Weld feature point recognition method, device and equipment

Abstract

The application relates to a weld feature point recognition method, device and equipment. For each contour point in a weld line contour point cloud corresponding to a workpiece weld, a target window corresponding to the contour point under preset chord length construction parameters is determined; two end contour points of the target window are connected to form a chord, and distances of each contour point in the target window to the chord are accumulated to obtain a first curvature value corresponding to the contour point calculated based on the chord length construction parameters; based on the first curvature value corresponding to each contour point in the weld line contour point cloud, a set of inflection points is determined from the weld line contour point cloud; for each current inflection point in the set, from a second neighborhood range of the current inflection point, a previous contour point and a subsequent contour point on both sides of the current inflection point are selected; a concave-convex judgment is performed on a geometric structure formed by the current inflection point, the previous contour point and the subsequent contour point, and if the geometric structure is in a concave shape, the current inflection point is determined as a weld feature point. The scheme can improve weld feature point recognition accuracy.

Description

Technical Field

[0001] This application relates to the field of automatic welding technology, and in particular to a method, apparatus and equipment for identifying weld feature points. Background Technology

[0002] With the rapid development of intelligent manufacturing and industrial automation, welding equipment is evolving towards high automation and intelligence. Among these advancements, automated welding systems, primarily based on robots, have been widely applied in fields such as automobile manufacturing, pressure vessels, energy equipment, and shipbuilding. The ability to autonomously identify weld seam trajectories is crucial, and various vision-based weld seam recognition technologies are gradually becoming core components of automated welding systems.

[0003] Current vision-based weld seam recognition methods mainly fall into the following two categories: 1) Deep learning-based recognition method based on line structured light This method utilizes a line-structured light vision sensor to acquire images of the weld area for deep learning model training. Based on the trained deep learning model, image features are automatically extracted to determine the weld location. This method is highly dependent on the quality and quantity of training data. Ambient light interference, welding reflections, high-temperature fumes, and the complexity of weld morphology can all lead to a decrease in the algorithm's recognition stability. Furthermore, its ability to distinguish similar geometric features is limited, making it prone to misidentifying weld feature points. For example, the right-angled edges of lapped plates can easily be misidentified as weld boundaries, and the right-angled structure of the workpiece itself in corner joints can also be confused with the actual weld area, resulting in weld location identification errors.

[0004] 2) Traditional methods based on light stripe extraction and geometric intersection This type of method relies on the light stripe contour formed by structured light projection. After extracting the center line of the light stripe through image processing algorithms, the intersection points are calculated to determine the weld feature points. However, this method also suffers from inaccurate weld identification. For example, when dealing with scenarios with complex local geometric relationships, such as lap joints or corner joints, this method is prone to misidentifying weld feature points.

[0005] In summary, current traditional methods all suffer from the problem of easily misidentifying weld feature points. Summary of the Invention

[0006] Therefore, it is necessary to provide a weld feature point identification method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the accuracy of weld feature point identification in response to the above-mentioned technical problems.

[0007] Firstly, this application provides a method for identifying weld feature points. The method includes: Obtain the point cloud of the weld line contour corresponding to the weld seam of the workpiece; For each contour point in the weld line contour point cloud, a target window corresponding to the contour point under the preset chord length construction parameters is determined. The target window includes the contour point and the neighboring contour points within the first neighborhood range of the contour point indicated by the chord length construction parameters. The two contour points at the two ends of the target window are connected to form a chord. The distances from each contour point in the target window to the chord are accumulated to obtain the first curvature value corresponding to the contour point calculated based on the chord length construction parameters. Based on the first curvature value corresponding to each contour point in the weld line contour point cloud, inflection points are determined from the weld line contour point cloud to obtain the inflection point set. For each current inflection point in the inflection point set, select the preceding and following contour points located on both sides of the current inflection point from the second neighborhood of the current inflection point in the weld line contour point cloud. The concavity and convexity of the geometric structure formed by the current inflection point, the previous contour point, and the subsequent contour point are judged. If the geometric structure is concave, the current inflection point is determined to be a weld feature point.

[0008] In one embodiment, determining the target window corresponding to the contour points under preset chord length construction parameters includes: Construct a sliding window corresponding to the preset chord length construction parameters, and control the sliding window to move on the weld line contour point cloud to obtain multiple target windows corresponding to each contour point in the weld line contour point cloud; Connect the two contour points at the ends of the target window to form a chord. Sum the distances from each contour point within the target window to the chord to obtain the first curvature value corresponding to the contour point, calculated based on the chord length parameters. This includes: For each target window, connect the two ends of the target window to form a chord, and sum the distances from each contour point in the target window to the chord to obtain the sub-curvature value of the contour point under the target window. The sub-curvature values ​​corresponding to the contour points under multiple target windows are added together to obtain the first curvature value corresponding to the contour points calculated based on the chord length construction parameters.

[0009] In one embodiment, there are multiple preset chord length construction parameters; for each contour point, a first curvature value corresponding to the contour point is calculated based on each chord length construction parameter; based on the first curvature values ​​corresponding to each contour point in the weld line contour point cloud, inflection points are determined from the weld line contour point cloud to obtain an inflection point set, including: For each chord length construction parameter, a set of curvature values ​​calculated based on the chord length construction parameter is determined; the set of curvature values ​​includes the first curvature value calculated based on the chord length construction parameter for each contour point in the weld line contour point cloud; Normalize the set of curvature values ​​to obtain the normalized curvature values ​​of each contour point under the chord length construction parameters; For each contour point in the weld line contour point cloud, the normalized curvature values ​​corresponding to the contour point under multiple chord length construction parameters are multiplied together to obtain the second curvature value corresponding to the contour point. Based on the second curvature value corresponding to each contour point, inflection points are determined from the weld line contour point cloud to obtain the inflection point set.

[0010] In one embodiment, inflection points are determined from the weld line contour point cloud based on the second curvature value corresponding to each contour point, resulting in a set of inflection points, including: Based on the second curvature value corresponding to each contour point, the position where the sign of the rate of change of curvature is flipped is determined so as to identify the inflection point from the weld line contour point cloud and obtain the inflection point set.

[0011] In one embodiment, inflection points are determined from the weld line contour point cloud based on the second curvature value corresponding to each contour point, resulting in a set of inflection points, including: Based on the second curvature value corresponding to each contour point, the inflection point is initially determined from the weld line contour point cloud, and a candidate inflection point set is obtained. For each inflection point in the candidate inflection point set, if the angle between the tangents of the two neighboring regions on both sides of the inflection point is greater than a preset angle threshold or the second curvature value corresponding to the inflection point is less than a preset curvature intensity threshold, then the inflection point is determined to be a pseudo inflection point. False inflection points are removed from the candidate inflection point set to obtain the final inflection point set.

[0012] In one embodiment, the concavity / convexity of the geometric structure formed by the current inflection point, the preceding contour point, and the following contour point is determined, including: The point with the smaller x-coordinate between the preceding and following contour points is defined as the geometric left point, and the point with the larger x-coordinate between the preceding and following contour points is defined as the geometric right point. Based on the geometric left-side point, the current inflection point, and the geometric right-side point, construct a first feature vector and a second feature vector; the first feature vector is constructed by pointing from the geometric left-side point to the current inflection point; the second feature vector is constructed by pointing from the current inflection point to the geometric right-side point. Calculate the cross product of the first and second eigenvectors in two dimensions; If the cross product of the two-dimensional vectors is greater than 0 or a preset positive value, the geometric structure formed by the left geometric point, the current inflection point, and the right geometric point is determined to be a concave shape.

[0013] In one embodiment, the method further includes: If the cross product of the two-dimensional vector is less than 0 or a preset negative value, the geometric structure is determined to be an upward convex shape, and the current inflection point is the edge feature point of the workpiece.

[0014] In one embodiment, obtaining the weld line contour point cloud corresponding to the workpiece weld includes: Obtain the initial weld line contour point cloud corresponding to the weld seam of the workpiece; Gaussian smoothing is applied to the coordinate data of each contour point in the initial weld line contour point cloud to obtain the weld line contour point cloud.

[0015] Secondly, this application also provides a weld feature point identification device. The device includes: The point cloud acquisition module is used to acquire the point cloud of the weld line contour corresponding to the weld seam of the workpiece. The curvature calculation module is used to determine the target window corresponding to each contour point in the weld line contour point cloud under the preset chord length construction parameters. The target window includes the contour point and the neighboring contour points within the first neighborhood range of the contour point indicated by the chord length construction parameters. The two contour points at the two ends of the target window are connected to form a chord. The distances from each contour point in the target window to the chord are accumulated to obtain the first curvature value corresponding to the contour point calculated based on the chord length construction parameters. The inflection point recognition module is used to determine the inflection points from the weld line contour point cloud based on the first curvature value corresponding to each contour point in the weld line contour point cloud, and obtain the inflection point set. The feature point recognition module is used to select, for each current inflection point in the inflection point set, the preceding and following contour points located on both sides of the current inflection point from the second neighborhood of the current inflection point in the weld line contour point cloud; and to determine the concavity and convexity of the geometric structure formed by the current inflection point, the preceding contour points and the following contour points. If the geometric structure is concave, the current inflection point is determined to be a weld feature point.

[0016] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the method described in the first aspect above.

[0017] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method described in the first aspect above.

[0018] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps of the method described in the first aspect above.

[0019] The aforementioned weld feature point recognition method, apparatus, computer equipment, storage medium, and computer program product acquire a weld line contour point cloud corresponding to the workpiece weld. For each contour point in the weld line contour point cloud, a target window corresponding to the contour point under preset chord length construction parameters is determined. The target window includes the contour point and neighboring contour points within a first neighborhood range indicated by the chord length construction parameters. The two ends of the contour points in the target window are connected to form a chord. The distances from each contour point within the target window to the chord are accumulated to obtain a first curvature value corresponding to the contour point calculated based on the chord length construction parameters. Based on the first curvature values ​​corresponding to each contour point in the weld line contour point cloud, inflection points are determined from the weld line contour point cloud to obtain a set of inflection points. That is, through high curvature feature point detection processing based on point-to-chord distance accumulation (CPDA), each inflection point in the weld line contour point cloud can be preliminarily identified. Furthermore, for each current inflection point in the inflection point set, from the second neighborhood of the current inflection point in the weld line contour point cloud, the preceding and following contour points located on both sides of the current inflection point are selected; the concavity / convexity of the geometric structure formed by the current inflection point, the preceding contour points, and the following contour points is judged. If the geometric structure is concave, the current inflection point is determined to be a weld feature point. That is, based on the initially identified inflection point set, accurate weld feature point identification processing is performed based on geometric concavity / convexity judgment. In summary, combining high curvature feature point detection processing based on point-to-chord distance accumulation (CPDA) with accurate weld feature point identification processing based on geometric concavity / convexity judgment can accurately identify weld feature points. Attached Figure Description

[0020] Figure 1 This is a simplified flowchart of a weld feature point identification method in one embodiment; Figure 2 This is a flowchart illustrating a weld feature point identification method in one embodiment; Figure 3 This is a schematic diagram illustrating the cumulative distance from a point to a chord in one embodiment; Figure 4 This is a schematic diagram illustrating the inflection point recognition effect in one embodiment; Figure 5 This is a structural block diagram of a weld feature point recognition device in one embodiment; Figure 6 This is a structural block diagram of the curvature calculation module in one embodiment; Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] This application proposes a weld feature point identification method, which mainly includes two parts: the first part is the high curvature feature point detection processing based on the cumulative point-to-chord distance (CPDA), and the second part is the accurate identification processing of weld feature points based on geometric concavity and convexity judgment (or non-weld point removal processing).

[0023] like Figure 1 As shown, firstly, the weld line contour point cloud corresponding to the workpiece weld is obtained. This can be used as the basic data for weld feature point recognition. Then, coarse localization processing of the weld feature points is performed based on the point cloud curvature information. This coarse localization processing corresponds to the first part of the processing described above. After coarse localization of the weld feature points, further precise recognition processing of the weld feature points can be performed based on geometric concavity and convexity judgment. This precise recognition processing of the weld feature points corresponds to the second part of the processing described above.

[0024] Specifically, in the first part of the processing, a discrete curvature estimation technique based on point-to-chord distance accumulation (CPDA) is employed to achieve stable detection of high-curvature feature points in the weld line contour point cloud. This avoids the high sensitivity to noise inherent in traditional high-order differential curvature calculations, and still achieves reliable curvature response even in industrial line scan point clouds (i.e., weld line contour point clouds) with speckle noise and jitter. That is, after the first part of the processing, a set of high-curvature feature points identified from the weld line contour point cloud corresponding to the workpiece weld is obtained, which can also be called the inflection point set. The inflection point set is equivalent to the coarse localization result of the weld feature points.

[0025] Furthermore, by combining the processing described in the second part above, weld feature points can be accurately identified from the inflection point set, or non-weld feature points can be removed. That is, based on the coarse positioning results of weld feature points, weld feature points can be identified more accurately. For example, in the second part of the processing, a vector cross product concavity / convexity judgment method based on a unified geometric reference direction is introduced, which eliminates the influence of workpiece posture, scanning direction, and data order on the judgment results. Even under complex working conditions, it can still stably distinguish weld feature points (i.e., concave feature points) from workpiece edge feature points (i.e., convex feature points), solving the problem that relying solely on curvature cannot distinguish between welds and edges.

[0026] It should be understood that this application combines high-curvature feature point detection processing based on point-to-chord distance accumulation (CPDA) with accurate weld feature point identification processing based on geometric concavity / convexity judgment. Compared to traditional methods based on light stripe extraction and geometric intersection, which are prone to misidentifying other intersection points as weld feature points, this method can more accurately distinguish and identify weld feature points from the inflection point set, thus improving the accuracy of weld feature point identification. Furthermore, compared to traditional deep learning recognition methods based on line structured light, this method does not require deep learning training. Instead, it combines the weld line contour point cloud with the first and second parts of the processing described above, reducing the impact of ambient light interference, welding reflection, high-temperature fumes, and the complexity of weld morphology. This avoids, to some extent, the limitations of deep learning models and improves the accuracy of weld feature point identification.

[0027] The weld feature point identification method in this application will be described in more detail below.

[0028] In one embodiment, such as Figure 2 As shown, a method for identifying weld feature points is provided. Taking the application of this method to computer equipment as an example, the method includes the following steps: S21, Obtain the weld line contour point cloud corresponding to the weld seam of the workpiece.

[0029] It should be understood that, based on laser 3D contour measurement technology, a line profiler can be used to scan the weld seam of a workpiece to obtain the weld seam surface line profile point cloud data, which can be simply referred to as the weld seam line profile point cloud. It should also be understood that the weld seam line profile point cloud obtained by scanning with a line profiler is an ordered point cloud on the XOZ plane in the camera coordinate system. Points on the XOZ plane have a value of 0 in the Y-axis direction (i.e., the vertical axis, which can represent the depth direction). Here, the X-axis is the horizontal axis, which can represent the left-right direction, and the Z-axis is the vertical axis, which can represent the height direction.

[0030] In some embodiments, to eliminate quantization noise or minute jagged edges generated during data acquisition while preserving the overall geometric features of the curve, Gaussian filtering can be used to smooth the coordinate data of the initially acquired weld line contour point cloud. It should be understood that traditional methods require large-scale Gaussian smoothing due to the lack of other auxiliary processing. In this case, because parameters are created based on multi-scale chord lengths to enhance the accuracy of corner / inflection point identification, only small-scale Gaussian smoothing is needed to obtain a stable and accurate curvature response. This effectively avoids the curve shrinkage effect caused by traditional Gaussian smoothing and, to some extent, helps to achieve high-precision positioning of the true physical location of weld feature points.

[0031] Specifically, the computer device can acquire the initial weld line contour point cloud corresponding to the weld seam of the workpiece; and perform Gaussian smoothing on the coordinate data of each contour point in the initial weld line contour point cloud to obtain the weld line contour point cloud. For example, for each contour point, the coordinate data of each contour point within the Gaussian filter neighborhood of the contour point (i.e., the neighborhood range indicated by the window size of the Gaussian filter) are weighted and averaged according to the corresponding target Gaussian weights to obtain the smoothed coordinate data of the contour point, which is equivalent to obtaining the smoothed contour point.

[0032] In some examples, the standard deviation of the Gaussian function is set to... The formula for calculating the Gaussian kernel is as follows: ; Where W is the half-width of the Gaussian kernel, and k is an integer index representing the offset of the current point from the center of the Gaussian kernel. It is the original Gaussian weight at position k.

[0033] To ensure energy conservation during the filtering process, the Gaussian kernel is normalized using the following formula: ; in, This represents the normalized target Gaussian weight at position k. This represents the sum of the original Gaussian weights at all locations. This represents the original Gaussian weight at any position m, where m∈[-W,W].

[0034] In some examples, it is assumed that the initial weld line profile point cloud consists of N points arranged in sequence: ;x i It is the i-th contour point P i The original coordinates (i.e., original x-coordinate / original x-coordinate) in the horizontal axis (X-axis) direction, z i It is P i The original coordinates (i.e., the original z-coordinate / original vertical coordinate) in the vertical axis (Z-axis) direction.

[0035] Furthermore, Gaussian smoothing is applied to the x and z coordinates of any contour point. For example, Gaussian smoothing can be performed according to the following formula: , ; in, For the contour point P i The original coordinates x in the X-axis direction i The smoothed coordinates obtained after Gaussian smoothing. For the contour point P i The original coordinates z in the Z-axis directioni The smoothed coordinates obtained after Gaussian smoothing. It is the contour point P i Within the Gaussian filter neighborhood, and with contour point P i The original x-coordinates of the contour points at a distance of k. It is the contour point P i Within the Gaussian filter neighborhood, and with contour point P i The original z-coordinate of the contour point at a distance k.

[0036] It should be understood that performing step S22 and subsequent processing based on the weld seam contour point cloud formed by the smoothed contour points can improve the accuracy of weld seam feature point recognition to a certain extent.

[0037] In other embodiments, step S22 and subsequent processing can also be performed directly based on the initially acquired weld line contour point cloud. This is not a limitation.

[0038] S22, for each contour point in the weld line contour point cloud, determine the target window corresponding to the contour point under the preset chord length construction parameters; connect the two contour points at the two ends of the target window to form a chord, and accumulate the distances from each contour point in the target window to the chord to obtain the first curvature value corresponding to the contour point calculated based on the chord length construction parameters.

[0039] Here, the chord length construction parameter is not a basic attribute parameter of the chord (such as the length of the chord), but rather a parameter used to construct the chord. That is, based on the chord length construction parameter, a chord used to calculate the first curvature value can be constructed. Each contour point, within the target window corresponding to the chord length construction parameter, includes that contour point and other contour points within the first neighborhood range of that contour point indicated by the chord length construction parameter (which can be denoted as adjacent contour points).

[0040] It should be understood that the chord length construction parameter determines the span of the window used to construct the chord. Therefore, the chord length construction parameter can indicate the first neighborhood range of the contour point, which also matches the span of the window used to construct the chord. The span of the window characterizes the number of sub-intervals that the window spans (i.e., covers / includes). Two adjacent contour points constitute a sub-interval. In other words, the first neighborhood range is equivalent to the range covered by the window used to construct the chord. For example, assuming the chord length construction parameter is L, the span of the window is also L, that is, the window covers L sub-intervals, then the window covers L+1 contour points (i.e., there are L+1 contour points within the first neighborhood range). In the embodiments of this application, the chord is constructed by connecting the contour points at both ends of the corresponding window.

[0041] In some embodiments, a contour point can correspond to multiple target windows under the same chord length construction parameters. Step S22 includes: constructing a sliding window corresponding to the preset chord length construction parameters, controlling the sliding window to move on the weld line contour point cloud, and obtaining multiple target windows corresponding to each contour point in the weld line contour point cloud. For each target window, the contour points at both ends of the target window are connected to form a chord, and the distances (i.e., vertical Euclidean distances) from each contour point in the target window to the chord are accumulated to obtain the sub-curvature value corresponding to the contour point under the target window; the sub-curvature values ​​corresponding to the contour point under multiple target windows are added together to obtain the first curvature value corresponding to the contour point calculated based on the chord length construction parameters.

[0042] It should be understood that, assuming the chord length construction parameter is L, in this embodiment, for each contour point, during the process from the initial covering of the contour point by the sliding window to the sliding out and no longer covering the contour point, L+1 target windows corresponding to the contour point can be formed, and each target window includes the contour point.

[0043] by Figure 3 For example, suppose there are a total of n contour points (i.e., P1 to P2). n The chord length is constructed with a parameter L=5, and the span of the sliding window is also 5, meaning the sliding window covers 5 sub-intervals (two adjacent contour points form one sub-interval), thus covering 6 contour points. For any contour point P... i The first target window is formed when the sliding window initially covers the contour point. That is, the first target window covers the contour point P. i-L To the contour point P i After the sliding window slides once, the resulting contour point P... i The corresponding second target window covers the contour point P. i-L+1 To the contour point P i+1 The sliding window continues to slide, creating a target window corresponding to each contour point with each slide, until the contour point becomes the last target window corresponding to that contour point, which is the end contour point of the sliding window. That is, contour point P i The corresponding last target window covers the contour point P i To the contour point P i+L Thus, the sliding window initially covers the contour point P. i Until it slides out and no longer covers the contour point P i During the process, the contour point P was formed. i The corresponding 6 target windows.

[0044] In some examples, for each contour point, the distances from each contour point within each target window corresponding to that contour point to the corresponding chord of the target window (i.e., the chord formed by connecting the two contour points at the two ends of the target window) are accumulated, and the total accumulated distance is used as the sub-curvature value of that contour point under that target window. Then, the sub-curvature values ​​corresponding to that contour point under multiple target windows are added together to obtain the first curvature value corresponding to that contour point calculated based on the chord length construction parameters.

[0045] For example, any contour point P i The corresponding first curvature value h L The formula for calculating (i) is as follows: ; W'∈[w i-L , w i-L+1 ,..., w i+L-1 , w i+L ]; Where W' represents the sliding window starting from the initial coverage contour point P. i Until it slides out and no longer covers the contour point P i During the process, the contour points P formed i The set of corresponding target windows. For example, w i-L It is the contour point P i The first target window corresponds to the contour point P. i To the contour point P i-L For example, w i-L+1 It is the contour point P i The corresponding second target window covers the contour point P. i-L+1 To the contour point P i+1 .

[0046] d i,w Represents the contour point P i The cumulative distance (i.e., the total accumulated distance) from each contour point within any target window w to the chord corresponding to the target window w. For example, Figure 3 In the middle, d i, w i-L Represents the contour point P i The corresponding first target window w i-L From each contour point within the window to the first target window w i-L The corresponding chord (i.e., the first target window w) i-L Two end contour points P i and P i-L The cumulative distance of the chord formed by the connection; d i, w i-L+1 Represents the contour point P i The corresponding second target window w i-L+1From each contour point within the window to the second target window w i-L+1 The cumulative distance value of the corresponding chord.

[0047] It should be noted that in this example, d i,w This is also equivalent to the contour point P. i The sub-curvature value corresponding to any target window w; using the above formula, the contour point P i The sub-curvature values ​​corresponding to each target window are added together to obtain the contour point P. i The corresponding first curvature value h L (i). First curvature value h L The larger (i) is, the more severe the curvature is near that point.

[0048] It should be understood that by using a sliding window to form multiple target windows for a contour point, and adding the sub-curvature values ​​corresponding to the contour point under each of the multiple target windows, the first curvature value corresponding to the contour point can be obtained more accurately.

[0049] In other embodiments, a contour point can be a target window under the same chord length construction parameters. For example, a target window with a span matching the chord length construction parameters is constructed with the contour point as the center. Then, the contour points at both ends of the target window are connected to form a chord, and the distances from each contour point within the target window to the chord are accumulated. The total accumulated distance is used as the first curvature value corresponding to the contour point calculated based on the chord length construction parameters.

[0050] The inventors of this application discovered through research that in fillet welds and lap welds, weld feature points are typically distributed in the corner areas or geometric abrupt changes of the workpiece, where the contour curvature is significantly higher than that of non-weld areas. Based on this significant difference in curvature characteristics of weld feature points, preliminary location and screening of weld feature points can be achieved by analyzing the workpiece contour curvature. Traditional curvature calculation methods rely on calculating the first or second derivative of the curve. However, methods that calculate curvature using derivatives are highly sensitive to local noise and minute jitter on the curve, easily leading to false positives. The method of this application uses discrete curvature estimation technology, that is, determining the degree of curvature at a point on the curve by the cumulative vertical distance from the point to the "chord," which overcomes the aforementioned defects of the derivative curvature calculation method.

[0051] S23, Based on the first curvature value corresponding to each contour point in the weld line contour point cloud, determine the inflection point from the weld line contour point cloud to obtain the inflection point set.

[0052] In some embodiments, "multi-scale point-to-chord distance accumulation" can be used to measure discrete curvature. That is, there are multiple preset chord length construction parameters; for each contour point, step S22 is performed based on each chord length construction parameter to calculate the first curvature value corresponding to the contour point based on each chord length construction parameter. Then, the second curvature value corresponding to the contour point is calculated by combining the first curvature values ​​calculated for the contour point based on all chord length construction parameters. Then, based on the second curvature values ​​corresponding to each contour point, inflection points are determined from the weld line contour point cloud to obtain an inflection point set.

[0053] It should be understood that the cumulative distance values ​​(i.e., the first curvature values) calculated based on chord length construction parameters of different sizes have significant differences in numerical magnitude. To further improve robustness, the curvature value sets calculated based on each chord length construction parameter can be normalized, and then the curvature normalization results corresponding to multiple chord length construction parameters can be combined to calculate the final curvature value (i.e., the second curvature value) of each contour point.

[0054] Specifically, for each chord length construction parameter, a set of curvature values ​​calculated based on the chord length construction parameter is determined; the set of curvature values ​​includes a first curvature value calculated for each contour point in the weld line contour point cloud based on the chord length construction parameter; the set of curvature values ​​is normalized to obtain the normalized curvature value corresponding to each contour point under the chord length construction parameter; for each contour point in the weld line contour point cloud, the normalized curvature values ​​corresponding to the contour point under multiple chord length construction parameters are multiplied to obtain the second curvature value corresponding to the contour point; based on the second curvature value corresponding to each contour point, inflection points are determined from the weld line contour point cloud to obtain a set of inflection points.

[0055] To facilitate understanding, an example is given below. Suppose we select three different chord length construction parameters, denoted as L1, L2, and L3 (e.g., L1=5, L2=15, L3=35). Based on each chord length construction parameter, we calculate the corresponding first curvature value for each contour point. The first curvature values ​​corresponding to all contour points form the set of curvature values ​​calculated under that chord length construction parameter (also called a curvature sequence). That is, the set of curvature values ​​corresponding to the chord length construction parameters L1, L2, and L3 is h. L1 h L2 h L3 For the three curvature value sets h L1 h L2 h L3 Normalization is performed separately to obtain the normalized curvature value set h' L1 ,h' L2 ,h' L3 This maps its numerical range to the interval [0, 1]. For any contour point P... i The contour point P iMultiply the normalized curvature values ​​corresponding to the chord length construction parameters L1, L2, and L3 respectively, that is, multiply the contour points P. i In h' L1 ,h' L2 ,h' L3 The corresponding curvature values ​​h' in each case L1 (i), h' L2 (i), h' L3 (i) Multiply to obtain the contour point P. i The corresponding second curvature value H(i) = h' L1 (i)×h' L2 (i) ×h' L3 (i).

[0056] It should be understood that the above product operation utilizes the characteristic that "strong inflection points exhibit high response at different scales", which can significantly enhance the signal strength of the real inflection point, while effectively suppressing noise interference that only occurs at a single scale.

[0057] In some examples, inflection points are determined from the weld line contour point cloud based on the second curvature values ​​corresponding to each contour point, resulting in an inflection point set. Specifically, based on the second curvature values ​​corresponding to each contour point, the positions where the sign of the rate of change of curvature is flipped are determined to identify inflection points from the weld line contour point cloud, resulting in an inflection point set. More specifically, a state machine algorithm is used to scan the positions where the sign of the rate of change of curvature is flipped (from positive to negative) in the target curvature sequence formed by the second curvature values ​​corresponding to each contour point, thereby accurately locating all local maxima (peaks) in the target curvature sequence. The contour points corresponding to these local maxima constitute the inflection point set.

[0058] It should be noted that the inflection point set obtained above can be either the final inflection point set or a candidate inflection point set. Furthermore, false inflection points (such as smooth rounded corners or noise on straight lines) can be identified and removed from the candidate inflection point set to obtain the final inflection point set. For example, false inflection points can be identified based on a curvature intensity threshold filtering strategy or a geometric angle verification strategy, as described below.

[0059] In some examples, for each inflection point in the candidate inflection point set, if the angle α between the tangents of the two neighboring areas of the inflection point is greater than a preset angle threshold δ (e.g., preset δ=170°) or the second curvature value corresponding to the inflection point is less than a preset curvature intensity threshold T, then... h (The data scale can be set to 0.0001), then the inflection point is determined to be a false inflection point; false inflection points are removed from the candidate inflection point set to obtain the final inflection point set. It should be understood that when the angle α between the tangents of the two neighboring regions on both sides of the inflection point is greater than the preset angle threshold δ, it means that it is close to a straight line, and the inflection point is determined to be a false inflection point (or false corner point) and is removed.

[0060] In other embodiments, a single preset chord length construction parameter is used. In this case, the inflection points can be directly determined from the weld line contour point cloud based on the first curvature value corresponding to the contour points calculated using this parameter, thus obtaining an inflection point set. For example, based on the first curvature value corresponding to each contour point, the position where the sign of the rate of change of curvature is flipped is determined to identify the inflection points from the weld line contour point cloud, obtaining an inflection point set. It should be understood that in this example, the obtained inflection point set can also be either the final inflection point set or a candidate inflection point set. If it is a candidate inflection point set, a curvature intensity threshold filtering strategy or a geometric angle verification strategy can be used to identify and eliminate false inflection points, obtaining the final inflection point set. The specific principles are described above and will not be repeated here.

[0061] It should be understood that the inflection points retained in the final set of inflection points in the above embodiments are the detected high-confidence inflection points. These high-confidence inflection points (also known as high-curvature feature points) typically correspond to weld feature points or edge feature points of the workpiece, as shown in the attached figure. Figure 4 As shown. Therefore, in order to achieve accurate positioning of weld feature points, the geometric concavity and convexity judgment method can be further combined to distinguish weld feature points from workpiece edge feature points, as detailed in the relevant descriptions of steps S24 and S25.

[0062] S24, for each current inflection point in the inflection point set, select the preceding and following contour points located on both sides of the current inflection point from the second neighborhood range of the current inflection point in the weld line contour point cloud.

[0063] Specifically, a local geometric neighborhood is constructed for each current inflection point, and two auxiliary points are selected from it. For example, a set neighborhood step size j is obtained (j is a positive integer, such as any positive integer from 5 to 10). For any current inflection point P to be determined... target From the weld seam contour point cloud, select based on neighborhood step size The determined second neighborhood contains two auxiliary points, which are the preceding and following contour points located on either side of the current inflection point, respectively. For example, suppose the current inflection point P... target If is the i-th contour point in the weld line contour point cloud, then select the preceding contour point P with index position ij from the weld line contour point cloud. pre And the subsequent contour point P at index position i+j post .

[0064] S25, determine the concavity / convexity of the geometric structure formed by the current inflection point, the previous contour point, and the subsequent contour point. If the geometric structure is concave, then determine that the current inflection point is a weld feature point.

[0065] It should be understood that vector direction correction based on coordinate order can be performed, that is, the two auxiliary points (preceding contour points and subsequent contour points) are geometrically ordered to force a unified reference direction for vector construction.

[0066] Specifically, the preceding contour point P pre and subsequent contour point P post The point with the smaller x-coordinate is defined as the geometric left-side point P. left , the preceding contour point P pre and subsequent contour point P post The point with the larger x-coordinate is defined as the geometric right-side point P. right That is, satisfying: x l =min(x pre x post ), x r =max(x pre x post ).

[0067] In other words, the preceding contour point P pre x-coordinate pre and subsequent contour point P post x-coordinate post The minimum value in the middle is used as the x-coordinate of the point on the left side of the geometry. l Its x-coordinate l The corresponding contour point is the geometric left point P. left ; the preceding contour point P pre x-coordinate pre and subsequent contour point P post x-coordinate post The largest value in the middle is used as the x-coordinate of the point on the right side of the geometry. r Its x-coordinate r The corresponding contour point is the geometric right-side point P. right .

[0068] Furthermore, based on the geometric left-side point, the current inflection point, and the geometric right-side point, two contiguous feature vectors are constructed sequentially: the first feature vector and the second feature vector. The first feature vector, V1, is constructed by pointing from the geometric left-side point to the current inflection point: V1 = P target -P left The second eigenvector V2 is constructed from the point on the geometric right side pointed to by the current inflection point: V2 = P right -P target .

[0069] Furthermore, the cross product of the first eigenvector V1 and the second eigenvector V2 in two dimensions is calculated. In the XZ plane coordinate system, the formula for calculating the cross product CP is: CP = V1 × V2.

[0070] Furthermore, feature attributes are determined based on the sign of the calculated cross product value CP (taking a right-handed Cartesian coordinate system as an example, with the positive Z-axis pointing vertically upwards). Specifically, if the cross product value CP of the two-dimensional vector is greater than 0 or a preset positive value (i.e., a certain positive small threshold), it indicates that a counterclockwise rotation has occurred from vector V1 to V2. Therefore, the geometric structure formed by the geometric left-side point, the current inflection point, and the geometric right-side point is determined to be a concave shape. In this case, the current inflection point is determined to be a weld feature point.

[0071] In some embodiments, if the cross product of two-dimensional vectors is less than 0 or a preset negative value (i.e. a certain negative small threshold), it indicates that a clockwise rotation has occurred from vector V1 to V2. Therefore, the geometric structure is determined to be an upward convex shape, and the current inflection point is the edge feature point of the workpiece.

[0072] It should be understood that the judgment based on concavity and convexity can effectively remove the workpiece edge feature points in the inflection point obtained in the previous step, and can effectively solve the problem of easily misidentifying the position of lap welds and corner welds.

[0073] The above method combines high curvature feature point detection processing based on point-to-chord distance accumulation (CPDA) with weld feature point accurate identification processing based on geometric concavity and convexity judgment, which can accurately identify weld feature points.

[0074] In addition, a cascaded recognition framework combining curvature screening and concavity / convexity discrimination is constructed to realize the automatic extraction and classification of weld feature points, reduce manual intervention, and improve the automation and intelligence level of the welding inspection and tracking system.

[0075] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0076] Based on the same inventive concept, this application also provides a weld feature point identification device for implementing the weld feature point identification method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more weld feature point identification device embodiments provided below can be found in the limitations of the weld feature point identification method described above, and will not be repeated here.

[0077] In one embodiment, such as Figure 5 As shown, a weld feature point recognition device is provided, comprising: Point cloud acquisition module 502 is used to acquire the point cloud of the weld line contour corresponding to the weld seam of the workpiece. The curvature calculation module 504 is used to determine the target window corresponding to each contour point in the weld line contour point cloud under the preset chord length construction parameters. The target window includes the contour point and the neighboring contour points within the first neighborhood range of the contour point indicated by the chord length construction parameters. The two contour points at both ends of the target window are connected to form a chord. The distances from each contour point in the target window to the chord are accumulated to obtain the first curvature value corresponding to the contour point calculated based on the chord length construction parameters. The inflection point recognition module 506 is used to determine the inflection points from the weld line contour point cloud based on the first curvature value corresponding to each contour point in the weld line contour point cloud, and obtain the inflection point set. The feature point recognition module 508 is used to select, for each current inflection point in the inflection point set, the preceding and following contour points located on both sides of the current inflection point from the second neighborhood of the current inflection point in the weld line contour point cloud; and to determine the concavity and convexity of the geometric structure formed by the current inflection point, the preceding contour points and the following contour points. If the geometric structure is concave, then the current inflection point is determined to be a weld feature point.

[0078] like Figure 6 As shown, in some embodiments, the curvature calculation module 504 includes: The window sliding control module 5042 is used to construct a sliding window corresponding to the preset chord length construction parameters, and control the sliding window to move on the weld line contour point cloud to obtain multiple target windows corresponding to each contour point in the weld line contour point cloud. The distance accumulation module 5044 is used to connect the two ends of the contour points of the target window to form a chord for each target window, and accumulate the distances from each contour point in the target window to the chord to obtain the sub-curvature value of the contour point under the target window. The sub-curvature accumulation module 5046 is used to add the sub-curvature values ​​corresponding to the contour points under multiple target windows to obtain the first curvature value corresponding to the contour points calculated based on the chord length construction parameters.

[0079] In some embodiments, there are multiple preset chord length construction parameters; for each contour point, a first curvature value corresponding to the contour point is calculated based on each chord length construction parameter; the inflection point identification module 506 is further used to determine a set of curvature values ​​calculated based on each chord length construction parameter; the set of curvature values ​​includes a first curvature value calculated based on the chord length construction parameter for each contour point in the weld line contour point cloud; the set of curvature values ​​is normalized to obtain the normalized curvature value corresponding to each contour point under the chord length construction parameter; for each contour point in the weld line contour point cloud, the normalized curvature values ​​corresponding to the contour point under multiple chord length construction parameters are multiplied to obtain a second curvature value corresponding to the contour point; based on the second curvature values ​​corresponding to each contour point, inflection points are determined from the weld line contour point cloud to obtain a set of inflection points.

[0080] In some embodiments, the inflection point recognition module 506 is further configured to determine the position where the sign of the rate of change of curvature is flipped based on the second curvature value corresponding to each contour point, so as to determine the inflection point from the weld line contour point cloud and obtain the inflection point set.

[0081] In some embodiments, the inflection point identification module 506 is further configured to preliminarily determine inflection points from the weld line contour point cloud based on the second curvature values ​​corresponding to each contour point, and obtain a candidate inflection point set; for each inflection point in the candidate inflection point set, if the angle between the tangents of the neighborhood on both sides of the inflection point is greater than a preset angle threshold or the second curvature value corresponding to the inflection point is less than a preset curvature intensity threshold, then the inflection point is determined to be a false inflection point; and the false inflection points are removed from the candidate inflection point set to obtain the final inflection point set.

[0082] In some embodiments, the feature point recognition module 508 is further configured to define the point with the smaller abscissa among the preceding and following contour points as the geometric left point, and the point with the larger abscissa among the preceding and following contour points as the geometric right point; construct a first feature vector and a second feature vector based on the geometric left point, the current inflection point, and the geometric right point; the first feature vector is constructed by pointing from the geometric left point to the current inflection point; the second feature vector is constructed by pointing from the current inflection point to the geometric right point; calculate the two-dimensional vector cross product value of the first feature vector and the second feature vector; if the two-dimensional vector cross product value is greater than 0 or a preset positive value, determine that the geometric structure formed by the geometric left point, the current inflection point, and the geometric right point is a concave shape.

[0083] In some embodiments, the feature point recognition module 508 is further used to determine that the geometric structure is convex and the current inflection point is a feature point on the edge of the workpiece if the cross product of the two-dimensional vector is less than 0 or a preset negative value.

[0084] In some embodiments, the point cloud acquisition module 502 is further configured to acquire the initial weld line contour point cloud corresponding to the workpiece weld; and to perform Gaussian smoothing on the coordinate data of each contour point in the initial weld line contour point cloud to obtain the weld line contour point cloud.

[0085] Each module in the aforementioned weld feature point recognition device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in the processor of a computer device in hardware form or independent of it, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0086] In one embodiment, a computer device is provided, the internal structure of which can be shown as follows: Figure 7 As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a weld seam feature point recognition method.

[0087] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the methods involved in the embodiments of this application.

[0088] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the methods involved in the embodiments of this application.

[0089] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the methods involved in the embodiments of this application.

[0090] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0091] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0092] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for identifying weld feature points, characterized in that, The method includes: Obtain the point cloud of the weld line contour corresponding to the weld seam of the workpiece; For each contour point in the weld line contour point cloud, a target window corresponding to the contour point under a preset chord length construction parameter is determined. The target window includes the contour point and neighboring contour points within the first neighborhood range of the contour point indicated by the chord length construction parameter. The two contour points at the two ends of the target window are connected to form a chord. The distances from each contour point in the target window to the chord are accumulated to obtain the first curvature value corresponding to the contour point calculated based on the chord length construction parameter. Based on the first curvature value corresponding to each contour point in the weld line contour point cloud, inflection points are determined from the weld line contour point cloud to obtain an inflection point set. For each current inflection point in the set of inflection points, select the preceding and following contour points located on both sides of the current inflection point from the second neighborhood of the current inflection point in the weld line contour point cloud. The concavity / convexity of the geometric structure formed by the current inflection point, the preceding contour point, and the following contour point is determined. If the geometric structure is concave, the current inflection point is determined to be a weld feature point.

2. The method according to claim 1, characterized in that, Determining the target window corresponding to the contour point under preset chord length construction parameters includes: Construct a sliding window corresponding to the preset chord length construction parameters, and control the sliding window to move on the weld line contour point cloud to obtain multiple target windows corresponding to each contour point in the weld line contour point cloud; The step of connecting the two contour points of the target window to form a chord, and summing the distances from each contour point within the target window to the chord to obtain the first curvature value corresponding to the contour point calculated based on the chord length construction parameters, includes: For each target window, the two ends of the target window are connected to form a chord, and the distances from each contour point in the target window to the chord are accumulated to obtain the sub-curvature value of the contour point under the target window. The sub-curvature values ​​corresponding to the contour points under the multiple target windows are added together to obtain the first curvature value corresponding to the contour points calculated based on the chord length construction parameters.

3. The method according to claim 1, characterized in that, There are multiple preset chord length construction parameters; for each contour point, the first curvature value corresponding to the contour point is calculated based on each chord length construction parameter. The inflection points are determined from the weld line contour point cloud based on the first curvature values ​​corresponding to each contour point, resulting in a set of inflection points, including: For each chord length construction parameter, a set of curvature values ​​calculated based on the chord length construction parameter is determined; the set of curvature values ​​includes a first curvature value calculated based on the chord length construction parameter for each contour point in the weld line contour point cloud; The curvature value set is normalized to obtain the normalized curvature value of each contour point under the chord length construction parameters; For each contour point in the weld line contour point cloud, the normalized curvature values ​​corresponding to the contour point under multiple chord length construction parameters are multiplied together to obtain the second curvature value corresponding to the contour point. Based on the second curvature value corresponding to each contour point, inflection points are determined from the weld line contour point cloud to obtain an inflection point set.

4. The method according to claim 3, characterized in that, The inflection points are determined from the weld line contour point cloud based on the second curvature value corresponding to each contour point, resulting in a set of inflection points, including: Based on the second curvature value corresponding to each contour point, the position where the sign of the rate of change of curvature is flipped is determined so as to identify the inflection point from the weld line contour point cloud and obtain the inflection point set.

5. The method according to claim 3, characterized in that, The inflection points are determined from the weld line contour point cloud based on the second curvature value corresponding to each contour point, resulting in a set of inflection points, including: Based on the second curvature value corresponding to each contour point, inflection points are initially determined from the weld line contour point cloud to obtain a candidate inflection point set. For each inflection point in the candidate inflection point set, if the angle between the tangents of the two neighboring regions on both sides of the inflection point is greater than a preset angle threshold or the second curvature value corresponding to the inflection point is less than a preset curvature intensity threshold, then the inflection point is determined to be a pseudo inflection point. The false inflection points are removed from the candidate inflection point set to obtain the final inflection point set.

6. The method according to claim 1, characterized in that, The step of determining the concavity / convexity of the geometric structure formed by the current inflection point, the preceding contour point, and the following contour point includes: The point with the smaller abscissa among the preceding and following contour points is defined as the geometric left point, and the point with the larger abscissa among the preceding and following contour points is defined as the geometric right point. Based on the geometric left-side point, the current inflection point, and the geometric right-side point, a first feature vector and a second feature vector are constructed; the first feature vector is constructed by pointing from the geometric left-side point to the current inflection point; the second feature vector is constructed by pointing from the current inflection point to the geometric right-side point. Calculate the cross product of the first feature vector and the second feature vector in two dimensions; If the cross product of the two-dimensional vector is greater than 0 or a preset positive value, the geometric structure formed by the geometric left point, the current inflection point, and the geometric right point is determined to be a concave shape.

7. The method according to claim 6, characterized in that, The method further includes: If the cross product of the two-dimensional vector is less than 0 or a preset negative value, the geometric structure is determined to be an upward convex shape, and the current inflection point is a feature point on the edge of the workpiece.

8. The method according to any one of claims 1 to 7, characterized in that, The step of obtaining the weld line contour point cloud corresponding to the workpiece weld includes: Obtain the initial weld line contour point cloud corresponding to the weld seam of the workpiece; Gaussian smoothing is applied to the coordinate data of each contour point in the initial weld line contour point cloud to obtain the weld line contour point cloud.

9. A weld seam feature point identification device, characterized in that, The device includes: The point cloud acquisition module is used to acquire the point cloud of the weld line contour corresponding to the weld seam of the workpiece. The curvature calculation module is used to determine the target window corresponding to each contour point in the weld line contour point cloud under a preset chord length construction parameter. The target window includes the contour point and the neighboring contour points within the first neighborhood range of the contour point indicated by the chord length construction parameter. The two contour points at both ends of the target window are connected to form a chord. The distances from each contour point in the target window to the chord are accumulated to obtain the first curvature value corresponding to the contour point calculated based on the chord length construction parameter. The inflection point recognition module is used to determine inflection points from the weld line contour point cloud based on the first curvature value corresponding to each contour point in the weld line contour point cloud, and obtain an inflection point set. The feature point recognition module is used to select, for each current inflection point in the inflection point set, the preceding and following contour points located on both sides of the current inflection point from the second neighborhood of the current inflection point in the weld line contour point cloud; and to determine the concavity and convexity of the geometric structure formed by the current inflection point, the preceding contour points, and the following contour points. If the geometric structure is concave, then the current inflection point is determined to be a weld feature point.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

Images