A steel bar welding point intelligent identification and mechanical arm self-adaptive positioning method and device

Abstract

The application provides a steel bar welding point intelligent identification and mechanical arm self-adaptive positioning method and device, relates to the technical field of intelligent control of mechanical arms, and comprises the following steps: collecting an image of a steel bar net, pre-processing the image, and generating a pre-processed image; segmenting a steel bar area from the pre-processed image and extracting a center line of the steel bar area; determining an initial position of a welding point based on the intersection of the center line, generating a target area containing the initial position; inputting the target area into a trained neural network model, and predicting a detection position of the welding point by the neural network model; determining the geometric structure features at the welding point by using the image information in the neighborhood of the detection position; correcting the detection position according to the geometric structure features to obtain the coordinates of the welding point. The application can effectively suppress the interference of site reflections, shadows and sundries, and realize high-precision intelligent identification of welding points without manual presetting of point positions.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology for robotic arms, and more specifically, to a method and apparatus for intelligent identification of steel bar welding points and adaptive positioning of robotic arms. Background Technology

[0002] In automated welding operations of steel reinforcement cages in construction, existing intelligent steel reinforcement welding robotic arms typically use monocular vision for simple spacing measurements. However, because steel reinforcement is a cylindrical metal component with strong surface reflection and blurred edges, the measurement error of the lateral distance between the robotic arm's end and the point to be welded is relatively large. Furthermore, traditional methods struggle to identify tilting, misalignment, or bending deformation of the steel reinforcement, resulting in significant positioning deviations at the welding points and a tendency for misalignment, incomplete welds, or missed welds. In addition, existing methods lack an intelligent welding point recognition mechanism, relying on manually preset points. There is a lack of real-time visual closed-loop correction between visual measurement data and robotic arm movement, and no capabilities for positioning verification, post-weld inspection, or anomaly classification processing, leading to poor robustness in complex construction site environments. Summary of the Invention

[0003] The purpose of this invention is to provide a method and apparatus for intelligent identification of rebar welding points and adaptive positioning of a robotic arm, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows:

[0004] Firstly, this application provides a method for intelligent identification of steel bar welding points, including:

[0005] Images of the reinforcing mesh are acquired, and the images are preprocessed to generate preprocessed images;

[0006] Based on the directional distribution and geometric size characteristics of the reinforcing bars in the preprocessed image, the reinforcing bar region is segmented from the preprocessed image, and the centerline of the reinforcing bar region is extracted;

[0007] The initial position of the welding point is determined based on the intersection of the center lines, and a target area containing the initial position is generated.

[0008] The target region is input into the trained neural network model, which then predicts the detection location of the welding point.

[0009] The geometric features of the welding point are determined by using image information in the neighborhood of the detection location, and the detection location is corrected according to the geometric features to obtain the coordinates of the welding point.

[0010] Secondly, this application also provides an adaptive positioning method for a rebar welding robotic arm, comprising:

[0011] The actual lateral distance between the end of the robotic arm and the welding point is calculated based on the monocular vision ranging principle.

[0012] Identify the reinforcing bars associated with the welding point, and determine the lateral movement correction amount and rotation correction amount based on the inclination angle of the associated reinforcing bars;

[0013] The robotic arm is driven to perform alignment actions according to the actual lateral distance, lateral movement correction amount, and rotation correction amount through visual feedback control.

[0014] Thirdly, this application also provides an intelligent identification device for steel bar welding points, comprising:

[0015] The image acquisition module is used to acquire images of the steel mesh and preprocess the images to generate a preprocessed image.

[0016] The segmentation and extraction module is used to segment the rebar region from the preprocessed image based on the directional distribution characteristics and geometric size characteristics of the rebar in the preprocessed image, and extract the center line of the rebar region;

[0017] An initial positioning module is used to determine the initial position of the welding point based on the intersection of the center lines and generate a target area containing the initial position.

[0018] The neural network detection module is used to input the target area into the trained neural network model, and the neural network model predicts the detection position of the welding point;

[0019] The correction module is used to determine the geometric structural features at the welding point using image information in the neighborhood of the detection location, and to correct the detection location based on the geometric structural features to obtain the coordinates of the welding point.

[0020] Fourthly, this application also provides an adaptive positioning device for a rebar welding robotic arm, comprising:

[0021] The lateral distance calculation module is used to calculate the actual lateral distance between the end of the robotic arm and the welding point based on the monocular vision ranging principle.

[0022] The correction amount determination module is used to determine the reinforcing bars associated with the welding point, and to determine the lateral movement correction amount and rotation correction amount based on the inclination angle of the associated reinforcing bars.

[0023] The alignment control module is used to drive the robotic arm to perform alignment actions according to the actual lateral distance, lateral movement correction amount, and rotation correction amount through visual feedback control.

[0024] The beneficial effects of this invention are as follows:

[0025] 1. This invention can effectively suppress interference from construction site reflections, shadows, and debris. In particular, it can achieve high-precision intelligent identification of welding points for inclined, misaligned, or bent steel bars without the need for manual pre-setting of points.

[0026] 2. This invention calculates the actual lateral distance based on the monocular vision ranging principle, determines the lateral movement correction amount and rotation correction amount according to the inclination angle of the reinforcing bar, and drives the robotic arm to perform alignment actions through visual feedback control, thereby improving the stability and automation of welding operations.

[0027] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a schematic diagram of the intelligent identification method for steel bar welding points described in this embodiment of the invention;

[0030] Figure 2 This is a schematic diagram of the adaptive positioning method for the rebar welding robotic arm described in this embodiment of the invention;

[0031] Figure 3 This is a schematic diagram of the intelligent identification device for steel bar welding points described in an embodiment of the present invention;

[0032] Figure 4 This is a schematic diagram of the adaptive positioning device for the rebar welding robotic arm described in an embodiment of the present invention; Detailed Implementation

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0034] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0035] Example 1:

[0036] See Figure 1 This embodiment provides a method for intelligent identification of steel bar welding points, including:

[0037] Before performing the recognition method, this embodiment first performs offline calibration on the monocular camera.

[0038] Specifically, the Zhang Zhengyou checkerboard calibration method is adopted, using a 9×6 checkerboard template, and the equivalent focal length of the camera is calculated through a calibration algorithm. and distortion coefficient After calibration, the camera is fixedly mounted on the robotic arm, ensuring that the camera's optical axis is strictly perpendicular to the plane of the steel mesh to be welded, and the working distance between the camera and the steel mesh is locked. .

[0039] S1. Start the camera to acquire the original image of the steel mesh, preferably, acquire a single frame image, preprocess the image to generate a preprocessed image;

[0040] In this embodiment, the image is preprocessed to generate a preprocessed image. The preprocessing includes sequentially performing distortion correction, grayscale conversion, reflection suppression, contrast enhancement, and noise filtering on the image.

[0041] Specifically, for any pixel in the image To eliminate geometric distortion caused by lens distortion, distortion correction is performed according to the following formula:

[0042] ;

[0043] in, , These are the corrected pixel coordinates.

[0044] Furthermore, the corrected color image is converted to a single-channel grayscale image to reduce data volume and eliminate color interference. For each pixel in the color image, its red component... Green components Blue component Calculate the grayscale value using the following formula. :

[0045] ;

[0046] Specifically, this embodiment uses a combination of homomorphic filtering and reflection mask weighted repair to suppress reflection:

[0047] First, a logarithmic transformation is performed on the grayscale image to convert the multiplicative illumination components into additive components. Let the grayscale value of a pixel in the original grayscale image be... The formula for logarithmic transformation is:

[0048] ;

[0049] In the formula, The transformed image, These are the converted pixels.

[0050] Then, a two-dimensional Fourier transform is performed on the transformed image to obtain the frequency domain image. ;

[0051] A Gaussian high-pass filter is used to filter the frequency domain image. The filter has the following form:

[0052] ;

[0053] In the formula, The parameter is used to control the cutoff frequency of the filter, which is used to transform the frequency domain image. With filters By multiplying point by point, the filtered frequency domain image is obtained. .

[0054] Furthermore, inverse Fourier transforms are performed sequentially in the frequency domain. and exponential restoration The reflection-suppressed image is obtained, where, Represents the spatial domain, This indicates an image with reduced reflection.

[0055] Preferably, a reflection mask is generated using a fixed threshold to further repair the affected area. This applies to the homomorphically filtered image. A threshold of 230 is set. When a pixel's grayscale value is greater than or equal to 230, the pixel is marked as a reflective area with a mask value of 1; otherwise, it is marked as 0. ;

[0056] Perform a 5×5 median filter on the pixels within the mask area, and then fuse the original grayscale image and the median-filtered image at the pixel level with a weight ratio of 7:3 to complete the highlight restoration.

[0057] Specifically, this embodiment uses fixed parameters for adaptive enhancement of CLAHE:

[0058] The image after reflection suppression is divided into several 16×16 pixel blocks, and a contrast limit threshold of 2.0 is set. A histogram is calculated within each block, and pixels exceeding the threshold are uniformly cropped and redistributed to other gray levels. Then, bilinear interpolation is performed on each block to eliminate block artifacts.

[0059] Specifically, this embodiment uses adaptive Gaussian filtering to remove random noise from the image.

[0060] Based on the above embodiments, this method further includes:

[0061] S2. Based on the directional distribution characteristics and geometric size characteristics of the reinforcing bars in the preprocessed image, segment the reinforcing bar region from the preprocessed image and extract the centerline of the reinforcing bar region;

[0062] Specifically, step S2 includes:

[0063] S21. Extract the main direction of the reinforcing bars in the preprocessed image and calculate the average pixel diameter of the reinforcing bars;

[0064] Specifically, the Sobel operator is used to calculate the horizontal and vertical gradients of the preprocessed image, thereby obtaining the gradient direction of each pixel.

[0065] A gradient orientation histogram is constructed, and the principal orientations of the reinforcing bars are automatically extracted using the K-means clustering algorithm. The width distribution of the reinforcing bars is statistically analyzed along directions perpendicular to each principal orientation, and the width value with the highest frequency is taken as the average pixel diameter of the reinforcing bar.

[0066] S22. Based on the main direction and average pixel diameter, the preprocessed image is binarized and segmented to obtain a binary mask for the reinforcing bars;

[0067] In this embodiment, the local window size is set according to the average pixel diameter of the reinforcing bars:

[0068] ;

[0069] In the formula, Let the side length be denoted as 'window'. The average pixel diameter of the reinforcing bar;

[0070] For each pixel, within a square window centered on it and with side length W, calculate the average gray value of all pixels within the window. As a local threshold, a pixel is marked as a foreground reinforcement only when it simultaneously satisfies both the global and local threshold conditions; otherwise, it is marked as background. This yields a preliminary binary mask.

[0071] S23. Apply geometric constraints to the binary mask of the reinforcing bars and filter them, and take the connected regions that satisfy the geometric constraints as the final segmented reinforcing bar regions;

[0072] Specifically, the geometric constraints are an aspect ratio greater than 4:1, width fluctuation less than 20%, and linear fitting residual less than 3 pixels.

[0073] Connected components that simultaneously satisfy all three conditions are retained as the final binary mask for the reinforcing bars.

[0074] Subsequently, a rectangular structural element is determined based on the average pixel diameter of the reinforcing bar. The rectangular structural element is then used to perform expansion and erosion operations on the mask in sequence to fill the internal holes and smooth the edges.

[0075] Finally, the Zhang-Suen parallel thinning algorithm is used to iteratively delete edge pixels from the closing operation results: pixels that meet specific boundary conditions are repeatedly deleted until no pixels can be deleted. At this point, the remaining foreground pixels form a single-pixel-width steel reinforcement centerline skeleton.

[0076] Based on the above embodiments, this method further includes:

[0077] S3. Determine the initial position of the welding point based on the intersection of the center lines, and generate a target area containing the initial position;

[0078] Specifically, step S3 includes:

[0079] S31. Perform distance transformation on the rebar region. For each foreground pixel within the mask, calculate its Euclidean distance to the nearest background pixel, thereby determining the distance value of each pixel within the rebar region:

[0080] S32. At the intersection of the centerlines, candidate points are selected based on the distance value, local gradient value, and centerline intersection angle matching degree corresponding to the intersection point:

[0081] Specifically, a 3×3 neighborhood window is used to traverse the skeleton image. If the foreground value of the center pixel in the window is 255 and the number of foreground pixels in its 8 neighborhoods is greater than 2, then the center pixel is marked as an intersection point.

[0082] A weighted scoring function is constructed based on the distance value, local gradient value, and centerline intersection angle matching degree corresponding to the intersection point, and the score of each intersection point is calculated. :

[0083] ;

[0084] In the formula, , , These are the weight values, which in this embodiment are 0.4, 0.3, and 0.3 respectively. This is the distance value. For local gradient values, This represents the maximum distance value in the distance transformation graph. This represents the maximum gradient magnitude in the preprocessed image.

[0085] S33. Perform non-maximum suppression on the candidate points, and determine the retained candidate points as the initial positions of the welding points;

[0086] Specifically, the selected candidate points are sorted from highest to lowest score, and non-maximum suppression is performed: if the distance between the current candidate point and the retained candidate points is less than the preset suppression radius, the current candidate point is discarded; otherwise, it is retained. The final retained candidate points are the initial positions of the welding points.

[0087] S34. Using the initial position as the center, cut out the target area according to the geometric dimensions of the reinforcing bar.

[0088] Based on the above embodiments, this method further includes:

[0089] S4. Input the target area into the trained neural network model, and the neural network model predicts the detection position of the welding point;

[0090] In this embodiment, the neural network model is a specialized model derived from YOLOv8n, specifically improved for rebar intersection features. Specifically, the neural network model includes a line enhancement feature extraction layer, a dynamic pooling layer, a direction-aware feature fusion layer, and a geometric constraint detection head.

[0091] Specifically, step S4 includes:

[0092] S41. Input the target region into the neural network model. The line enhancement feature extraction layer performs parallel directional gradient convolution on the input image and outputs a multi-scale line feature map.

[0093] In this embodiment, the ROI is scaled proportionally to 640×640 pixels. After scaling, the image edges are filled using the local grayscale mean of the steel bars, and the filled image data is normalized to the [0,1] interval as the input to the neural network.

[0094] The neural network employs the C2f architecture of YOLOv8n. In the feature extraction stage, the neural network comprises two parallel processing branches. The first branch is a standard C2f module, used to extract general multi-scale features. The second branch is a rebar line enhancement module, used to perform directional gradient convolution on the input image to extract linear responses along the main direction of the rebar. The output feature maps of the two branches are added element-wise to obtain the fused enhanced features.

[0095] The neural network generates multi-scale feature maps of sizes 80×80 pixels, 40×40 pixels, and 20×20 pixels, respectively, through downsampling layers with strides of 8, 16, and 32, denoted as P3, P4, and P5. All feature maps enhance the linear structure response while suppressing noise interference such as speckles, reflections, and oil stains.

[0096] S42. The dynamic pooling layer selects the size of the pooling kernel according to the average pixel diameter of the reinforcing bar, performs pooling on the multi-scale linear feature map, and outputs multi-receptive field fusion features.

[0097] In this embodiment, the dynamic pooling layer adds the function of dynamically selecting the pooling kernel size based on the YOLOv8n structure.

[0098] Specifically, the size of the pooling kernel is selected based on the average pixel diameter of the reinforcing bars:

[0099] When D is small, a smaller pooling kernel should be selected;

[0100] When D is large, a larger pooling kernel should be selected;

[0101] For each scale of feature map, max pooling is performed using a pooling kernel of the selected size. The results obtained from different pooling kernel sizes are then concatenated along the channel dimension to output a multi-receptive field fused feature. This fused feature simultaneously includes local details and global contextual information, and the pooling kernel size matches the width of the reinforcing bars, avoiding feature loss or receptive field redundancy.

[0102] S43. The direction-aware feature fusion layer applies adaptive weights to the multi-receptive-field fusion feature map according to the main direction of the reinforcing bar, and outputs the enhanced feature map;

[0103] Specifically, the direction-aware feature fusion layer adopts a bidirectional feature pyramid structure to achieve top-down and bottom-up feature transfer and fusion.

[0104] During the fusion process, the orientation-aware feature fusion layer applies orientation-adaptive weights to different spatial locations on the feature map using the main direction of the reinforcing bars. Specifically, for each pixel location on the feature map, its gradient direction at the corresponding location in the preprocessed image is calculated:

[0105] If the gradient direction is consistent with the main direction of the reinforcing bar, a higher weight is assigned to enhance the feature response at the corresponding location;

[0106] If the direction deviates significantly, a lower weight is assigned to suppress noise and background interference.

[0107] Through a top-down path, high-level semantic information is passed to the low-level feature map; then, through a bottom-up path, low-level localization information is passed to the high-level feature map. The aforementioned directional adaptive weights are applied during each layer fusion. In the final enhanced feature map, the response of the rebar intersection region is significantly enhanced, while background clutter is effectively suppressed.

[0108] S44. The geometric constraint detection head outputs the center coordinates of the prediction box as the detection position of the welding point based on the enhanced feature map;

[0109] Specifically, for each spatial location on the enhanced feature map, determine whether it falls on the centerline skeleton after being mapped back to the original image:

[0110] If so, predict the horizontal and vertical offsets of the spatial location relative to the feature point location, as well as the width and height of the prediction box.

[0111] The feature point coordinates are added to the predicted horizontal and vertical offsets to obtain the pixel coordinates of the weld point center in the image. For all predicted weld point centers, results with a confidence level greater than 0.5 are retained first, and then non-maximum suppression is performed to remove redundant detection boxes with overlapping positions. Then, the retained weld point center coordinates are subjected to a spatial logical AND operation with the binary mask of the rebar obtained in step S2 to eliminate false targets falling outside the mask, retaining only the weld points located inside the rebar area. The final output coordinates are the detection positions of the weld points.

[0112] Based on the above embodiments, this method further includes:

[0113] S5. Determine the geometric structural features at the welding point using image information in the neighborhood of the detection location, and correct the detection location based on the geometric structural features to obtain the coordinates of the welding point.

[0114] Specifically, step S5 includes:

[0115] S51. Extract the edge points of the reinforcing bars within the neighborhood of the detection location;

[0116] In this embodiment, a square search window is set centered on each weld point detection location. Within this square search window, edge extraction is performed on the binary mask of the reinforcing bars. Specifically, the Canny edge detection operator is used to extract all edge points within the window. Then, the edge points are classified according to the main direction of the reinforcing bars: edge points with an angle less than 30° to the main direction are grouped together, corresponding to reinforcing bars aligned with the main direction; edge points with an angle between 60° and 120° to the main direction are grouped together, corresponding to reinforcing bars perpendicular to the main direction. This results in two sets of edge points for the two reinforcing bars, respectively. and .

[0117] S52. The edge points are fitted with straight lines using the robust estimation method to obtain the central axis of each steel bar constituting the welding point;

[0118] Specifically, for each set of edge points, a line fit is performed using the least squares method. To suppress noise and outlier interference, IGGIII robust estimation is incorporated during the fitting process. Preferably, the robustness parameter is set to... After iterative calculations, two straight lines were finally obtained. These two lines represent the central axes of the two reinforcing bars in the image, respectively. and .

[0119] S53. Project the detection positions vertically onto each central axis, and take the midpoint of each projection point as the corrected welding point;

[0120] In this embodiment, the detection position of the welding point is projected vertically onto two central axes, resulting in a projection point on each axis. The midpoint of the line connecting these two projection points is then taken as the corrected welding point position. Through this geometric constraint, the detection accuracy can be improved to the sub-pixel level.

[0121] S54. In order to ensure the stability of the welding point coordinates, 5 to 8 frames of images are continuously acquired, and the welding point coordinates obtained from each frame of image are filtered by a sliding window to remove abnormal coordinate values.

[0122] S55. Perform a validity check on the filtered coordinates, and use the coordinates that pass the validity check as the coordinates of the welding point;

[0123] In this embodiment, the weld point coordinates undergo triple validity checks. Only if all checks pass is the weld point considered valid. The triple validity checks are as follows:

[0124] 1. Region Validity: Verify whether the coordinates of the welding point are located inside the binary mask of the reinforcing bar;

[0125] 2. Axis Distance Validity: Calculate the coordinates of the welding point to the two center axes. and distance and .when pixels and At the pixel level, the axial distance validity is satisfied.

[0126] 3. Coordinate stability: For the coordinates of welding points in multiple consecutive frames of output, calculate the variance of the coordinates in the most recent 5 consecutive frames. If the variance is less than a preset threshold, the coordinates are considered stable.

[0127] If the welding point coordinates of the current frame simultaneously meet all three conditions mentioned above, then these coordinates will be used as the final welding point coordinates. .

[0128] Example 2:

[0129] like Figure 2 As shown, this embodiment also provides an adaptive positioning method for a rebar welding robotic arm, enabling the robotic arm to automatically align and complete welding, specifically including:

[0130] S1. Calculate the actual lateral distance between the end of the robotic arm and the welding point based on the monocular vision ranging principle, wherein the end of the robotic arm is a two-finger gripping mechanism;

[0131] Specifically, the reference pixel coordinates of the center point of the two-finger gripping mechanism in the image are calibrated as follows: Calculate the pixel lateral difference between the end of the robotic arm and the welding point:

[0132] ;

[0133] Based on the pinhole imaging model of a monocular camera, with a known camera focal length... Fixed working distance between the camera and the steel reinforcement plane Under these conditions, the conversion relationship between pixel distance and actual physical distance is as follows:

[0134] ;

[0135] In the formula, This represents the actual horizontal distance.

[0136] Based on the above embodiments, this method further includes:

[0137] S2. Determine the reinforcing bars associated with the welding point, and determine the lateral movement correction amount and rotation correction amount based on the inclination angle of the associated reinforcing bars;

[0138] Specifically, step S2 includes:

[0139] S21. Extract the centerline of the reinforcing bar associated with the welding point, calculate the inclination angle of the centerline, and use the inclination angle as the rotation correction amount. ;

[0140] S22. Determine the vertical distance between the end of the robotic arm and the welding point. The lateral movement correction amount is calculated based on the tilt angle and vertical distance:

[0141] ;

[0142] In the formula, This is the amount of lateral movement correction;

[0143] Therefore, the total lateral displacement required by the robotic arm At the same time, the end effector of the robotic arm needs to rotate. The angle is adjusted so that the axis of the welding torch is perpendicular to the vertical rib.

[0144] Based on the above embodiments, this method further includes:

[0145] S3. Through visual feedback control, drive the robotic arm to perform alignment actions according to the actual lateral distance, lateral movement correction amount, and rotation correction amount;

[0146] First, the robotic arm determines the total displacement. and rotational correction amount The initial alignment process is performed, during which real-time images are continuously acquired, and the real-time lateral error between the current robotic arm end effector and the welding point is calculated. and angle error An incremental PID controller is used to calculate the control quantity. When the absolute value of the lateral error is less than 0.5 mm and the absolute value of the angular error is less than 0.3° in three consecutive frames, the alignment is considered complete. At this time, the welding torch at the end of the robotic arm is aligned with the welding point to perform welding.

[0147] Example 3:

[0148] like Figure 3 As shown, this embodiment provides an intelligent identification device for steel bar welding points, including:

[0149] The image acquisition module is used to acquire images of the steel mesh and preprocess the images to generate a preprocessed image.

[0150] The segmentation and extraction module is used to segment the rebar region from the preprocessed image based on the directional distribution characteristics and geometric size characteristics of the rebar in the preprocessed image, and extract the center line of the rebar region;

[0151] An initial positioning module is used to determine the initial position of the welding point based on the intersection of the center lines and generate a target area containing the initial position.

[0152] The neural network detection module is used to input the target area into the trained neural network model, and the neural network model predicts the detection position of the welding point;

[0153] The correction module is used to determine the geometric structural features at the welding point using image information in the neighborhood of the detection location, and to correct the detection location based on the geometric structural features to obtain the coordinates of the welding point.

[0154] Based on the above embodiments, the segmentation and extraction module includes:

[0155] The directional diameter calculation unit is used to extract the main direction of the reinforcing bars in the preprocessed image and calculate the average pixel diameter of the reinforcing bars;

[0156] The binarization segmentation unit is used to perform binarization segmentation on the preprocessed image based on the main direction and the average pixel diameter to obtain a binary mask for the reinforcing bars.

[0157] The geometric constraint filtering unit is used to apply geometric constraints to the binary mask of the reinforcing bars and filter the connected regions that satisfy the geometric constraints as the final segmented reinforcing bar regions.

[0158] Based on the above embodiments, the initial positioning module includes:

[0159] A distance transformation unit is used to perform distance transformation on the steel reinforcement area to determine the distance value of each pixel point in the steel reinforcement area;

[0160] The candidate point filtering unit is used to filter candidate points at the intersection of the center lines based on the distance value, local gradient value, and center line intersection angle matching degree corresponding to the intersection point.

[0161] The non-maximum suppression unit is used to perform non-maximum suppression on the candidate points and determine the retained candidate points as the initial positions of the welding points.

[0162] The target area segmentation unit is used to segment the target area based on the geometric dimensions of the reinforcing bars, with the initial position as the center.

[0163] Based on the above embodiments, the neural network detection module is specifically used for:

[0164] The target region is input into the neural network model. The line enhancement feature extraction layer outputs a multi-scale line feature map, the dynamic pooling layer outputs a multi-receptive field fusion feature, the direction-aware feature fusion layer outputs an enhanced feature map, and the geometric constraint detection head outputs the center coordinates of the prediction box as the detection position of the welding point.

[0165] Based on the above embodiments, the correction module includes:

[0166] An edge point extraction unit is used to extract the edge points of the reinforcing bars within the neighborhood of the detection location.

[0167] An axis fitting unit is used to perform linear fitting on the edge points using a robust estimation method to obtain the central axis of each steel bar constituting the welding point.

[0168] The projection correction unit is used to vertically project the detection position onto each of the central axes, and take the midpoint of each projection point as the corrected welding point.

[0169] The multi-frame filtering unit is used to continuously acquire multiple frames of images and perform sliding window filtering on the welding point coordinates obtained from each frame of image to remove abnormal coordinate values.

[0170] The validity judgment unit is used to perform validity judgment on the filtered coordinates and use the coordinates that pass the validity judgment as the coordinates of the welding point.

[0171] It should be noted that the specific manner in which each module performs its operation in the apparatus described in the above embodiments has been described in detail in the embodiments of the method, and will not be elaborated here.

[0172] Example 4:

[0173] like Figure 4 As shown, this embodiment provides an adaptive positioning device for a rebar welding robotic arm, comprising:

[0174] The lateral distance calculation module is used to calculate the actual lateral distance between the end of the robotic arm and the welding point based on the monocular vision ranging principle.

[0175] The correction amount determination module is used to determine the reinforcing bars associated with the welding point, and to determine the lateral movement correction amount and rotation correction amount based on the inclination angle of the associated reinforcing bars.

[0176] The alignment control module is used to drive the robotic arm to perform alignment actions according to the actual lateral distance, lateral movement correction amount, and rotation correction amount through visual feedback control.

[0177] Based on the above embodiments, the correction amount determination module includes:

[0178] The centerline extraction unit is used to extract the centerline of the reinforcing bar associated with the welding point, calculate the tilt angle of the centerline, and use the tilt angle as the rotation correction amount.

[0179] The lateral correction calculation unit is used to determine the vertical distance between the end of the robotic arm and the welding point, and to calculate the lateral movement correction amount based on the tilt angle and the vertical distance.

[0180] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0181] 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 technical scope 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 intelligent identification of steel bar welding points, characterized in that, include: Images of the reinforcing mesh are acquired, and the images are preprocessed to generate preprocessed images; Based on the directional distribution and geometric size characteristics of the reinforcing bars in the preprocessed image, the reinforcing bar region is segmented from the preprocessed image, and the centerline of the reinforcing bar region is extracted; The initial position of the welding point is determined based on the intersection of the center lines, and a target area containing the initial position is generated. The target region is input into the trained neural network model, which then predicts the detection location of the welding point. The geometric features of the welding point are determined by using image information in the neighborhood of the detection location, and the detection location is corrected according to the geometric features to obtain the coordinates of the welding point; in, The target region is input into a trained neural network model, which then predicts the detection location of the welding point, including: The target region is input into the neural network model, and the line enhancement feature extraction layer performs parallel directional gradient convolution on the input image to output a multi-scale line feature map. The dynamic pooling layer selects the size of the pooling kernel based on the average pixel diameter of the reinforcing bar, performs pooling on the multi-scale linear feature map, and outputs multi-receptive field fused features. The direction-aware feature fusion layer applies adaptive weights to the multi-receptive field fusion feature map according to the main direction of the reinforcing bar, and outputs the enhanced feature map. The geometrically constrained detection head outputs the center coordinates of the prediction box as the detection position of the welding point based on the enhanced feature map.

2. The intelligent identification method for steel bar welding points according to claim 1, characterized in that, The image preprocessing includes: performing distortion correction, grayscale conversion, reflection suppression, contrast enhancement, and noise filtering on the image in sequence.

3. The intelligent identification method for steel bar welding points according to claim 1, characterized in that, Based on the directional distribution and geometric dimensions of the reinforcing bars in the preprocessed image, the reinforcing bar region is segmented from the preprocessed image, including: Extract the main direction of the reinforcing bars in the preprocessed image and calculate the average pixel diameter of the reinforcing bars; Based on the main direction and average pixel diameter, the preprocessed image is binarized and segmented to obtain a binary mask for the reinforcing bars; Geometric constraints are applied to the binary mask of the reinforcing bars for screening, and the connected regions that satisfy the geometric constraints are taken as the final segmented reinforcing bar regions.

4. The intelligent identification method for steel bar welding points according to claim 1, characterized in that, The initial position of the welding point is determined based on the intersection of the center lines, and a target region containing the initial position is generated, including: Perform distance transformation on the steel reinforcement area to determine the distance value of each pixel within the steel reinforcement area; At the intersection of the center lines, candidate points are selected based on the distance value, local gradient value, and center line intersection angle matching degree corresponding to the intersection point; Non-maximum suppression is applied to the candidate points, and the retained candidate points are determined as the initial positions of the welding points; Centered on the initial position, the target area is cut out according to the geometric dimensions of the reinforcing bars.

5. The intelligent identification method for steel bar welding points according to claim 1, characterized in that, The geometric features at the welding point are determined using image information within the neighborhood of the detection location. The detection location is then corrected based on these geometric features to obtain the coordinates of the welding point. This process includes: Extract the edge points of the reinforcing bars within the neighborhood of the detection location; The edge points are fitted with straight lines using the robust estimation method to obtain the central axis of each steel bar constituting the welding point; The detection positions are vertically projected onto each of the central axes, and the midpoint of each projection point is taken as the corrected welding point. Multiple frames of images are continuously acquired, and the coordinates of the welding points obtained from each frame are filtered by a sliding window to remove abnormal coordinate values. The filtered coordinates are validated, and the coordinates that pass the validation are used as the coordinates of the welding point.

6. An adaptive positioning method for a robotic arm for welding reinforcing bars, wherein the coordinates of the welding point are obtained by means of the method described in any one of claims 1-5, characterized in that, include: The actual lateral distance between the end of the robotic arm and the welding point is calculated based on the monocular vision ranging principle. Identify the reinforcing bars associated with the welding point, and determine the lateral movement correction amount and rotation correction amount based on the inclination angle of the associated reinforcing bars; The robotic arm is driven to perform alignment actions according to the actual lateral distance, lateral movement correction amount, and rotation correction amount through visual feedback control.

7. The adaptive positioning method for a rebar welding robotic arm according to claim 6, characterized in that, Identify the reinforcing bars associated with the weld point, and determine the lateral movement correction amount and rotation correction amount based on the inclination angle of the associated reinforcing bars, including: Extract the centerline of the reinforcing bar associated with the welding point, calculate the tilt angle of the centerline, and use the tilt angle as the rotation correction amount; Determine the vertical distance between the end of the robotic arm and the welding point, and calculate the lateral movement correction amount based on the tilt angle and the vertical distance.

8. A smart identification device for steel bar welding points, used to implement the smart identification method for steel bar welding points according to any one of claims 1-5, characterized in that, include: The image acquisition module is used to acquire images of the steel mesh and preprocess the images to generate a preprocessed image. The segmentation and extraction module is used to segment the rebar region from the preprocessed image based on the directional distribution characteristics and geometric size characteristics of the rebar in the preprocessed image, and extract the center line of the rebar region; An initial positioning module is used to determine the initial position of the welding point based on the intersection of the center lines and generate a target area containing the initial position. The neural network detection module is used to input the target area into the trained neural network model, and the neural network model predicts the detection position of the welding point; The correction module is used to determine the geometric structural features at the welding point using image information in the neighborhood of the detection location, and to correct the detection location based on the geometric structural features to obtain the coordinates of the welding point.

9. An adaptive positioning device for a rebar welding robotic arm, which uses the device described in claim 8 to obtain the coordinates of the welding point, characterized in that... include: The lateral distance calculation module is used to calculate the actual lateral distance between the end of the robotic arm and the welding point based on the monocular vision ranging principle. The correction amount determination module is used to determine the reinforcing bars associated with the welding point, and to determine the lateral movement correction amount and rotation correction amount based on the inclination angle of the associated reinforcing bars. The alignment control module is used to drive the robotic arm to perform alignment actions according to the actual lateral distance, lateral movement correction amount, and rotation correction amount through visual feedback control.

Images