process of substituting the final optimal bend line position, final optimal bend angle, and final optimal bend band width into the continuous bend transformation and point cloud registration residual calculation process to obtain the final predicted point cloud, final registration error, and final bend angle specifically includes: Substitute the final optimal bend line position, the final optimal bend angle, and the final optimal bend strip width into the continuous bend transformation, and re-perform the transformation on the reference point set to generate the final predicted point cloud. Substituting the final optimal bend line position, final optimal bend angle, and final optimal bend band width into the point cloud registration residual calculation process, we obtain the final registration error. The final optimal bending angle is converted to degrees according to the conversion relationship between radians and degrees, and the final bending angle is output.

[0045] Optimal parameters , , Substituting the continuous bending transformation, for the reference point set Re-perform the transformation to obtain the final predicted point cloud. ; Substituting the optimal parameters into the point cloud registration residual formula yields the final registration error. Specifically: ; The final bending angle will be output in degrees as follows: ;in, The final bending angle in angle measurement.

[0046] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0047] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A high-precision visual measurement method for the bending angle of metal shells based on point cloud registration, characterized in that, include: Establish a measurement coordinate system, and write the theoretical length of the rectangular metal plate, the theoretical width of the rectangular metal plate, the point cloud of the measured surface after bending, the origin reference marker, the reference marker in the length direction, and the reference marker in the width direction. Complete the point cloud coordinate transformation, boundary filtering, reference sampling step size calculation, and reference point set construction. Divide the measurement width after bending into columns along the direction of the measurement width after bending, assign each measured point to the corresponding measurement width after bending column, obtain the measured average folding height and bending strength of each measurement width after bending column, and obtain the initial position of the bending line and the initial bending band width. The range of parameters to be determined, including candidate bend line position, candidate bend angle, and candidate bend band width, is entered. A continuous local bend angle distribution is constructed based on the positional difference between the pre-bend width coordinate and the candidate bend line position, forming a continuous bend field. Specifically: The current candidate bend line position is denoted as... The current candidate bend angle is denoted as The current candidate bending strip width is denoted as And limited to: , ;in, Used as a reference sampling step size; This represents the theoretical width of the rectangular metal plate. For each pre-bend width coordinate, calculate the signed distance from the pre-bend width coordinate to the bend line, specifically: ;in, For the first Width sampling coordinates before bending To the bend line position The signed distance; Based on the current candidate bending band width Before bending, a continuous local corner distribution is constructed in the width direction, specifically as follows: ;in, For the first Width sampling coordinates before bending Distribution of local bending angles at the location; This refers to the sampling sequence number in the width direction before bending; Perform a continuous bending transformation on the reference point set to generate a predicted point cloud; Establish the nearest point correspondence between the predicted points and the measured points to obtain the point cloud registration residual; The compactness residual and column height residual of the bending strip are obtained by taking the left and right projection boundaries of the bending strip in the measurement coordinates after bending, the outward diffusion distance of the measurement width column after bending, the predicted average folding height of the measurement width column after bending, and the measured average folding height of the measurement width column after bending. The total residual is constructed based on the point cloud registration residual, the bending band compaction residual, and the column height residual. The final optimal bending line position, the final optimal bending angle, and the final optimal bending band width are obtained through multiple rounds of candidate parameter search. Substituting the final optimal bend line position, final optimal bend angle, and final optimal bend band width into the continuous bend transformation and point cloud registration residual calculation process, we obtain the final predicted point cloud, final registration error, and final bend angle.

2. The high-precision visual measurement method for the bending angle of a metal shell based on point cloud registration according to claim 1, characterized in that, The establishment of the measurement coordinate system involves inputting the theoretical length and width of the rectangular metal plate, the point cloud of the measured surface after bending, the origin reference marker, the length direction reference marker, and the width direction reference marker. This includes point cloud coordinate transformation, boundary filtering, reference sampling step size calculation, and reference point set construction. Specifically, this includes: The center of the measurement surface of the rectangular metal plate before bending is set as the origin of the coordinate system. The length direction of the rectangular metal plate is set as the positive direction of the length direction of the coordinate system. The width direction of the rectangular metal plate is set as the positive direction of the width direction of the coordinate system. The normal direction of the plate surface is set as the positive normal direction of the coordinate system. The measurement surface area before bending is limited to the rectangular area enclosed by half of the theoretical length of the rectangular metal plate in the length direction and half of the theoretical width of the rectangular metal plate in the width direction. Write the factory-fixed 3D point cloud device coordinate system of the vision system into the original coordinate system of the vision system, and write the point cloud of the bent surface obtained by the vision system. At the same time, write the three-dimensional coordinates of the origin reference mark point, the length direction reference mark point, and the width direction reference mark point in the original coordinate system of the vision system onto the fixture. Construct a unit vector in the length direction according to the direction from the origin reference point to the length reference point. Construct a unit vector in the width direction by removing the projection of the width reference point onto the unit vector in the length direction. Construct a unit vector in the normal direction according to the cross product direction of the unit vector in the length direction and the unit vector in the width direction. When the origin reference mark coincides with the length direction reference mark, or when the length of the width direction auxiliary vector after removing the projection is zero, the final optimal bending angle, final bending angle, final optimal bending line position, final optimal bending band width, and final registration error are marked as invalid, and the measurement calculation is terminated. For each origin point, calculate its projection relative to the origin reference point on the unit vectors of the length direction, width direction, and normal direction, and obtain the length direction coordinates, width direction coordinates, and normal coordinates of each origin point in the measurement coordinate system. The transformed points whose length direction coordinates are within ±1 / 2 of the theoretical length of the rectangular metal plate and whose width direction coordinates are within ±3 / 2 of the theoretical width of the rectangular metal plate are retained. The retained points are renumbered and a measured point cloud is formed. Under the condition that the total number of retained measured points is not less than one, the square root of the product of the theoretical length and the theoretical width of the rectangular metal plate is divided by the total number of retained measured points, and the square root result is compared with half of the theoretical width of the rectangular metal plate. The minimum value between the square root result and half of the theoretical width of the rectangular metal plate is taken as the reference sampling step size. Divide the theoretical length of the rectangular metal plate by the reference sampling step size and the theoretical width of the rectangular metal plate by the reference sampling step size, round up the results, and increment them by one to obtain the total number of sampling points in the length direction and the total number of sampling points in the width direction before bending. Then, divide the theoretical length of the rectangular metal plate by the total number of sampling points in the length direction minus one, and divide the theoretical width of the rectangular metal plate by the total number of sampling points in the width direction before bending minus one to obtain the sampling interval in the length direction and the sampling interval in the width direction before bending. Write these intervals into the reference point set according to the sampling sequence number in the length direction and the sampling sequence number in the width direction.

3. The high-precision visual measurement method for the bending angle of a metal shell based on point cloud registration according to claim 2, characterized in that, The process involves dividing the measurement width into columns along the direction of the post-bend measurement width, assigning each measured point to its corresponding column, obtaining the measured average folding height and bending strength for each column, and determining the initial position of the bending line and the initial bending band width. Specifically, this includes: The result of dividing three times the theoretical width of the rectangular metal plate by the reference sampling step size is rounded up and then incremented by one to obtain the total number of discrete columns in the direction of the measured width after bending. The spacing between the measured width columns after bending is obtained by dividing three times the theoretical width of the rectangular metal plate by the total number of discrete columns minus one. The starting width coordinate is taken as three-half of the theoretical width of the rectangular metal plate, and written into the center of each measured width column after bending according to the spacing between the measured width columns after bending. For each measured point, compare its width direction coordinate with the center of all the measured width columns after bending, and write the measured width column after bending with the smallest distance as the corresponding nearest measured width column after bending; when multiple measured width columns after bending reach the minimum distance at the same time, write the measured width column after bending with the smallest column number value. For each column of measured width after bending, if there are measured points that should be included in the current column, the absolute values ​​of the normal coordinates of all measured points in the current column are averaged to obtain the average folded height of the current column. If there are no measured points that should be included in the current column, the average folded height of the non-empty column closest to the center of the current column is written into the current column. If only the left side has the nearest non-empty column, the average folded height of the left nearest non-empty column is written. If only the right side has the nearest non-empty column, the average folded height of the right nearest non-empty column is written. If the distances from the nearest non-empty columns on both sides to the center of the current column are the same, the arithmetic mean of the average folded heights of the nearest non-empty columns on both sides is written. Calculate the bending strength for all measured width columns after bending. The bending strength of the first measured width column after bending is written as the absolute value of the difference between the measured average folded height of the second measured width column after bending and the measured average folded height of the first measured width column after bending. The bending strength of the last measured width column after bending is written as the absolute value of the difference between the measured average folded height of the last measured width column after bending and the measured average folded height of the second-to-last measured width column after bending. The bending strength of the measured width columns after bending in the middle is written as the absolute value of the difference between the measured average folded height of the next measured width column after bending and the measured average folded height of the previous measured width column after bending. Take the maximum bending strength from the bending width measurement columns excluding the first and last bending width measurement columns to obtain the maximum bending strength value; when the maximum bending strength value is zero, mark the final optimal bending angle, final bending angle, final optimal bending line position, final optimal bending band width, and final registration error as invalid, and terminate the measurement calculation. Write the column corresponding to the maximum bending strength after bending as the peak column; when the bending strength of multiple columns of measured widths after bending reaches the maximum value at the same time, first write the column of measured widths after bending with the smallest center coordinate value. If the center coordinates of the columns of measured widths after bending are still the same, write the column of measured widths after bending with the smallest column number value; then restrict the center of the column of measured widths after bending corresponding to the peak column to within a range of plus or minus half of the theoretical width of the rectangular metal plate to obtain the initial position of the bending line; Half of the bending strength of the peak column is written as the half-peak threshold. Continuously search to the left from the peak column for the smallest consecutive bend width measurement column number for all bend strengths not lower than the half-peak threshold. Continuously search to the right from the peak column for the largest consecutive bend width measurement column number for all bend strengths not lower than the half-peak threshold. Multiply the difference between the largest and smallest consecutive bend width measurement column numbers by the bend width measurement column spacing to obtain the initial bending band width. When the initial bending band width is less than the reference sampling step size, it is written as the reference sampling step size; when the initial bending band width is greater than the theoretical width of the rectangular metal plate, it is written as the theoretical width of the rectangular metal plate.

4. The high-precision visual measurement method for the bending angle of a metal shell based on point cloud registration according to claim 3, characterized in that, The step of performing a continuous bending transformation on the reference point set to generate a predicted point cloud specifically includes: For each reference point, perform continuous local bend angle rotation around a rotating straight line whose width direction coordinate is equal to the candidate bend line position and whose normal coordinate is equal to zero. Directly write the length direction coordinate of the reference point as the length direction coordinate of the predicted point. Add the product of the signed distance and the cosine value of the corresponding continuous local bend angle to the candidate bend line position and write it as the width direction coordinate of the predicted point. Write the product of the signed distance and the sine value of the corresponding continuous local bend angle as the normal coordinate of the predicted point to generate the corresponding predicted point. A continuous bending transformation is performed on all reference points to form a predicted point cloud.

5. The high-precision visual measurement method for the bending angle of a metal shell based on point cloud registration according to claim 4, characterized in that, The process of establishing the nearest point correspondence between predicted and measured points to obtain point cloud registration residuals specifically includes: For each predicted point, search for the measured point cloud that is closest to the predicted point and write the measured point with the smallest distance as the unique nearest corresponding point of the current predicted point; when the distance between multiple measured points and the same predicted point reaches the minimum value at the same time, write the measured point with the smallest measured point number. The point cloud registration residual is obtained by summing the squared distances between all predicted points and their unique nearest corresponding points, and then averaging the sum using the number of reference points in the reference point set.

6. The high-precision visual measurement method for the bending angle of a metal shell based on point cloud registration according to claim 5, characterized in that, The bending tape compactness residual and column height residual are obtained by using the left and right projected boundaries of the bending tape in the post-bending measurement coordinates, the outward diffusion distance of the post-bending measurement width column, the predicted average folding height of the post-bending measurement width column, and the measured average folding height of the post-bending measurement width column. Specifically, this includes: For each signed distance located inside the bend, the projection width coordinates of the bend in the measurement coordinates after bending are calculated point by point according to the combination relationship between the candidate bend line position, the signed distance and the cosine value of the continuous local bend angle corresponding to the signed distance. The minimum value is written as the left projection boundary and the maximum value is written as the right projection boundary. For each column center of the measured width after bending, compare the positional relationship between the current column center and the left and right projection boundaries. When the current column center falls between the left and right projection boundaries, write the out-of-band diffusion distance as zero; when the current column center is to the left of the left projection boundary, write the difference between the left projection boundary and the current column center as the out-of-band diffusion distance; when the current column center is to the right of the right projection boundary, write the difference between the current column center and the right projection boundary as the out-of-band diffusion distance. The product of the square of the out-of-band diffusion distance of each of the measured width columns after bending and the square of the bending strength of the same measured width column after bending is accumulated, and then normalized by the total number of discrete columns in the measured width direction after bending and the square of the theoretical width of the rectangular metal plate to obtain the compact residual of the bending band. For each prediction point, compare its width direction coordinate with the distance of the center of all the bend measurement width columns, and write the bend measurement width column with the smallest distance as the nearest bend measurement width column for the current prediction point; when multiple bend measurement width columns reach the minimum distance at the same time, write the bend measurement width column with the smallest column number value. For each column of post-bend measurement width, if there is a predicted point that should be included in the current post-bend measurement width column, the absolute values ​​of the normal coordinates of all predicted points in the current post-bend measurement width column are averaged to obtain the predicted average folding height of the current post-bend measurement width column. If there is no predicted point that should be included in the current post-bend measurement width column, the predicted average folding height of the non-empty post-bend measurement width column closest to the center of the current post-bend measurement width column is written into the current post-bend measurement width column. If there is only a nearest non-empty post-bend measurement width column on the left, the predicted average folding height of the nearest non-empty post-bend measurement width column on the left is written. If there is only a nearest non-empty post-bend measurement width column on the right, the predicted average folding height of the nearest non-empty post-bend measurement width column on the right is written. If the distances from the nearest non-empty post-bend measurement width columns on both the left and right sides to the center of the current post-bend measurement width column are the same, the arithmetic mean of the predicted average folding heights of the nearest non-empty post-bend measurement width columns on both the left and right sides is written. The average of the squares of the differences between the measured average folded height and the predicted average folded height of each of the measured widths after bending in all the bends is summed to obtain the column height residual.

7. The high-precision visual measurement method for the bending angle of a metal shell based on point cloud registration according to claim 6, characterized in that, The total residual is constructed from the point cloud registration residual, the bend band compaction residual, and the column height residual. The final optimal bend line position, the final optimal bend angle, and the final optimal bend band width are obtained through multiple rounds of candidate parameter search. Specifically, this includes: The total residual is obtained by adding the point cloud registration residual, the bending band compaction residual, and the column height residual. One-quarter of the reference sampling step length is written as the search interval for the first round bend line position and the search interval for the first round bend band width, respectively. The result of dividing pi by 180 is written as the search interval for the first round bend angle. Based on the distance from the initial position of the bending line to the negative half-width position of the rectangular metal plate and the distance from the positive half-width position of the rectangular metal plate to the initial position of the bending line, divide each distance by the maximum integer number obtained after the first round of bending line position search interval. Generate candidate values ​​for the first round of bending line positions by offsetting the initial position of the bending line by integer steps. Limit candidate values ​​that exceed the range of the positive and negative half-width of the rectangular metal plate to the range of the positive and negative half-width of the rectangular metal plate. Then delete duplicate values ​​to form the first round of bending line position candidate set. Based on the distance from the initial bending strip width to the reference sampling step length and the distance from the theoretical width of the rectangular metal plate to the initial bending strip width, respectively, divide by the maximum integer number of steps obtained after the first round of bending strip width search interval, generate candidate values ​​for the first round of bending strip width by offsetting by integer steps around the initial bending strip width, and restrict candidate values ​​that exceed the range from the reference sampling step length to the theoretical width of the rectangular metal plate to the range from the reference sampling step length to the theoretical width of the rectangular metal plate, and then delete duplicate values ​​to form the candidate set for the first round of bending strip width; Take values ​​between zero and pi according to the first round of bending angle search interval, delete zero and pi, and form the first round of bending angle candidate set; The candidate sets for the first-round bend line position, the first-round bend band width, and the first-round bend angle are combined using Cartesian products to form a first-round candidate parameter set. The total residual is calculated for each first-round candidate parameter set. The first-round candidate parameter set with the smallest total residual is selected as the center bend line position, center bend angle, and center bend band width obtained in the first round search. When the total residuals of multiple first-round candidate parameter sets simultaneously reach their minimum values, the first-round candidate parameter set with the smallest point cloud registration residual value is selected. When the point cloud registration residuals are the same, the first-round candidate parameter set with the smallest bend band compaction residual value is selected. When both the point cloud registration residual and the bend band compaction residual are the same, the first-round candidate parameter set with the smallest column height residual value is selected. If they are still tied, the first-round candidate parameter set with the smallest bend band width value is selected. The search intervals for the first round of bend line position, bend band width, and bend angle are halved to obtain the search intervals for the second round of bend line position, bend band width, and bend angle. Candidate values ​​for the second round of bend line positions are generated by offsetting the center bend line position obtained in the first round by integer steps of -2, -1, 0, +1, and +2. Candidate values ​​exceeding ±1 / 2 of the theoretical width of the rectangular metal plate are restricted to within ±1 / 2 of the theoretical width of the rectangular metal plate. Duplicate values ​​are removed to form the candidate set for the second round of bend line positions. The width of the center bend obtained in the first round of search is offset by integer steps of -2, -1, zero, +1, and +2 to generate candidate values ​​for the width of the second round of bend. Candidate values ​​that exceed the range of the reference sampling step size to the theoretical width of the rectangular metal plate are restricted to the range of the reference sampling step size to the theoretical width of the rectangular metal plate. After deleting duplicate values, the second round of candidate values ​​for the width of the bend is formed. The center bend angle obtained in the first round of search is offset by integer steps of -2, -1, zero, +1, and +2 to generate candidate values ​​for the second round of bend. Candidate values ​​that do not belong to zero and pi are deleted to form the second round of candidate values ​​for the bend. The candidate sets for the second round of bending line positions, bending band widths, and bending angles are combined using Cartesian products to form a second-round candidate parameter set. The total residual is calculated for each candidate parameter set in this set. The candidate parameter set with the smallest total residual is selected as the center bending line position, center bending angle, and center bending band width obtained in the second round of search. When the total residuals of multiple candidate parameter sets simultaneously reach their minimum values, the candidate parameter set with the smallest point cloud registration residual value is selected. If the point cloud registration residuals are the same, the candidate parameter set with the smallest bending band compaction residual value is selected. If both the point cloud registration residual and the bending band compaction residual are the same, the candidate parameter set with the smallest column height residual value is selected. If they are still tied, the candidate parameter set with the smallest bending band width value is selected. The search intervals for the second round of bend line position, bend strip width, and bend angle are halved to obtain the search intervals for the third round of bend line position, bend strip width, and bend angle. Candidate values ​​for the third round of bend line position are generated by offsetting the center bend line position obtained in the second round of search by integer steps of -2, -1, 0, +1, and +2. Candidate values ​​exceeding the theoretical width plus or minus half of the rectangular metal plate are restricted to the range of the theoretical width plus or minus half of the rectangular metal plate. After deleting duplicate values, the candidate set for the third round of bend line position is formed. The third round of candidate values ​​for the bend width is generated by offsetting the center bend width obtained in the second round of search by integer steps of -2, -1, zero, +1, and +2. Candidate values ​​that exceed the range of the reference sampling step size to the theoretical width of the rectangular metal plate are restricted to the range of the reference sampling step size to the theoretical width of the rectangular metal plate. After deleting duplicate values, the third round of candidate values ​​for the bend width is formed. The third round of candidate values ​​for the bend angle is generated by offsetting the center bend angle obtained in the second round of search by integer steps of -2, -1, zero, +1, and +2. Candidate values ​​that do not belong to the range between zero and pi are deleted, forming the third round of candidate values ​​for the bend angle. The candidate sets for the third-round bend line position, the third-round bend band width, and the third-round bend angle are combined using a Cartesian product to form a third-round candidate parameter set. The total residual is calculated for each third-round candidate parameter set. The third-round candidate parameter set with the smallest total residual is selected as the final optimal bend line position, final optimal bend angle, and final optimal bend band width. When the total residual of multiple third-round candidate parameter sets reaches its minimum value simultaneously, the third-round candidate parameter set with the smallest point cloud registration residual value is selected. When the point cloud registration residuals are the same, the third-round candidate parameter set with the smallest bend band compaction residual value is selected. When both the point cloud registration residual and the bend band compaction residual are the same, the third-round candidate parameter set with the smallest column height residual value is selected. If they are still tied, the third-round candidate parameter set with the smallest bend band width value is selected.

8. The high-precision visual measurement method for the bending angle of a metal shell based on point cloud registration according to claim 7, characterized in that, The process of substituting the final optimal bend line position, final optimal bend angle, and final optimal bend band width into the continuous bend transformation and point cloud registration residual calculation process to obtain the final predicted point cloud, final registration error, and final bend angle specifically includes: Substitute the final optimal bend line position, the final optimal bend angle, and the final optimal bend strip width into the continuous bend transformation, and re-perform the transformation on the reference point set to generate the final predicted point cloud. Substituting the final optimal bend line position, final optimal bend angle, and final optimal bend band width into the point cloud registration residual calculation process, we obtain the final registration error. The final optimal bending angle is converted to degrees according to the conversion relationship between radians and degrees, and the final bending angle is output.

Citation Information

Patent Citations

  • Method and device for measuring bending angle of numerical control bending workpiece based on binocular camera

    CN115330845A

  • Detection method of sheet metal model and electronic equipment

    CN121415112A