A laser vision-based T-type corner weld identification method and system

Abstract

The application discloses a T-shaped corner weld identification method and system based on laser vision, and belongs to the technical field of automatic welding and machine vision, and comprises the following steps: collecting images, performing enhancement processing on the collected images, and extracting an effective laser stripe area; center lines are extracted from the effective laser stripe area, and the center line coordinates are converted into three-dimensional point cloud data; the three-dimensional point cloud data is subjected to geometric feature optimization processing, and a RANSAC algorithm with spatial geometric constraints is used to fit and identify the geometric features of a T-shaped corner weld structure; based on the geometric features of the T-shaped corner weld structure, the three-dimensional coordinates of the key nodes of the weld are calculated through linear end point distance constraints, three-dimensional intersection calculation and Z-axis height screening. The application can realize automatic and high-precision acquisition of the position of the weld, realize automatic conversion of the weld from a two-dimensional image to a three-dimensional robot coordinate system, and significantly improve the degree of welding automation.

Description

Technical Field

[0001] This invention relates to the fields of automated welding and machine vision technology, and in particular to a method and system for identifying T-shaped fillet welds based on laser vision. Background Technology

[0002] In automated welding processes in shipbuilding, steel structures, and construction machinery, accurate identification of T-fillet welds is crucial for achieving automated welding. Currently, laser vision-based weld identification technology is widely used, but existing technologies have a prominent problem: they cannot effectively determine the unique geometric relationships of T-fillet welds.

[0003] Specifically, while existing laser vision weld identification methods can extract laser stripes and locate individual welds, they generally lack a mechanism for recognizing the spatial perpendicular relationship between two welds in a T-type fillet weld. In complex industrial environments, due to the lack of effective discrimination of this specific geometric relationship, existing technologies struggle to accurately distinguish T-type fillet welds from other types of weld structures, leading to unreliable weld identification results and directly impacting the accuracy of subsequent welding path planning. Summary of the Invention

[0004] In view of the above-mentioned shortcomings in the current field of automated welding and machine vision technology, the present invention provides a T-shaped fillet weld recognition method and system based on laser vision, which can realize the automatic and high-precision acquisition of weld position, realize the automatic conversion of weld from two-dimensional image to three-dimensional robot coordinate system, and significantly improve the degree of welding automation.

[0005] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:

[0006] A method for identifying T-shaped fillet welds based on laser vision includes the following steps:

[0007] Acquire images, enhance the acquired images, and extract the effective laser stripe areas;

[0008] Extract the center line from the effective laser stripe area and convert the center line coordinates into three-dimensional point cloud data;

[0009] The geometric features of the three-dimensional point cloud data are optimized, and the geometric features of the T-shaped fillet weld structure are identified by fitting based on the RANSAC algorithm with spatial geometric constraints.

[0010] Based on the geometric features of the T-shaped fillet weld structure, the three-dimensional coordinates of the key nodes of the weld are calculated by means of straight line endpoint distance constraints, three-dimensional intersection calculation, and Z-axis height screening.

[0011] According to one aspect of the present invention, the enhancement processing of the acquired image and the extraction of the effective laser stripe region includes:

[0012] Convert the color image to a grayscale image and then perform Gaussian filtering for smoothing.

[0013] A one-dimensional linear filtering operator is used to perform convolution operations on the image;

[0014] The filtering results are binarized, and effective laser stripe regions are selected through connected component analysis.

[0015] According to one aspect of the present invention, the enhancement processing of the acquired image can be implemented in a variety of ways, such as, but not limited to, convolution enhancement based on specific weight operators, or edge detection enhancement based on Sobel operators.

[0016] According to one aspect of the present invention, the extraction of the center line from the effective laser stripe region is performed using a distance transformation method, specifically including:

[0017] Calculate the distance from each foreground pixel to the nearest background pixel;

[0018] Scan the distance map row by row and take the pixel with the largest distance value in each row as the center point;

[0019] A sequence of center points is formed that is continuously distributed along the weld seam.

[0020] According to one aspect of the present invention, the conversion of centerline coordinates into three-dimensional point cloud data includes:

[0021] The pixel coordinates are transformed to the camera coordinate system using the camera intrinsic parameter matrix and the laser plane equation.

[0022] Transform the coordinates to the robot coordinate system using the hand-eye matrix and the robotic arm pose matrix.

[0023] According to one aspect of the present invention, the geometric feature optimization processing of the three-dimensional point cloud data includes:

[0024] Perform Euclidean cluster analysis on the 3D point cloud to screen effective weld point cloud clusters;

[0025] The point cloud was smoothed using moving average and Gaussian kernel covariance filtering.

[0026] Spatial outliers are eliminated based on radius constraints.

[0027] According to one aspect of the present invention, the geometric features of the T-type fillet weld structure fitted and identified based on the RANSAC algorithm with spatial geometric constraints include:

[0028] Multiple straight lines are iteratively fitted using the RANSAC algorithm;

[0029] Calculate the spatial angle between the direction vectors of each line;

[0030] When there is a set of straight lines whose included angles deviate from 90° less than a set threshold, it is determined to be a T-type fillet weld.

[0031] According to one aspect of the invention, the set threshold is 10°, and the number of interior points of the fitted line is greater than 100.

[0032] According to one aspect of the present invention, the calculation of the three-dimensional coordinates of the key nodes of the weld through straight line endpoint distance constraints, three-dimensional intersection calculation, and Z-axis height screening includes:

[0033] Calculate the intersection point of two straight lines in three-dimensional space;

[0034] The validity of intersection points is determined by distance threshold constraints;

[0035] The intersection point with the largest Z-coordinate value is selected as the reference point for the weld.

[0036] According to one aspect of the present invention, before calculating the three-dimensional coordinates of the key nodes of the weld through straight line endpoint distance constraints, three-dimensional intersection calculation, and Z-axis height screening, a geometric verification step is further included:

[0037] Calculate the minimum Euclidean distance between the endpoints of two straight lines;

[0038] When the minimum distance is less than or equal to a preset threshold, it is determined to be a valid weld structure, and the step of calculating the three-dimensional coordinates of the key nodes of the weld is performed; when the minimum distance is greater than the preset threshold, it is determined to be an invalid weld structure and the process is terminated.

[0039] A laser vision-based T-shaped fillet weld recognition system, implemented based on the aforementioned laser vision-based T-shaped fillet weld recognition method, includes:

[0040] The acquisition module acquires images, enhances the acquired images, and extracts the effective laser stripe areas.

[0041] The extraction module extracts the center line from the effective laser stripe area and converts the center line coordinates into three-dimensional point cloud data;

[0042] The feature module performs geometric feature optimization processing on the three-dimensional point cloud data and identifies the geometric features of the T-shaped fillet weld structure based on the RANSAC algorithm with spatial geometric constraints.

[0043] The calculation module, based on the geometric features of the T-shaped fillet weld structure, calculates the three-dimensional coordinates of the key nodes of the weld through straight line endpoint distance constraints, three-dimensional intersection calculation, and Z-axis height screening.

[0044] Advantages of this invention: This invention organically combines laser stripe extraction, centerline refinement, 3D reconstruction, point cloud optimization, and geometric structure discrimination to form a complete, closed-loop weld identification method. This method enables automatic and accurate identification and positioning of T-shaped fillet welds, effectively solving the core problem of existing technologies' inability to discern weld geometric relationships. By strengthening the response of the weld center region through a linear operator with specific weight distribution, background interference is effectively suppressed, improving the stability of laser stripe extraction. Combined with Euclidean clustering, Gaussian kernel covariance filtering, and radius-constrained outlier removal, the quality of the weld point cloud is improved, enhancing system robustness. Through multi-line fitting and angle constraints, the geometric features of T-shaped fillet welds are accurately identified. This method, through enhanced laser stripe extraction, 3D point cloud optimization and reconstruction, RANSAC-based linear feature extraction, and spatial geometric constraint discrimination, significantly improves the accuracy and robustness of T-shaped fillet weld identification in complex welding environments, providing reliable key node coordinates for automated welding path planning, ultimately achieving the beneficial effect of improving welding quality and efficiency. Attached Figure Description

[0045] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating a laser vision-based T-shaped fillet weld identification method according to Embodiments 1 and 2 of the present invention.

[0047] Figure 2 This is a schematic flowchart of a T-shaped fillet weld recognition system based on laser vision, as described in Embodiment 3 of the present invention.

[0048] Figure 3 These are laser visual images of the T-shaped weld seam described in Embodiments 1 and 2 of the present invention;

[0049] Figure 4 This is the original image for comparing the filter effects described in Embodiment 1 of the present invention;

[0050] Figure 5 This is a custom filter diagram comparing the filter effects described in Embodiment 1 of the present invention;

[0051] Figure 6 This is a Gaussian filter diagram comparing the effects of the filter described in Embodiment 1 of the present invention;

[0052] Figure 7 This is the visual image of laser centerline extraction as described in Embodiment 1 of the present invention;

[0053] Figure 8 This is a visual image of the point cloud after Euclidean clustering as described in Embodiment 1 of the present invention.

[0054] Figure 9 This is the original pixel point cloud image described in Embodiment 1 of the present invention;

[0055] Figure 10 This is the pixel point cloud map after window + Gaussian smoothing as described in Embodiment 1 of the present invention;

[0056] Figure 11 This is a visual image of the smoothed three-dimensional centerline of the weld as described in Embodiment 1 of the present invention;

[0057] Figure 12 This is a diagram illustrating the effect of laser centerline extraction as described in Embodiment 1 of the present invention.

[0058] Figure 13 This is a rendering of the unconstrained identification points described in Embodiment 1 of the present invention;

[0059] Figure 14 This is a schematic diagram of a two-dimensional weld point as described in Embodiment 1 of the present invention. Detailed Implementation

[0060] 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.

[0061] Example 1

[0062] like Figure 1 As shown, a method for identifying T-shaped fillet welds based on laser vision is described, the method comprising the following steps:

[0063] Step S1: Acquire an image, perform enhancement processing on the acquired image, and extract the effective laser stripe area;

[0064] Step S1 involves acquiring an image, enhancing the acquired image, and extracting the effective laser stripe region, specifically including:

[0065] The implementation conditions of this embodiment include at least the following: a vision sensor, consisting of a camera, a line laser, and filters, is mounted on the flange of a robotic arm; the camera's intrinsic parameters are calibrated using the Zhang Zhengyou calibration method; the laser plane's representation in the camera coordinate system is fitted using the least squares method; and the hand-eye transformation matrix is ​​calibrated based on the pointing method. The robotic arm moves the vision sensor to scan the T-shaped flat weld seam, and the camera captures 2D images of the laser illuminating the weld seam in real time at a preset frame rate. Figure 3 As shown, the coordinates of the robotic arm flange at the time of shooting are acquired synchronously.

[0066] The process involves acquiring and inputting a 2D image. First, channel detection is performed on the input 2D image to determine if it is a color image. If the input is a color image, it is converted to a grayscale image. Then, a 3*3 Gaussian filter is applied to the grayscale image for smoothing to suppress imaging noise and sudden changes in local brightness.

[0067] Based on this, a one-dimensional custom linear filtering operator [-1-1-1-1161-1-1-1-1] is used to perform convolution operations on the image. This operator enhances the grayscale response of the weld center region with a larger center weight, while suppressing the areas on both sides of the weld and the background region through negative weights, thereby significantly enhancing the weld strip structure and reducing background interference. After filtering, the enhancement result is binarized using a fixed threshold, retaining only the high grayscale response region, so that the laser stripes are initially separated from the background.

[0068] Next, connected component analysis is performed on the binary image to statistically analyze the area of ​​each connected region. Small, isolated noise regions are then removed based on a preset minimum area threshold, retaining only candidate laser stripe regions with continuity and a certain scale. Finally, the filtered connected components are synthesized into an output binary image, thus extracting the effective laser stripe regions.

[0069] like Figure 4 , Figure 5 and Figure 6 The diagram shown is a comparison of the filtering effect of this embodiment with that of a conventional Gaussian filter.

[0070] In practical applications, the enhancement process can employ Gaussian filter kernels of different sizes, and the connected component analysis can be combined with morphological opening operations to further remove noise points.

[0071] Step S2: Extract the center line from the effective laser stripe area and convert the center line coordinates into three-dimensional point cloud data;

[0072] Step S2 extracts the center line from the effective laser stripe region and converts the center line coordinates into three-dimensional point cloud data, including:

[0073] The center line is extracted from the effective laser stripe region using the distance transformation method, specifically including:

[0074] Based on the binarized image obtained in step S1, the laser center point is extracted using the distance transform method. The minimum distance from each foreground pixel (255) to the nearest background pixel (0) is calculated and normalized, mapping the distance values ​​uniformly to the 0-1 interval. The distance map is scanned row by row, and the pixel with the largest distance value is searched in each row. This point corresponds to the geometric center position of the laser stripe in that row. If there is a valid foreground response in the current row, the pixel coordinates corresponding to the maximum distance are saved as the center point. By extracting the maximum distance points row by row, a set of center point sequences continuously distributed along the weld seam is finally obtained, realizing the refinement and positioning of the laser center line, such as... Figure 7 As shown.

[0075] The extracted laser center pixel can be transformed into the camera coordinate system based on the camera intrinsic parameter matrix M and the laser plane matrix P. Transform the hand-eye matrix H and the corresponding TCP transformation matrix T of the image into the robot coordinate system. .

[0076] The conversion formula is as follows:

[0077]

[0078] In the formula, parameters A, B, and c are coefficients in the laser plane equation Ax + By + Cz = 1 obtained through calibration; fx is the focal length in the x-direction; fy is the focal length in the y-direction; (u0, v0) is the position of the optical axis origin; (u, v) is the position of a point on the 2D image; T is the 4*4 TCP transformation matrix, which can be obtained through upconversion using a robot teach pendant and is a known matrix; H is the 4*4 hand-eye matrix, which can be obtained through robot photo calibration and is a known matrix.

[0079] In practical applications, the centerline extraction can be achieved using the Steger method or the skeleton extraction method, and the coordinate transformation can be implemented based on different hand-eye calibration methods.

[0080] Step S3: Perform geometric feature optimization processing on the three-dimensional point cloud data, and fit and identify the geometric features of the T-shaped fillet weld structure based on the RANSAC algorithm with spatial geometric constraints.

[0081] Step S3 involves geometric feature optimization of the three-dimensional point cloud data, and fitting and identifying the geometric features of the T-shaped fillet weld structure based on the RANSAC algorithm with spatial geometric constraints. Specifically, this includes:

[0082] First, Euclidean clustering was performed on the weld seam point cloud in the robot coordinate system. Based on a distance threshold of 10mm and a minimum point count constraint of 250, spatially close points were divided into several independent clusters. The visual image of the point cloud after Euclidean clustering is shown below. Figure 8As shown, each clustering result is filtered, and discrete small clusters with fewer than a set threshold are removed, leaving only weld candidate clusters with sufficient number of points and continuity.

[0083] For clusters that meet the conditions, based on the previously established two-dimensional to three-dimensional point correspondence, the point indices within them are mapped back to the original two-dimensional image coordinates, and the pixel coordinate sequence of the weld center point is regenerated.

[0084] The obtained two-dimensional center pixel sequence was smoothed using a 5*5 window moving average method to reduce the impact of image noise and local jitter on the weld trajectory. Based on this, the smoothed two-dimensional center points were reconstructed into a point cloud, and further spatial smoothing and structural optimization were performed on the point cloud using Gaussian kernel covariance filtering. This resulted in weld centerline data with better continuity and lower noise. The generated point cloud is compared to... Figure 9 and Figure 10 As shown.

[0085] The obtained weld centerline data was subjected to outlier removal based on a 10mm radius constraint. This removed spatially isolated outliers and those with insufficient neighborhood points, making the weld point cloud more compact and stable. Furthermore, if a reference point from the previous frame exists, a KD-Tree search was used to further remove point clouds within a 5mm radius of the reference point to eliminate its impact on subsequent line fitting. Finally, the cleaned weld point cloud was smoothed using Gaussian kernel covariance filtering to further reduce local jitter and noise, resulting in the desired image. Figure 11 The three-dimensional centerline of the weld is shown to be more continuous and smooth in space.

[0086] The RANSAC-based line segmentation method iteratively fits lines to the point cloud. In each round, it extracts the optimal line from the remaining point cloud and removes the interior points of that line until there are insufficient remaining points to continue fitting. For each successfully fitted line with more than 100 interior points, the algorithm saves its model parameters and corresponding interior point cloud information, forming a line feature set. After extracting multiple lines, the algorithm pairs all lines together, calculates the spatial angle between their direction vectors, finds the pair of lines closest to being perpendicular, and calculates the difference from 90°. If the minimum deviation is less than a set angle threshold of 10°, it is considered that a T-shaped fillet weld satisfying geometric constraints exists in the current point cloud; otherwise, the processing of the current frame is terminated. Figures 12 to 13 The diagram shown is a comparison of the effects of angle constraints.

[0087] In practical applications, the point cloud optimization can employ a statistical outlier removal method, and the line fitting can be configured with different distance thresholds and maximum number of iterations.

[0088] Step S4: Based on the geometric features of the T-shaped fillet weld structure, calculate the three-dimensional coordinates of the key nodes of the weld by means of straight line endpoint distance constraints, three-dimensional intersection calculation and Z-axis height screening.

[0089] For each point cloud obtained in step S3, it is first projected onto the corresponding line model; Euclidean clustering is performed on the projected point cloud to extract continuous main line segments in space, and statistical outlier removal is performed on the line segments to further remove abnormal points; the endpoint positions of each line are calculated, and the start and end points of the line in space are obtained by the endpoint extraction method. The minimum Euclidean distance between the endpoints of two lines is calculated to evaluate the degree of proximity between the two line segments in space; when the minimum distance is greater than the preset threshold of 50mm, the two lines are considered to be separated from each other and do not constitute an effective weld structure, and the processing of the current frame is terminated; otherwise, the next step is continued.

[0090] The algorithm then iterates through pairs of lines, calculating the intersection point of each pair in 3D space. A 10mm distance threshold is used to determine if the intersection point is near a point within the line, thus eliminating numerically unstable or geometrically invalid intersections. For intersections meeting the criteria, the algorithm further uses the Z-axis coordinate as a filtering criterion, retaining only those with larger Z-values ​​to avoid misidentifying false intersections located in low-lying or noisy areas as valid weld nodes. If multiple candidate intersections exist, the intersection with the largest Z-value is selected as the final reference point and processed as a critical weld node. This 3D reference point is then mapped back to the image coordinate system through an inverse coordinate transformation, resulting in... Figure 14 The corresponding two-dimensional pixel position is shown, and the pixel and its three-dimensional coordinates are stored in the algorithm output structure as welding reference points.

[0091] In practical applications, the extraction of key nodes can be combined with curve fitting to optimize the intersection coordinates, and the distance threshold can be adjusted based on the actual weld size.

[0092] Advantages of implementing the embodiments of the present invention:

[0093] Achieve automatic and high-precision acquisition of weld seam positions. Through the close integration of laser vision sensors and robot motion, the weld seam is automatically converted from a two-dimensional image to a three-dimensional robot coordinate system, eliminating the need for manual teaching and significantly improving the level of welding automation.

[0094] This method effectively enhances the robustness of laser stripe extraction. A combination of a custom one-dimensional linear filtering operator and connected component screening significantly improves the laser stripe response of the weld seam, suppresses background and noise interference, and provides stable input for subsequent centerline extraction.

[0095] A stable and reliable 3D point cloud of weld seams is constructed. Through a multi-level point cloud processing workflow, including Euclidean clustering, moving average smoothing, Gaussian kernel covariance filtering, and outlier removal, the continuity and stability of the weld seam point cloud are significantly improved.

[0096] Accurately identify the geometric structure of T-type fillet welds. Through RANSAC multi-line fitting and spatial angle constraints, two approximately perpendicular weld lines can be reliably extracted, and the existence of a T-type fillet weld structure can be determined accordingly.

[0097] It can reliably acquire key node information of welds. By using straight-line endpoint distance constraints, 3D intersection calculations, and Z-axis height filtering, it can accurately determine weld intersection points in complex point cloud environments, providing reliable key reference points for subsequent welding path planning.

[0098] Example 2

[0099] like Figure 1 As shown, a method for identifying T-shaped fillet welds based on laser vision is described, the method comprising the following steps:

[0100] Step S1: Acquire an image, perform enhancement processing on the acquired image, and extract the effective laser stripe area;

[0101] Step S1 specifically includes: The vision sensor, consisting of a camera, a line laser, and a filter, is mounted at the end of a robotic arm. The camera's intrinsic parameters are calibrated using the Zhang Zhengyou calibration method, and the laser plane equation is calibrated using the planar target method. The robotic arm moves the vision sensor, projecting laser stripes onto the surface of the T-shaped fillet weld to be identified. The camera acquires images of the laser stripes at a preset frame rate and simultaneously records the pose of the robotic arm's end effector at the moment of capture.

[0102] The acquired RGB image was converted to a grayscale image using color space conversion. A 3×3 Gaussian kernel was then used to filter and smooth the grayscale image to suppress noise. Subsequently, the Sobel operator was applied to the smoothed image for edge detection to enhance the edge response of the laser stripes. The edge-enhanced image was then binarized using a local adaptive thresholding method to better accommodate brightness variations in different regions of the image. Finally, connected component analysis was performed on the binary image, and the pixel area of ​​each connected region was calculated. Valid laser stripe regions were selected based on a set minimum area threshold, while noisy connected regions with excessively small areas were removed.

[0103] Step S2: Extract the center line from the effective laser stripe area and convert the center line coordinates into three-dimensional point cloud data;

[0104] Step S2 specifically includes: extracting the center line from the binarized effective laser stripe region using a distance transform method. First, the Euclidean distance from each foreground pixel (grayscale value 255) to its nearest background pixel (grayscale value 0) in the image is calculated, generating a distance map. Then, the distance map is scanned line by line, searching for the pixel with the largest distance value in each line. If a foreground pixel exists in that line, its coordinates are recorded as the center point of that line. This method yields a series of pixel-level center point sequences distributed along the direction of the laser stripes.

[0105] The extracted 2D center pixels are transformed to 3D coordinates in the camera coordinate system based on the camera intrinsic parameter matrix and the calibrated laser plane equation. Then, by combining the hand-eye calibration matrix and the robot arm end-effector pose (i.e., the transformation matrix of the tool coordinate system relative to the base coordinate system) recorded synchronously during image acquisition, the 3D points in the camera coordinate system are transformed to the robot base coordinate system to form the 3D point cloud data of the weld.

[0106] Step S3: Perform geometric feature optimization processing on the three-dimensional point cloud data, and fit and identify the geometric features of the T-shaped fillet weld structure based on the RANSAC algorithm with spatial geometric constraints.

[0107] Step S3 specifically includes: firstly, performing Euclidean clustering analysis on the three-dimensional point cloud in the robot base coordinate system, setting a distance threshold of 10mm and a minimum point count threshold of 250, dividing spatially adjacent points into different candidate clusters, and removing discrete point clusters with fewer than 250 points to screen out effective weld point clouds.

[0108] The filtered point cloud clusters are projected back into a 2D image sequence, and the center point sequence is smoothed using a 5×5 window moving average method. The smoothed points are then reconstructed into a 3D point cloud. Subsequently, Gaussian kernel covariance filtering is used to spatially smooth and structurally optimize the point cloud. Based on a neighborhood search with a radius of 10mm, spatial outliers with too few points in the neighborhood are removed. If a weld reference point calculated in the previous frame exists, it is used as the center, and historical point clouds within a 5mm radius are removed to avoid duplicate accumulation. Finally, Gaussian kernel covariance filtering is applied again to the processed point cloud to obtain the optimized final weld point cloud.

[0109] The RANSAC algorithm with spatial geometric constraints is used to fit straight lines to the optimized point cloud. A straight line model is iteratively fitted from the point cloud, extracting the line with the most interior points at each step and removing its interior points from the remaining point cloud, until no valid straight line can be fitted. All straight line models with more than 100 interior points and their interior points are saved. The angle between the direction vectors of all fitted straight lines is calculated, and the set of straight lines with the closest angle to 90 degrees is found. If the actual angle of this set of straight lines deviates from 90 degrees by less than a set angle threshold of 10°, it is determined that there are two approximately perpendicular straight line structures in the current point cloud that conform to the geometric characteristics of a T-shaped fillet weld.

[0110] Step S4: Based on the geometric features of the T-shaped fillet weld structure, calculate the three-dimensional coordinates of the key nodes of the weld by means of straight line endpoint distance constraints, three-dimensional intersection calculation, and Z-axis height screening;

[0111] Step S4 specifically includes: For the two identified straight lines that satisfy the perpendicularity condition, the least squares method is used to accurately fit the three-dimensional spatial line equations based on their respective interior point clouds. The intersection point of the two spatial lines is calculated as a candidate key node. A distance threshold of 10mm is set to verify whether the intersection point is located near the interior point clouds of both lines simultaneously (i.e., the distance from the intersection point to both lines is within the threshold).

[0112] For candidate intersection points that satisfy the distance constraint, compare their Z-axis coordinates (usually perpendicular to the worktable plane in the robot's base coordinate system). Select the intersection point with the largest Z-coordinate value as the final critical node of the weld (i.e., the vertex of the T-shaped fillet weld). To increase robustness, further geometric verification can be performed: calculate the minimum Euclidean distance between the endpoints of the two lines. If this minimum distance is greater than 50mm, the two lines are determined to be too far apart in space and do not constitute a valid T-shaped fillet weld structure; the current identification result is invalid.

[0113] Finally, the calculated 3D coordinates of the key nodes are output as reference positions for robot welding.

[0114] The advantages of this embodiment are as follows: By extracting the centerline row by row based on distance transformation, combining the 3D coordinate transformation of the camera model and robot pose, optimizing the point cloud based on spatial clustering and filtering, and introducing RANSAC line fitting and geometric verification with angle threshold constraints, a complete and feasible T-shaped fillet weld recognition scheme is constructed. This method can stably extract laser stripes under complex lighting and noise environments, accurately identify mutually perpendicular weld features through strict geometric rules, and finally calculate the accurate 3D coordinates of key weld points, providing a reliable and efficient visual guidance solution for automated welding. Example

[0115] like Figure 2As shown, a T-shaped fillet weld recognition system based on laser vision includes:

[0116] The acquisition module M1 acquires images, enhances the acquired images, and extracts the effective laser stripe areas.

[0117] The extraction module M2 extracts the center line from the effective laser stripe area and converts the center line coordinates into three-dimensional point cloud data;

[0118] The feature module M3 performs geometric feature optimization processing on the three-dimensional point cloud data and identifies the geometric features of the T-shaped fillet weld structure based on the RANSAC algorithm with spatial geometric constraints.

[0119] The calculation module M4, based on the geometric features of the T-shaped fillet weld structure, calculates the three-dimensional coordinates of the key nodes of the weld through straight line endpoint distance constraints, three-dimensional intersection calculation, and Z-axis height screening.

[0120] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying T-shaped fillet welds based on laser vision, characterized in that, The laser vision-based T-shaped fillet weld identification method includes the following steps: Acquire images, enhance the acquired images, and extract the effective laser stripe areas; Extract the center line from the effective laser stripe area and convert the center line coordinates into three-dimensional point cloud data; The geometric features of the three-dimensional point cloud data are optimized, and the geometric features of the T-shaped fillet weld structure are identified by fitting based on the RANSAC algorithm with spatial geometric constraints. Based on the geometric features of the T-shaped fillet weld structure, the three-dimensional coordinates of the key nodes of the weld are calculated by straight line endpoint distance constraints, three-dimensional intersection calculation, and Z-axis height screening. The calculation of the three-dimensional coordinates of key weld nodes through straight-line endpoint distance constraints, three-dimensional intersection calculation, and Z-axis height screening includes: Calculate the intersection point of two straight lines in three-dimensional space; The validity of intersection points is determined by distance threshold constraints; Select the intersection point with the largest Z-coordinate value as the weld reference point; Furthermore, a geometric verification step is included before calculating the three-dimensional coordinates of the critical nodes of the weld: Calculate the minimum Euclidean distance between the endpoints of two straight lines; When the minimum distance is less than or equal to a preset threshold, it is determined to be a valid weld structure, and the step of calculating the three-dimensional coordinates of the key nodes of the weld is performed; when the minimum distance is greater than the preset threshold, it is determined to be an invalid weld structure and the process is terminated.

2. The method for identifying T-shaped fillet welds based on laser vision according to claim 1, characterized in that, The process of enhancing the acquired image and extracting the effective laser stripe region includes: Convert the color image to a grayscale image and then perform Gaussian filtering for smoothing. A one-dimensional linear filtering operator is used to perform convolution operations on the image; The filtering results are binarized, and effective laser stripe regions are selected through connected component analysis.

3. The method for identifying T-shaped fillet welds based on laser vision according to claim 1, characterized in that, The extraction of the center line from the effective laser stripe region is performed using a distance transformation method, specifically including: Calculate the distance from each foreground pixel to the nearest background pixel; Scan the distance map row by row and take the pixel with the largest distance value in each row as the center point; A sequence of center points is formed that is continuously distributed along the weld seam.

4. The method for identifying T-shaped fillet welds based on laser vision according to claim 1, characterized in that, The process of converting centerline coordinates into three-dimensional point cloud data includes: The pixel coordinates are transformed to the camera coordinate system using the camera intrinsic parameter matrix and the laser plane equation. Transform the coordinates to the robot coordinate system using the hand-eye matrix and the robotic arm pose matrix.

5. The method for identifying T-shaped fillet welds based on laser vision according to claim 1, characterized in that, The geometric feature optimization processing of the three-dimensional point cloud data includes: Perform Euclidean cluster analysis on the 3D point cloud to screen effective weld point cloud clusters; The point cloud was smoothed using moving average and Gaussian kernel covariance filtering. Spatial outliers are eliminated based on radius constraints.

6. The method for identifying T-shaped fillet welds based on laser vision according to claim 1, characterized in that, The geometric features of the T-type fillet weld structure fitted and identified by the RANSAC algorithm with spatial geometric constraints include: Multiple straight lines are iteratively fitted using the RANSAC algorithm; Calculate the spatial angle between the direction vectors of each line; When there is a set of straight lines whose included angles deviate from 90° less than a set threshold, it is determined to be a T-type fillet weld.

7. The laser vision-based T-shaped fillet weld identification method according to claim 6, characterized in that, The set threshold is 10°, and the number of interior points of the fitted straight line is greater than 100.

8. A T-shaped fillet weld recognition system based on laser vision, characterized in that, The laser vision-based T-type fillet weld recognition system is implemented based on the laser vision-based T-type fillet weld recognition method according to any one of claims 1 to 7, including: The acquisition module acquires images, enhances the acquired images, and extracts the effective laser stripe areas. The extraction module extracts the center line from the effective laser stripe area and converts the center line coordinates into three-dimensional point cloud data; The feature module performs geometric feature optimization processing on the three-dimensional point cloud data and identifies the geometric features of the T-shaped fillet weld structure based on the RANSAC algorithm with spatial geometric constraints. The calculation module, based on the geometric features of the T-shaped fillet weld structure, calculates the three-dimensional coordinates of the key nodes of the weld through straight line endpoint distance constraints, three-dimensional intersection calculation, and Z-axis height screening.

Images