An Adaptive Correction Method for Robot Visual Perception

Abstract

This invention provides an adaptive correction method for robot vision perception, belonging to the field of track-type welding robot technology. The method includes: S1: acquiring a single frame of two-dimensional data; S2: extracting the right and left bevel lines and calculating their intersection and angle; S3: finding the left and right pipe body lines; S4: calculating the bevel depth and the number of welding layers; S5: hand-eye calibration; S6: adjusting the welding torch position in real time to ensure the torch is always directly above the intersection and maintains a certain distance; S7: performing the first weld of the V-shaped bevel; S8: acquiring a single frame of two-dimensional data; S9: extracting the right and left bevel lines; S10: extracting the pixels at the bottom of the bevel; S11: adjusting the welding torch position in real time according to the position of each sampling point; S12: repeating steps S8-S11 to weld layer by layer. This method has strong compatibility, can calculate welding deviations and correct path parameters in real time, and effectively solves the problems of large fluctuations and poor stability in welding accuracy existing in the industry.

Description

Technical Field

[0001] This invention relates to the field of track-mounted welding robot technology, and more specifically to an adaptive correction method for robot visual perception. Background Technology

[0002] When using a track-mounted welding robot for pipe welding, the ideal welding torch swing amplitude is usually a fixed value set based on experience. However, if there are issues such as track installation deviation or welding torch relative fixation loosening, the welding torch swing amplitude set in this case cannot effectively achieve pipe beveling. Furthermore, the welding stability of the welding torch swing amplitude set based on experience is poor, and it is impossible to achieve high-precision welding work according to the actual situation of the weld.

[0003] In existing technologies, patent CN121492076A provides a welding system and method where laser vision scanning and welding operations are performed simultaneously. The sensor is located close to the welding torch, and the strong arc light, high-temperature fumes, and metal spatter generated during welding directly cover and interfere with the laser signal, leading to failure in weld feature extraction, data distortion, and model calculation errors. This makes it impossible to maintain long-term stable operation under harsh field conditions. Over-reliance on pre-established all-position 3D models for path planning constitutes an open-loop control method, which is incapable of handling random disturbances such as misalignment of pipe sections, irregular bevels, and temporary deformations on-site. The deviation between the model and actual working conditions accumulates continuously, ultimately leading to a decrease in welding accuracy.

[0004] Patent CN121360938A describes welding equipment that relies entirely on mechanical methods for welding. It lacks vision, lasers, sensors, and algorithms, making it a traditional semi-automated device without any intelligent or adaptive capabilities. Problems such as installation deviations, track offsets, pipe misalignment, and mechanical loosening cannot be automatically compensated for or corrected. Inaccurate positioning will result in weld misalignment and leaks. Furthermore, it cannot recognize changes in bevel depth, width, or cross-section, cannot automatically plan the number of welding layers based on the bevel width, and cannot dynamically adjust parameters such as welding torch amplitude and welding speed.

[0005] Therefore, we provide an adaptive correction method for robot visual perception. Summary of the Invention

[0006] The purpose of this invention is to provide an adaptive correction method for robot visual perception, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] This invention provides an adaptive correction method for robot visual perception, comprising the following steps: S1: Use a line laser profile scanner to scan the V-shaped bevel pipe at a fixed frequency to obtain a single frame of two-dimensional data of the V-shaped bevel pipe; S2: Based on the single-frame two-dimensional data obtained in step S1, the edge probability space clustering algorithm of multi-scale operator is used to extract the right bevel line and the left bevel line, and the intersection point and angle of the two bevel lines are calculated. S3: Remove pixels near the right bevel line and the left bevel line, find the left pipe body line to the left of the left bevel line, and find the right pipe body line to the right of the right bevel line. S4: Calculate the bevel depth and the number of weld layers; S5: Perform hand-eye calibration of the line laser profile scanner and welding torch; S6: Calculate the position of the intersection of the right bevel line and the left bevel line in real time for each frame of two-dimensional data, and adjust the welding gun position in real time according to the bevel depth and the real-time position of the intersection point in step S4. S7: Perform the first weld of the V-groove. After the first weld is completed, the V-groove will appear as an approximate trapezoid. S8: Use a line laser profile scanner to scan the V-bevel pipe at a fixed frequency to obtain a single frame of two-dimensional data of the V-bevel pipe; S9: Based on the single-frame two-dimensional data from step S8, the right slope line and the left slope line are extracted using the edge probability space clustering algorithm with multi-scale operators. S10: Remove pixels near the right bevel line and the left bevel line. Find the left pipe body line to the left of the left bevel line and the right pipe body line to the right of the right bevel line. Then remove pixels near the left pipe body line and the right pipe body line. The remaining pixels are the bottom pixels of the bevel. S11: Use the manifold sampling algorithm to sample the pixels at the bottom of the bevel, obtain several physical sampling points, calculate the position of each physical sampling point, and adjust the welding gun position in real time according to the position of each sampling point. S12: Repeat steps S8-S11 to weld layer by layer.

[0009] Furthermore, the specific process of step S2 includes: S21: Based on the single-frame two-dimensional data obtained in step S1, use the Canny edge detection algorithm to extract edge pixels in the image; S22: For the edge pixels extracted in step S21, use the Hough line detection algorithm to fit them into standard short line segments. After the line fitting is completed, extract the start and end coordinates of each short line segment. S23: Calculate the slope of each short straight line segment, and perform angle clustering based on the slope of the short straight line segment. According to the V-shaped bevel of the V-shaped bevel pipe, the clusters are divided into three types, including the short straight line segment with the highest slope, the short straight line segment with the lowest slope, and the short straight line segment with the middle slope. The bevel line of the V-shaped bevel is represented by the straight line with the highest slope and the straight line with the lowest slope. The short straight line segment with the middle slope is removed, and the short straight line segment with the highest slope and the short straight line segment with the lowest slope are retained. S24: Based on the short line segments with the highest slope and the short line segments with the lowest slope in step S23, divide the endpoints of the line into the endpoint group to which the short line with the highest slope belongs and the endpoint group to which the short line with the lowest slope belongs. Fit these two endpoint groups using the least squares method respectively. The line fitted by the endpoint group to which the short line with the highest slope belongs is the right slope line, and the line fitted by the endpoint group to which the short line with the lowest slope belongs is the left slope line. S25: Calculate the location of the intersection point and the included angle between the right bevel line and the left bevel line; The equation of the right bevel line is defined as follows:

[0010] in, This indicates the slope of the straight line at the right bevel. This represents the intercept of the straight line at the right bevel. and Represents the coordinates of any point on the straight line at the right bevel; The equation of the straight line at the left bevel is defined as follows:

[0011] in, This indicates the slope of the straight line at the left bevel. This represents the intercept of the straight line at the left bevel. and Represents the coordinates of any point on the straight line at the left bevel; The intersection point of the right bevel line and the left bevel line is calculated to satisfy:

[0012] in, and Indicates the coordinates of the intersection point; The angle calculation for the right bevel line and the left bevel line satisfies:

[0013] in, This indicates the angle between the right bevel line and the left bevel line.

[0014] Furthermore, step S3 specifically includes: S31: Set a distance threshold range and remove pixels near the right and left bevel lines that are within the distance threshold range; S32: Among the remaining pixels, separate the pixels to the left of the left bevel line and the pixels to the right of the right bevel line, using the following formula:

[0015] in, The discriminant value; Indicates the slope of the straight line at the left or right bevel. Indicates the intercept of the left or right bevel line; and Represents the coordinates of a pixel; like The pixel is on the right side of the straight line at the right slope. The pixel is to the left of the straight line at the left slope. S33: Based on the pixels to the left of the left bevel line and the pixels to the right of the right bevel line selected in step S32, the pixels to the left of the left bevel line are fitted into a straight line using the least squares method, which is the left pipe body straight line. The pixels to the right of the right bevel line are fitted into a straight line using the least squares method, which is the right pipe body straight line.

[0016] Furthermore, step S4 specifically includes: S41: Define the equation of the straight line of the left tube as follows:

[0017] in, Indicates the slope of the straight line on the left side of the tube; Indicates the intercept of the straight line on the left side of the tube; and Represents the coordinates of any point on the straight line of the left tube; The equation of the right tube straight line is defined as follows:

[0018] in, Indicates the slope of the straight line on the right side of the tube; Indicates the intercept of the right tube's straight line; and Represents the coordinates of any point on the straight line of the right tube; Calculate the distances from the intersection of the right bevel line and the left bevel line to the left pipe body line and the right pipe body line, and take the smaller value as the bevel depth. The formula is:

[0019] in, Indicates the bevel depth; S42: Calculate the number of weld layers using the following formula:

[0020] in, Indicates the number of welding layers; This indicates the maximum allowable thickness for a single weld pass.

[0021] Furthermore, step S5 specifically includes: A front-end sensing field is constructed using a line laser profile scanner. An industrial control computer performs a nonlinear mapping transformation from the pixel coordinate system of the line laser profile scanner to the execution space coordinate system of the welding torch. The mapping relationship is as follows:

[0022]

[0023] in, Indicates the instantaneous time of data acquisition by the line laser profile scanner; Indicates the actual arrival of the welding torch. The physical location and time point scanned by the timeline laser profile scanner; The welding path points are located in the coordinate system of the line laser profile scanner. The target point for the welding torch in the base coordinate system of the welding robot; Hand-eye calibration matrix; This is the real-time pose matrix for the welding robot. This indicates the distance that can be perceived in advance, as determined by hand-eye calibration. This indicates the speed of the welding robot's movement; Indicates the time of delayed compensation.

[0024] Furthermore, step S6 specifically includes: S61: Based on the position calculation of the intersection point in step S25, calculate the position of the intersection point of the right slope line and the left slope line of each frame of two-dimensional data in real time; S62: The perpendicular line from the intersection of the right bevel line and the left bevel line to the left pipe body line is denoted as... The perpendicular line from the intersection of the right bevel line and the left bevel line to the right pipe body line is denoted as... Keep the shorter of the two perpendicular lines, denoted as ,satisfy:

[0025]

[0026]

[0027] S63: The slope of the straight line of the tube corresponding to the short perpendicular line satisfies:

[0028] in, This indicates the slope of the straight line in the pipe corresponding to the short perpendicular line; S64: The welding torch is always directly above the intersection point and maintains a certain distance. This distance is defined as the welding height. Based on the intersection point position and the welding height, the welding path point in the line laser profile scanner coordinate system is obtained. The formula is:

[0029] in, Indicates the welding height; Based on the welding path points in the coordinate system of the line laser contour scanner, the execution target point of the welding torch in the base coordinate system of the welding robot is obtained, using the following formula: .

[0030] Furthermore, step S9 specifically includes: S91: Based on the single-frame two-dimensional data obtained in step S8, the edge pixels in the image are extracted using the Canny edge detection algorithm; S92: For the edge pixels extracted in step S81, use the Hough line detection algorithm to fit them into standard short line segments. After the line fitting is completed, extract the start and end coordinates of each short line segment. S93: Calculate the slope of each short straight line segment, and perform angle clustering based on the slope of the short straight line segments. According to the approximate trapezoidal shape of the V-shaped bevel pipe, the clusters are divided into three types, including the short straight line segments with the highest slope, the short straight line segments with the lowest slope, and the short straight line segments with the middle slope. The bevel line is represented by the straight line with the highest slope and the straight line with the lowest slope. The short straight line segments with the middle slope are removed, and the short straight line segments with the highest slope and the short straight line segments with the lowest slope are retained. S94: Based on the short line segments with the highest and lowest slopes in step S93, divide the endpoints of the lines into the endpoint group to which the short line with the highest slope belongs and the endpoint group to which the short line with the lowest slope belongs. Fit these two endpoint groups using the least squares method. The line fitted by the endpoint group to which the short line with the highest slope belongs is the right slope line, and the line fitted by the endpoint group to which the short line with the lowest slope belongs is the left slope line.

[0031] Furthermore, step S10 specifically includes: S101: Set a distance threshold range and remove pixels near the right and left bevel lines that are within the distance threshold range; S102: Among the remaining pixels, separate the pixels to the left of the left bevel line and the pixels to the right of the right bevel line, using the following formula:

[0032] in, The discriminant value; Indicates the slope of the straight line at the left or right bevel. Indicates the intercept of the left or right bevel line; and Represents the coordinates of a pixel; like The pixel is on the right side of the straight line at the right slope. The pixel is to the left of the straight line at the left slope. S103: Based on the pixels on the left side of the left bevel line and the pixels on the right side of the right bevel line selected in step S102, the pixels on the left side of the left bevel line are fitted into a straight line using the least squares method, which is the left pipe body straight line. The pixels on the right side of the right bevel line are fitted into a straight line using the least squares method, which is the right pipe body straight line. S104: Set a distance threshold range and remove pixels that are within the distance threshold range near the straight lines of the left and right pipe bodies. The remaining pixels are the bottom pixels of the bevel.

[0033] Furthermore, step S11 specifically includes: S111: Assume the pipe axis is... The axis is used to calculate the range of pixels at the bottom of the bevel along the pipe axis, satisfying the following:

[0034] in, This indicates the axial coordinate of the starting pixel on the left side of the bottom of the bevel; Indicates the axial coordinate of the left-hand end pixel at the bottom of the bevel; Generate on this interval A sampling point covering the entire width of the bevel, satisfying:

[0035] in, Indicates the first The axial coordinates of each sampling point; Indicates an index; S112: Using the sampling points as guides, perform a nearest neighbor search within the pixels of the actual feature pixels at the bottom of the bevel to extract physical sampling points that fit the actual shape of the bevel, satisfying:

[0036] in, These represent physical sampling points that closely match the actual shape of the bevel. This represents the set of pixels representing the true features at the bottom of the bevel. This represents the actual feature pixels at the bottom of the bevel to be compared; Indicates the axial pixel coordinate; And record the coordinates of each physical sampling point. ,in, Represents pixel coordinates in the depth direction; S113: Define the equation of the straight line of the left tube as follows:

[0037] in, Indicates the slope of the straight line on the left side of the tube; Indicates the intercept of the straight line on the left side of the tube; and Represents the coordinates of any point on the straight line of the left tube; The equation of the right tube straight line is defined as follows:

[0038] in, Indicates the slope of the straight line on the right side of the tube; Indicates the intercept of the right tube's straight line; and Represents the coordinates of any point on the straight line of the right tube; The perpendicular line from each physical sampling point to the straight line of the left pipe body is denoted as... The perpendicular line from each physical sampling point to the straight line of the right pipe body is denoted as... Keep the shorter of the two perpendicular lines, denoted as ,satisfy:

[0039]

[0040]

[0041] S114: The slope of the straight line of the pipe body corresponding to the short perpendicular line satisfies:

[0042] in, This indicates the slope of the straight line in the pipe corresponding to the short perpendicular line; S115: The welding torch is always above the physical sampling point and maintains a certain distance, which is defined as the welding height. Based on the position of the physical sampling point and the welding height, the welding path point in the line laser profile scanner coordinate system is obtained. The formula is:

[0043] in, Indicates the welding height; Based on the welding path points in the coordinate system of the line laser contour scanner, the execution target point of the welding torch in the base coordinate system of the welding robot is obtained, using the following formula: .

[0044] Compared with the prior art, the present invention has the following technical effects: 1. In this invention, an adaptive correction method with a closed-loop feature of perception-decision-execution is constructed. This method can calculate welding deviation and correct path parameters in real time, upgrading traditional blind welding to dynamic tracking, significantly improving the robustness and automation level of welding operations under complex working conditions, and effectively solving the problems of large fluctuations and poor stability in welding accuracy in the industry.

[0045] 2. In this invention, the method uses a multi-scale operator edge probability space clustering algorithm to extract the right bevel line and the left bevel line. By fragmenting the two-dimensional contour information and using Hough short line fitting in the dual space to reconstruct the endpoints, it can extract sub-pixel level feature points from the original image full of noise interference, which greatly enhances the anti-distortion ability and analytical accuracy of the recognition algorithm and achieves high-precision decoupling of the bevel line segment.

[0046] 3. In this invention, the method constructs a nonlinear mapping transformation with "sensing ahead and execution lagging", which avoids nonlinear interference from electromagnetic and optical fields such as strong arc light and high temperature dust at the physical level. Combined with efficient digital image parallel processing logic, it can realize millisecond-level fast parsing of single frame data, inject key feature parameters into the control core in real time, and form a correction feedback chain with an extremely high refresh rate, which perfectly meets the stringent requirements of high-speed welding for real-time response.

[0047] 4. In this invention, the method adopts a manifold sampling algorithm based on the concept of manifold. Unlike traditional algorithms that rely on mathematical interpolation to generate pseudo observation points, the manifold sampling algorithm strictly distributes points uniformly in the real edge point cloud space, ensuring that each control sampling point originates from physical entity observation. This real restoration of the bevel morphology avoids poor weld formation caused by interpolation errors, making the control commands more consistent with the real physical boundary, and the processing results have higher reliability and process consistency.

[0048] 5. In this invention, the method has strong compatibility and does not require modification to the mechanical structure of existing welding robots when applied. Attached Figure Description

[0049] Figure 1 This is a flowchart of the adaptive correction method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the V-shaped bevel in an embodiment of the present invention; Figure 3This is a schematic diagram showing that the V-shaped bevel of an embodiment of the present invention is approximately trapezoidal; Figure 4 This is a schematic diagram of a track-type welding robot welding a pipe according to an embodiment of the present invention.

[0050] In the diagram: 1. Welding robot, 2. Fixing plate, 3. Line laser profile scanner, 4. Motion track, 5. Pipe, 51. Left pipe body straight line, 52. Right pipe body straight line, 53. Left bevel straight line, 54. Right bevel straight line, 55. Bottom of bevel. Detailed Implementation

[0051] 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 a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0052] In this article, terms such as "left," "right," "up," "down," "front," and "back" are established based on the positional relationships shown in the attached drawings. Depending on the attached drawings, the corresponding positional relationships may also change. Therefore, they should not be interpreted as an absolute limitation on the scope of protection.

[0053] Please see Figures 1 to 3 This invention provides an adaptive correction method for robot visual perception, comprising the following steps: S1: Use a line laser profile scanner 3 to scan the V-shaped bevel pipe at a fixed frequency to obtain a single frame of two-dimensional data of the V-shaped bevel pipe. The fixed frequency is set according to the actual situation.

[0054] Specifically, the line laser emitted by the line laser profile scanner 3 should, as far as possible, pass through the axis of the pipe 5. To avoid interference from the welding arc light, the line laser emitted by the line laser profile scanner 3 should be positioned at a certain distance in front of the welding torch.

[0055] S2: Based on the single-frame two-dimensional data obtained in step S1, the right bevel line 54 and the left bevel line 53 are extracted using a multi-scale operator edge probability space clustering algorithm, and the intersection point and angle of the two bevel lines are calculated. The multi-scale operator edge probability space clustering algorithm performs global gradient field analysis through the Canny operator, performs peak search in the dual space formed by the Hough transform, and uses probabilistic clustering to extract the optimal analytical solution for the bevel boundary.

[0056] Specifically, the process of step S2 includes: S21: Based on the single-frame two-dimensional data obtained in step S1, the edge pixels in the image are extracted using the Canny edge detection algorithm. At this time, these edge pixels are represented as tortuous and continuous pixels.

[0057] S22: For the edge pixels extracted in step S21, use the Hough line detection algorithm to fit them into standard short line segments. After the line fitting is completed, extract the start and end coordinates of each short line segment. S23: Calculate the slope of each short straight line segment, and perform angle clustering based on the slope of the short straight line segment. According to the V-shaped bevel of the V-shaped bevel pipe, the clusters are divided into three types, including the short straight line segment with the highest slope, the short straight line segment with the lowest slope, and the short straight line segment with the middle slope. The bevel line of the V-shaped bevel is represented by the straight line with the highest slope and the straight line with the lowest slope. The short straight line segment with the middle slope is removed, and the short straight line segment with the highest slope and the short straight line segment with the lowest slope are retained. S24: Based on the short line segments with the highest slope and the short line segments with the lowest slope in step S23, the endpoints of the lines are divided into the endpoint group to which the short line with the highest slope belongs and the endpoint group to which the short line with the lowest slope belongs. The two endpoint groups are fitted using the least squares method respectively. The line fitted by the endpoint group to which the short line with the highest slope belongs is the right slope line 54, and the line fitted by the endpoint group to which the short line with the lowest slope belongs is the left slope line 53. S25: Calculate the intersection point and included angle of the right bevel line 54 and the left bevel line 53; Define the equation of the right bevel line 54 as follows:

[0058] in, This represents the slope of the right-hand slope line 54; This represents the intercept of the right bevel line 54; and This represents the coordinates of any point on the straight line 54 at the right bevel. Define the equation of line 53 at the left bevel as follows:

[0059] in, This represents the slope of the straight line 53 at the left bevel. This represents the intercept of the straight line 53 at the left bevel. and This represents the coordinates of any point on the straight line 53 at the left bevel. The intersection point of the right bevel line 54 and the left bevel line 53 is calculated to satisfy:

[0060] in, and Indicates the coordinates of the intersection point; The angle calculations for right bevel line 54 and left bevel line 53 satisfy the following:

[0061] in, This indicates the angle between the right bevel line 54 and the left bevel line 53.

[0062] S3: Remove pixels near the right bevel line 54 and the left bevel line 53. Find the left pipe body line 51 to the left of the left bevel line 53 and the right pipe body line 52 to the right of the right bevel line 54.

[0063] Specifically, step S3 includes: S31: Set a distance threshold range and remove pixels near the right bevel line 54 and the left bevel line 53 that are within the distance threshold range. At this time, only the pipe body pixels are in the figure.

[0064] S32: Among the remaining pixels, separate the pixels to the left of the left bevel line 53 and the pixels to the right of the right bevel line 54, using the following formula:

[0065] in, The discriminant value; This indicates the slope of the left bevel line 53 or the right bevel line 54; This indicates the intercept of the left bevel line 53 or the right bevel line 54; and Represents the coordinates of a pixel; like The pixel is to the right of line 54 at the right slope. The pixel is to the left of line 53 at the left slope. S33: Based on the pixels to the left of the left bevel line 53 and the pixels to the right of the right bevel line 54 selected in step S32, the pixels to the left of the left bevel line 53 are fitted into a straight line using the least squares method, which is the left pipe body straight line 51. The pixels to the right of the right bevel line 54 are fitted into a straight line using the least squares method, which is the right pipe body straight line 52.

[0066] S4: Calculate the bevel depth and the number of weld layers.

[0067] Specifically, step S4 includes: S41: Define the equation of the left tube straight line 51 as follows:

[0068] in, This represents the slope of the straight line 51 on the left side of the tube; This indicates the intercept of the straight line 51 on the left side of the tube; and This represents the coordinates of any point on the straight line 51 of the left tube. Define the equation of straight line 52 on the right tube as follows:

[0069] in, This represents the slope of the straight line 52 on the right side of the tube; This represents the intercept of the straight line 52 on the right side of the tube. and This represents the coordinates of any point on the straight line 52 of the right tube. Calculate the distance from the intersection of the right bevel line 54 and the left bevel line 53 to the left pipe body line 51 and the right pipe body line 52, and take the smaller value as the bevel depth. The formula is:

[0070] in, Indicates the bevel depth; S42: Calculate the number of weld layers using the following formula:

[0071] in, Indicates the number of welding layers; This indicates the maximum allowable thickness for a single weld pass.

[0072] S5: Since the line laser emitted by the line laser profile scanner 3 is always in front of the welding torch, it is necessary to perform hand-eye calibration on the line laser profile scanner 3 and the welding torch before welding.

[0073] Specifically, step S5 includes: A front-end sensing field is constructed using a line laser profile scanner 3. The nonlinear mapping transformation from the pixel coordinate system of the line laser profile scanner 3 to the execution space coordinate system of the welding torch is completed by the industrial control computer. This compensates for the physical lag in the welding process and ensures that the welding path point in the coordinate system of the line laser profile scanner 3 strictly corresponds to the execution target point of the welding torch in the base coordinate system of the welding robot 1.

[0074] The mapping relationship is as follows:

[0075]

[0076] in, The instantaneous time of data acquisition by the line laser profile scanner 3; Indicates the actual arrival of the welding torch. The physical location and time point scanned by the timeline laser contour scanner 3; The welding path points in the 3-coordinate system of the line laser profile scanner; The target point for the welding torch in the welding robot's 1-base coordinate system; Hand-eye calibration matrix; The real-time pose matrix for welding robot 1; This indicates the distance that can be perceived in advance, as determined by hand-eye calibration. This indicates the speed of the welding robot 1. Indicates the time of delayed compensation.

[0077] S6: Calculate the position of the intersection of the right bevel line 54 and the left bevel line 53 in real time for each frame of two-dimensional data. Based on the bevel depth and the real-time position of the intersection in step S4, adjust the position of the welding torch in real time to ensure that the welding torch is always directly above the intersection and at a certain distance.

[0078] Specifically, step S6 includes: S61: Based on the position calculation of the intersection point in step S25, calculate the position of the intersection point of the right slope line 54 and the left slope line 53 in each frame of two-dimensional data in real time. S62: The perpendicular line from the intersection of the right bevel line 54 and the left bevel line 53 to the left pipe body line 51 is denoted as... The perpendicular line from the intersection of the right bevel line 54 and the left bevel line 53 to the right pipe body line 52 is denoted as... Keep the shorter of the two perpendicular lines, denoted as ,satisfy:

[0079]

[0080]

[0081] S63: The slope of the straight line of the tube corresponding to the short perpendicular line satisfies:

[0082] in, This indicates the slope of the straight line in the pipe corresponding to the short perpendicular line; S64: The welding torch is always directly above the intersection point and maintains a certain distance. This distance is defined as the welding height (determined by the specific welding process). Based on the intersection point position and the welding height, the welding path point in the 3D coordinate system of the line laser profile scanner is obtained. The formula is:

[0083] in, Indicates the welding height; Based on the welding path points in the 3-coordinate system of the line laser contour scanner, the execution target point of the welding torch in the 1-base coordinate system of the welding robot is obtained, using the following formula: .

[0084] S7: Perform the first weld of the V-groove. After the first weld is completed, the V-groove will appear as an approximate trapezoid.

[0085] S8: Use a line laser profile scanner 3 to scan the V-shaped bevel pipe at a fixed frequency to obtain a single frame of two-dimensional data of the V-shaped bevel pipe.

[0086] S9: Based on the single-frame two-dimensional data from step S8, the right slope line 54 and the left slope line 53 are extracted using the edge probability spatial clustering algorithm with multi-scale operators.

[0087] Specifically, step S9 includes: S91: Based on the single-frame two-dimensional data obtained in step S8, the edge pixels in the image are extracted using the Canny edge detection algorithm; S92: For the edge pixels extracted in step S81, use the Hough line detection algorithm to fit them into standard short line segments. After the line fitting is completed, extract the start and end coordinates of each short line segment. S93: Calculate the slope of each short straight line segment, and perform angle clustering based on the slope of the short straight line segments. According to the approximate trapezoidal shape of the V-shaped bevel pipe, the clusters are divided into three types, including the short straight line segments with the highest slope, the short straight line segments with the lowest slope, and the short straight line segments with the middle slope. The bevel line is represented by the straight line with the highest slope and the straight line with the lowest slope. The short straight line segments with the middle slope are removed, and the short straight line segments with the highest slope and the short straight line segments with the lowest slope are retained. S94: Based on the short line segments with the highest and lowest slopes in step S93, the endpoints of the lines are divided into the endpoint group to which the short line with the highest slope belongs and the endpoint group to which the short line with the lowest slope belongs. The two endpoint groups are fitted using the least squares method. The line fitted by the endpoint group to which the short line with the highest slope belongs is the right slope line 54, and the line fitted by the endpoint group to which the short line with the lowest slope belongs is the left slope line 53.

[0088] S10: Remove pixels near the right bevel line 54 and the left bevel line 53. Find the left pipe body line 51 to the left of the left bevel line 53 and the right pipe body line 52 to the right of the right bevel line 54. Then remove pixels near the left pipe body line 51 and the right pipe body line 52. The remaining pixels are the bottom pixels of the bevel.

[0089] Specifically, step S10 includes: S101: Set a distance threshold range and remove pixels near the right bevel line 54 and the left bevel line 53 that are within the distance threshold range; S102: Among the remaining pixels, separate the pixels to the left of the left bevel line 53 and the pixels to the right of the right bevel line 54, using the following formula:

[0090] in, The discriminant value; This indicates the slope of the left bevel line 53 or the right bevel line 54; This indicates the intercept of the left bevel line 53 or the right bevel line 54; and Represents the coordinates of a pixel; like The pixel is to the right of line 54 at the right slope. The pixel is to the left of line 53 at the left slope. S103: Based on the pixels to the left of the left bevel line 53 and the pixels to the right of the right bevel line 54 selected in step S102, the pixels to the left of the left bevel line 53 are fitted into a straight line using the least squares method, which is the left pipe body straight line 51. The pixels to the right of the right bevel line 54 are fitted into a straight line using the least squares method, which is the right pipe body straight line 52. S104: Set a distance threshold range and remove pixels that are within the distance threshold range near the straight line 51 on the left pipe body and the straight line 52 on the right pipe body. The remaining pixels are the bottom pixels of the bevel.

[0091] S11: Use the manifold sampling algorithm to sample the pixels at the bottom of the bevel, obtain several physical sampling points, calculate the position of each physical sampling point, and adjust the welding gun position in real time according to the position of each sampling point.

[0092] Specifically, step S11 includes: S111: Let the axial direction of pipe 5 be... The axis is used to calculate the range of pixels at the bottom of the bevel along the 5-axis of the pipe, satisfying the following:

[0093] in, This indicates the axial coordinate of the starting pixel at the bottom left of the bevel (55). This indicates the axial coordinate of the leftmost pixel at the bottom of the bevel (55). Generate on this interval A sampling point covering the entire width of the bevel, satisfying:

[0094] in, Indicates the first The axial coordinates of each sampling point; Indicates an index; S112: Using the sampling points as guides, perform a nearest neighbor search within the pixels of the actual feature pixels at the bottom of the bevel to extract physical sampling points that fit the actual shape of the bevel, thereby correcting the deviation between the sampling points and the actual workpiece. Satisfy:

[0095] in, These represent physical sampling points that closely match the actual shape of the bevel. This represents the set of pixels representing the true features at the bottom of the bevel. This represents the actual feature pixels at the bottom of the bevel to be compared; Indicates the axial pixel coordinate; And record the coordinates of each physical sampling point. ,in, Represents pixel coordinates in the depth direction; S113: Define the equation of the left tube straight line 51 as follows:

[0096] in, This represents the slope of the straight line 51 on the left side of the tube; This indicates the intercept of the straight line 51 on the left side of the tube; and This represents the coordinates of any point on the straight line 51 of the left tube. Define the equation of straight line 52 on the right tube as follows:

[0097] in, This represents the slope of the straight line 52 on the right side of the tube; This represents the intercept of the straight line 52 on the right side of the tube. and This represents the coordinates of any point on the straight line 52 of the right tube. The perpendicular line from each physical sampling point to the straight line 51 on the left pipe body is denoted as... The perpendicular line from each physical sampling point to the straight line 52 on the right pipe body is denoted as... Keep the shorter of the two perpendicular lines, denoted as ,satisfy:

[0098]

[0099]

[0100] S114: The slope of the straight line of the pipe body corresponding to the short perpendicular line satisfies:

[0101] in, This indicates the slope of the straight line in the pipe corresponding to the short perpendicular line; S115: The welding torch is always positioned above the physical sampling point of each frame of data and maintains a certain distance. This distance is defined as the welding height (determined by the welding process). Based on the physical sampling point position and the welding height, the welding path point in the 3D coordinate system of the line laser profile scanner is obtained. The formula is:

[0102] in, Indicates the welding height; Based on the welding path points in the 3-coordinate system of the line laser contour scanner, the execution target point of the welding torch in the 1-base coordinate system of the welding robot is obtained, using the following formula: .

[0103] S12: Repeat steps S8-S11 to weld layer by layer.

[0104] Specifically, this invention constructs an adaptive correction method with a closed-loop feature of perception-decision-execution. This method can calculate welding deviations and correct path parameters in real time, upgrading traditional blind welding to dynamic tracking. This significantly improves the robustness and automation level of welding operations under complex conditions, effectively solving the problems of large fluctuations and poor stability in welding accuracy that exist in the industry. This method uses a multi-scale operator edge probability space clustering algorithm to extract the right bevel line 54 and the left bevel line 53. By fragmenting the two-dimensional contour information and using Hough short line fitting in the dual space to reconstruct the endpoints, it can extract sub-pixel-level feature points from the original image full of noise interference. This greatly enhances the anti-distortion ability and analytical accuracy of the recognition algorithm, and achieves high-precision decoupling of the bevel line segments. This method constructs a nonlinear mapping transformation with "sensing ahead and execution lagging," avoiding nonlinear interference from electromagnetic and optical fields such as strong arc light and high-temperature fumes at the physical level. Combined with efficient digital image parallel processing logic, it can achieve millisecond-level rapid parsing of single-frame data, injecting key feature parameters into the control core in real time, forming a correction feedback chain with an extremely high refresh rate, perfectly adapting to the stringent real-time response requirements of high-speed welding. This method adopts a manifold sampling algorithm based on manifold thinking. Unlike traditional algorithms that rely on mathematical interpolation to generate pseudo-observation points, the manifold sampling algorithm strictly distributes points uniformly within the real edge point cloud space, ensuring that each control sampling point originates from physical entity observation. This realistic restoration of the bevel morphology avoids poor weld formation caused by interpolation errors, making the control commands more closely match the real physical boundaries, and the processing results have higher reliability and process consistency. This method has strong compatibility and does not require modification to the mechanical structure of existing welding robots.

[0105] For details, please refer to Figure 4 , Figure 4 The diagram shows welding robot 1 welding the V-groove of pipe 5, which is a V-groove pipe. A motion track 4 is installed on pipe 5, near the V-groove. Welding robot 1 is slidably mounted on the motion track 4, allowing it to move along it. A fixing plate 2 is located on one side of welding robot 1, and a line laser profile scanner 3 is mounted on the fixing plate 2.

[0106] The above embodiments merely illustrate the basic principles and characteristics of the present invention, but are not limited to the above implementation schemes. It should be understood that those skilled in the art can make various changes and modifications to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive correction method for robot visual perception, characterized in that, Includes the following steps: S1: Use a line laser profile scanner (3) to scan the V-shaped bevel pipe at a fixed frequency to obtain single-frame two-dimensional data of the V-shaped bevel pipe; S2: Based on the single-frame two-dimensional data obtained in step S1, the right bevel line (54) and the left bevel line (53) are extracted using the edge probability spatial clustering algorithm of multi-scale operators, and the intersection point and angle of the two bevel lines are calculated. S3: Remove the pixels near the right bevel line (54) and the left bevel line (53), find the left pipe body line (51) to the left of the left bevel line (53), and find the right pipe body line (52) to the right of the right bevel line (54). S4: Calculate the bevel depth and the number of weld layers; S5: Perform hand-eye calibration of the line laser profile scanner (3) and welding torch; S6: Calculate the position of the intersection of the right bevel line (54) and the left bevel line (53) in each frame of two-dimensional data in real time, and adjust the welding gun position in real time according to the bevel depth and the real-time position of the intersection in step S4. S7: Perform the first weld of the V-groove. After the first weld is completed, the V-groove will appear as an approximate trapezoid. S8: Use a line laser profile scanner (3) to scan the V-shaped bevel pipe at a fixed frequency to obtain single-frame two-dimensional data of the V-shaped bevel pipe; S9: Based on the single-frame two-dimensional data from step S8, the right slope line (54) and the left slope line (53) are extracted using the edge probability space clustering algorithm of multi-scale operators. S10: Remove the pixels near the right bevel line (54) and the left bevel line (53), find the left pipe body line (51) to the left of the left bevel line (53), find the right pipe body line (52) to the right of the right bevel line (54), and then remove the pixels near the left pipe body line (51) and the right pipe body line (52). The remaining points are the pixels at the bottom of the bevel. S11: Use the manifold sampling algorithm to sample the pixels at the bottom of the bevel, obtain several physical sampling points, calculate the position of each physical sampling point, and adjust the welding gun position in real time according to the position of each sampling point. S12: Repeat steps S8-S11 to weld layer by layer.

2. The adaptive correction method for robot visual perception according to claim 1, characterized in that, The specific process of step S2 includes: S21: Based on the single-frame two-dimensional data obtained in step S1, use the Canny edge detection algorithm to extract edge pixels in the image; S22: For the edge pixels extracted in step S21, use the Hough line detection algorithm to fit them into standard short line segments. After the line fitting is completed, extract the start and end coordinates of each short line segment. S23: Calculate the slope of each short straight line segment, and perform angle clustering based on the slope of the short straight line segment. According to the V-shaped bevel of the V-shaped bevel pipe, the clusters are divided into three types, including the short straight line segment with the highest slope, the short straight line segment with the lowest slope, and the short straight line segment with the middle slope. The bevel line of the V-shaped bevel is represented by the straight line with the highest slope and the straight line with the lowest slope. The short straight line segment with the middle slope is removed, and the short straight line segment with the highest slope and the short straight line segment with the lowest slope are retained. S24: Based on the short line segments with the highest slope and the short line segments with the lowest slope in step S23, the endpoints of the lines are divided into the endpoint group to which the short line with the highest slope belongs and the endpoint group to which the short line with the lowest slope belongs. The two endpoint groups are fitted using the least squares method. The line fitted by the endpoint group to which the short line with the highest slope belongs is the right slope line (54), and the line fitted by the endpoint group to which the short line with the lowest slope belongs is the left slope line (53). S25: Calculate the intersection point and included angle of the right slope line (54) and the left slope line (53); The equation of the right bevel line (54) is defined as follows: in, This represents the slope of the right-hand slope line (54); This represents the intercept of the right bevel line (54); and Represents the coordinates of any point on the right slope line (54); The equation of the left slope line (53) is defined as follows: in, This represents the slope of the straight line (53) at the left bevel. This represents the intercept of the left bevel line (53); and Represents the coordinates of any point on the straight line (53) at the left slope; The intersection point of the right bevel line (54) and the left bevel line (53) is calculated to satisfy: in, and Indicates the coordinates of the intersection point; The angle calculations for the right bevel line (54) and the left bevel line (53) satisfy the following: in, This indicates the angle between the right slope line (54) and the left slope line (53).

3. The adaptive correction method for robot visual perception according to claim 2, characterized in that, Step S3 specifically includes: S31: Set a distance threshold range and remove pixels that are within the distance threshold range near the right slope line (54) and the left slope line (53); S32: Among the remaining pixels, separate the pixels to the left of the left bevel line (53) and the pixels to the right of the right bevel line (54), using the following formula: in, The discriminant value; This indicates the slope of the left slope line (53) or the right slope line (54); This indicates the intercept of the left bevel line (53) or the right bevel line (54); and Represents the coordinates of a pixel; like The pixel is to the right of the straight line (54) at the right slope. The pixel is to the left of the straight line (53) at the left slope; S33: Based on the pixels to the left of the left bevel line (53) and the pixels to the right of the right bevel line (54) selected in step S32, the pixels to the left of the left bevel line (53) are fitted into a straight line using the least squares method, which is the left pipe body straight line (51). The pixels to the right of the right bevel line (54) are fitted into a straight line using the least squares method, which is the right pipe body straight line (52).

4. The adaptive correction method for robot visual perception according to claim 2, characterized in that, Step S4 specifically includes: S41: Define the equation of the left tube straight line (51) as follows: in, This represents the slope of the straight line (51) on the left side of the tube; This represents the intercept of the straight line (51) on the left side of the tube; and Represents the coordinates of any point on the straight line (51) of the left tube; The equation of the right tube straight line (52) is defined as follows: in, This represents the slope of the straight line (52) on the right side of the tube; This represents the intercept of the right tube straight line (52); and Represents the coordinates of any point on the straight line (52) of the right tube; Calculate the distance from the intersection of the right bevel line (54) and the left bevel line (53) to the left pipe body line (51) and the right pipe body line (52), and take the smaller value as the bevel depth. The formula is: in, Indicates the bevel depth; S42: Calculate the number of weld layers using the following formula: in, Indicates the number of welding layers; This indicates the maximum allowable thickness for a single weld pass.

5. The adaptive correction method for robot visual perception according to claim 2, characterized in that, Step S5 specifically includes: A front-end sensing field is constructed using a line laser profile scanner (3), and a nonlinear mapping transformation from the pixel coordinate system of the line laser profile scanner (3) to the execution space coordinate system of the welding torch is completed by an industrial control computer. The mapping relationship is as follows: in, The instantaneous time of data acquisition by the line laser profile scanner (3); Indicates the actual arrival of the welding torch. The physical location time point scanned by the timeline laser profile scanner (3); The welding path points are in the coordinate system of the line laser profile scanner (3); The target point for the welding torch in the base coordinate system of the welding robot (1); Hand-eye calibration matrix; For the welding robot (1) real-time pose matrix; This indicates the distance that can be perceived in advance, as determined by hand-eye calibration. This indicates the speed of movement of the welding robot (1); Indicates the time of delayed compensation.

6. The adaptive correction method for robot visual perception according to claim 4, characterized in that, Step S6 specifically includes: S61: Based on the position calculation of the intersection point in step S25, calculate in real time the position of the intersection point of the right slope line (54) and the left slope line (53) in each frame of two-dimensional data; S62: The perpendicular line from the intersection of the right bevel line (54) and the left bevel line (53) to the left pipe body line (51) is denoted as S62. The perpendicular line from the intersection of the right bevel line (54) and the left bevel line (53) to the right pipe body line (52) is denoted as... Keep the shorter of the two perpendicular lines, denoted as ,satisfy: S63: The slope of the straight line of the tube corresponding to the short perpendicular line satisfies: in, This indicates the slope of the straight line in the pipe corresponding to the short perpendicular line; S64: The welding torch is always directly above the intersection point and maintains a certain distance. This distance is defined as the welding height. Based on the intersection point position and the welding height, the welding path point in the coordinate system of the line laser profile scanner (3) is obtained. The formula is: in, Indicates the welding height; Based on the welding path points in the coordinate system of the line laser profile scanner (3), the execution target point of the welding torch in the base coordinate system of the welding robot (1) is obtained, and the formula is: 。 7. The adaptive correction method for robot visual perception according to claim 1, characterized in that, Step S9 specifically includes: S91: Based on the single-frame two-dimensional data obtained in step S8, the edge pixels in the image are extracted using the Canny edge detection algorithm; S92: For the edge pixels extracted in step S81, use the Hough line detection algorithm to fit them into standard short line segments. After the line fitting is completed, extract the start and end coordinates of each short line segment. S93: Calculate the slope of each short straight line segment, and perform angle clustering based on the slope of the short straight line segments. According to the approximate trapezoidal shape of the V-shaped bevel pipe, the clusters are divided into three types, including the short straight line segments with the highest slope, the short straight line segments with the lowest slope, and the short straight line segments with the middle slope. The bevel line is represented by the straight line with the highest slope and the straight line with the lowest slope. The short straight line segments with the middle slope are removed, and the short straight line segments with the highest slope and the short straight line segments with the lowest slope are retained. S94: Based on the short line segments with the highest slope and the short line segments with the lowest slope in step S93, the endpoints of the lines are divided into the endpoint group to which the short line with the highest slope belongs and the endpoint group to which the short line with the lowest slope belongs. The two endpoint groups are fitted using the least squares method. The line fitted by the endpoint group to which the short line with the highest slope belongs is the right slope line (54), and the line fitted by the endpoint group to which the short line with the lowest slope belongs is the left slope line (53).

8. The adaptive correction method for robot visual perception according to claim 7, characterized in that, Step S10 specifically includes: S101: Set the distance threshold range and remove pixels that are within the distance threshold range near the right slope line (54) and the left slope line (53); S102: Among the remaining pixels, separate the pixels to the left of the left bevel line (53) and the pixels to the right of the right bevel line (54), using the following formula: in, The discriminant value; This indicates the slope of the left slope line (53) or the right slope line (54); This indicates the intercept of the left bevel line (53) or the right bevel line (54); and Represents the coordinates of a pixel; like The pixel is to the right of the straight line (54) at the right slope. The pixel is to the left of the straight line (53) at the left slope; S103: Based on the pixels to the left of the left bevel line (53) and the pixels to the right of the right bevel line (54) selected in step S102, the pixels to the left of the left bevel line (53) are fitted into a straight line using the least squares method, which is the left pipe body straight line (51). The pixels to the right of the right bevel line (54) are fitted into a straight line using the least squares method, which is the right pipe body straight line (52). S104: Set the distance threshold range and remove the pixels near the left tube line (51) and the right tube line (52) that are within the distance threshold range. At this time, the remaining pixels are the bottom pixels of the bevel.

9. The adaptive correction method for robot visual perception according to claim 8, characterized in that, Step S11 specifically includes: S111: Let the axial direction of pipe (5) be... Calculate the range of pixels at the bottom of the bevel along the axial direction of the pipe (5), satisfying: in, This indicates the axial coordinate of the starting pixel on the left side of the bottom of the bevel (55); Indicates the axial coordinate of the left end pixel at the bottom of the bevel (55); Generate on this interval A sampling point covering the entire width of the bevel, satisfying: in, Indicates the first The axial coordinates of each sampling point; Indicates an index; S112: Using the sampling points as guides, perform a nearest neighbor search within the pixels of the actual feature pixels at the bottom of the bevel to extract physical sampling points that fit the actual shape of the bevel, satisfying: in, These represent physical sampling points that closely match the actual shape of the bevel. This represents the set of pixels representing the true features at the bottom of the bevel. This represents the actual feature pixels at the bottom of the bevel to be compared; Indicates the axial pixel coordinate; And record the coordinates of each physical sampling point. ,in, Represents pixel coordinates in the depth direction; S113: Define the equation of the left tube straight line (51) as follows: in, This represents the slope of the straight line (51) on the left side of the tube; This represents the intercept of the straight line (51) on the left side of the tube; and Represents the coordinates of any point on the straight line (51) of the left tube; The equation of the right tube straight line (52) is defined as follows: in, This represents the slope of the straight line (52) on the right side of the tube; This represents the intercept of the right tube straight line (52); and Represents the coordinates of any point on the straight line (52) of the right tube; The perpendicular line from each physical sampling point to the straight line (51) on the left pipe body is denoted as... The perpendicular line from each physical sampling point to the straight line (52) of the right pipe body is denoted as... Keep the shorter of the two perpendicular lines, denoted as ,satisfy: S114: The slope of the straight line of the pipe body corresponding to the short perpendicular line satisfies: in, This indicates the slope of the straight line in the pipe corresponding to the short perpendicular line; S115: The welding torch is always above the physical sampling point and maintains a certain distance. This distance is defined as the welding height. Based on the position of the physical sampling point and the welding height, the welding path point in the coordinate system of the line laser profile scanner (3) is obtained. The formula is: in, Indicates the welding height; Based on the welding path points in the coordinate system of the line laser profile scanner (3), the execution target point of the welding torch in the base coordinate system of the welding robot (1) is obtained, and the formula is: 。

Images