Point cloud edge segmentation method and system suitable for complex industrial parts

Abstract

The invention provides a point cloud edge segmentation method and system suitable for a complex industrial part, and the method comprises the steps: extracting a point cloud neighborhood point estimation normal vector of each point in point cloud data after the sampling processing of a voxel grid, carrying out the analysis processing based on the point cloud neighborhood point estimation normal vector, and obtaining an enhanced curvature; utilizing a Canny edge detection algorithm to extract an edge region in a two-dimensional view after two-dimensional processing of the point cloud data, and utilizing a communication component to screen to obtain an effective edge region; and screening the enhanced curvature and the effective edge region based on a preset threshold to obtain a segmentation result. According to the method, the extraction precision and accuracy of the point cloud edge features are effectively improved, omission and redundancy of the edge features are avoided, the reliability of an edge extraction result is improved, parts of complex geometric structures can be processed, and the method is suitable for the fields of industrial detection, reverse engineering and the like.

Description

Technical Field

[0001] The present invention belongs to the technical field of three-dimensional point cloud processing, and in particular to a point cloud edge segmentation method and system suitable for complex industrial parts. Background Art

[0002] With the rapid development of industrial manufacturing, the demand for the design and manufacture of complex parts is increasing. Especially in the fields of aerospace, automobile manufacturing and precision instruments, the geometric structure of parts is becoming increasingly complex, and the requirements for processing accuracy and quality control are also higher. As a key tool in industrial inspection and reverse engineering, 3D point cloud technology can effectively capture the external morphology and structural information of parts. Point cloud data obtained by laser scanning, optical measurement and other technologies contains rich 3D coordinate information, which provides data support for subsequent geometric analysis, dimensional measurement and defect detection.

[0003] Due to the complexity of the geometric shapes of complex industrial parts, industrial products often present a variety of geometric characteristics such as planes, edges, bumps, holes and other complex surfaces. At the same time, the conversion between geometric elements is very complex, and it is difficult to accurately extract the boundaries and corners in the image using traditional segmentation algorithms based on geometric criteria. In addition, complex structures often have areas with large turning changes, which brings great challenges to the design of point cloud edge segmentation algorithms; the point cloud data of industrial parts facing massive data processing, the number of which is as high as one million, poses a great challenge to its computing power. How to improve the efficiency of subsequent point cloud processing calculations and reduce data redundancy while ensuring the accuracy of point cloud segmentation is an important issue faced in actual engineering.

[0004] Traditional point cloud segmentation methods use region growing or clustering algorithms, such as dividing geometric features based on the similarity or distance between points. When faced with industrial parts with complex structures or fine edges, such methods cannot accurately capture subtle geometric features and have strong parameter dependence, especially at edge feature points, which can easily lead to over-segmentation or under-segmentation. However, curvature can effectively reflect the geometric features of local areas of point clouds, especially at edges and corners, where there are significant changes. Therefore, curvature analysis can distinguish between planes, edges and corners. However, segmentation methods that only rely on curvature have great errors in the extracted edge point features in industrial parts processing, resulting in poor reliability and low accuracy of the segmentation results. Summary of the invention

[0005] In order to solve the problem that the existing methods for segmenting industrial parts have great errors in extracting edge point features, resulting in poor reliability and low accuracy of segmentation results, the present invention provides a point cloud edge segmentation method and system suitable for complex industrial parts.

[0006] To achieve the above object, the present invention provides the following technical solutions: The present invention proposes a point cloud edge segmentation method suitable for complex industrial parts, comprising the following steps: Extract the estimated normal vector of the point cloud neighborhood point of each point in the point cloud data after the voxel grid downsampling processing, and perform analysis and processing based on the estimated normal vector of the point cloud neighborhood point to obtain enhanced curvature; The Canny edge detection algorithm is used to extract the edge area in the two-dimensional view after the point cloud data is processed in two dimensions, and the connected components are used to filter and obtain the effective edge area; The enhanced curvature and the effective edge area are screened based on a preset threshold to obtain a segmentation result.

[0007] Preferably, extracting the estimated normal vector of the point cloud neighborhood point of each point in the point cloud data after the voxel grid downsampling process includes: Collect 3D point cloud data of complex workpieces; Divide the 3D space where the 3D point cloud data is located into multiple voxels , count each voxel All points within , select each voxel The centroid point in , using the centroid point Replace each voxel All points within , the centroid point Retain in the downsampled point cloud to obtain a downsampled point cloud dataset; Extract the downsampled point cloud data set, each centroid point Estimated normal vector of the point cloud neighborhood point.

[0008] Preferably, the step of selecting each voxel The centroid point in The process is:

[0009] Where n is the voxel The number of points in Voxel The point in .

[0010] Preferably, in the extracted downsampled point cloud data set, each centroid point The estimated normal vector of the point cloud neighborhood point includes: Based on the centroid point , determine the centroid point The set of neighboring points ; For the neighborhood point set Perform centralization and calculate the covariance matrix ; For the covariance matrix Perform eigenvalue decomposition to obtain eigenvalues ​​and eigenvectors corresponding to the eigenvalues, that is, obtain estimated normal vectors of point cloud neighborhood points; The enhanced curvature is obtained by performing analysis based on the eigenvalue and the estimated normal vector of the point cloud neighborhood point.

[0011] Preferably, the covariance matrix The calculation process is:

[0012] in, Neighborhood point set The number of midpoints, Neighborhood point set The mean of the midpoints, is the centroid value.

[0013] Preferably, the analyzing and processing based on the eigenvalue and the estimated normal vector of the point cloud neighborhood point to obtain the enhanced curvature includes: Obtaining a neighborhood point normal vector of a first neighborhood point corresponding to a point whose estimated normal vector is a neighborhood point of the point cloud; Calculate the difference angle between the estimated normal vector of the point cloud neighborhood point and the normal vector of the domain point ; Calculate the curvature based on the eigenvalue; Based on the difference angle and the curvature, and the enhanced curvature is calculated.

[0014] Preferably, the extracting edge areas in the two-dimensional view after two-dimensional processing of the point cloud data by using the Canny edge detection algorithm comprises: Based on the set resolution and the size of the plane, and construct the projection plane; Projecting the point cloud data onto the projection plane to obtain a two-dimensional image; Performing Gaussian filtering on the two-dimensional image to obtain a smoothed image; The gradient amplitude of the smoothed image is calculated using the Sobel operator and the gradient direction , and the gradient amplitude along the gradient direction Perform non-maximum suppression and retain local maxima; Based on the set threshold, the suppressed gradient amplitude is Detect and get edge image ; The edge image is constructed using connected components Screening is performed to obtain the effective edge area.

[0015] Preferably, the edge image is processed by using connected components. Screening is performed to obtain effective edge areas, including: Using connected components to map edge images Perform analysis and screen the complete edge area; Get the perimeter of the connected area in the entire edge region , based on a preset perimeter threshold The perimeter of the connected region Compare and retain the perimeter of the connected area Greater than the perimeter threshold The edge area is obtained by calculating the effective edge area. Preferably, the screening of the enhanced curvature and the effective edge area based on a preset threshold to obtain a segmentation result includes: Obtain the second area point corresponding to the point of enhanced curvature, and obtain the neighborhood point set by statistics , calculate the neighborhood point set Mean curvature of the medium enhanced curvature ; For the mean curvature Normalization is performed so that the curvature mean Go to the same interval; The transformed curvature mean The converted curvature mean values ​​greater than the first screening threshold are retained. , extract the retained curvature mean The corresponding point is the first edge point; Matching edge points in the effective edge area with projection points of the edge area in the corresponding two-dimensional view; Calculate the two-dimensional Euclidean distance between two points that can be matched; Comparing the two-dimensional Euclidean distance with a preset second screening threshold, retaining edge points in the valid edge region whose two-dimensional Euclidean distance is less than the second screening threshold, and obtaining second edge points; The first edge points and the second edge points are collected to construct an edge point set to obtain the segmentation result.

[0016] The present invention proposes a point cloud edge segmentation system suitable for complex industrial parts, which is used to implement the above method, including: The first processing unit is configured as follows: Used to extract the estimated normal vector of the point cloud neighborhood point of each point in the point cloud data after voxel grid downsampling processing, perform analysis and processing based on the estimated normal vector of the point cloud neighborhood point to obtain enhanced curvature, and extract edge feature points based on the enhanced curvature; The second processing unit is configured as follows: It is used to extract the edge area in the two-dimensional view after two-dimensional processing of point cloud data using the Canny edge detection algorithm, and obtain the effective edge area by filtering using connected components; The third processing unit is configured as follows: Used to filter the edge feature points and the effective edge area based on a preset threshold to obtain a segmentation result; Output unit, configured as: Used to output segmentation results.

[0017] Compared with the prior art, the present invention has the following beneficial technical effects: The present invention proposes a point cloud edge segmentation method suitable for complex industrial parts. The method effectively reduces data redundancy and enhances curvature features through voxel grid downsampling processing and point cloud neighborhood point estimation normal vector, laying a solid foundation for subsequent processing. Combined with the processing of two-dimensional views by the Canny edge detection algorithm, the mature image processing technology is fully utilized to convert three-dimensional point cloud data into two-dimensional information, which is convenient for capturing subtle edge features. The connected component screening step further purifies the edge area, effectively removes noise and irrelevant details, and ensures the continuity and integrity of the edge. Based on the preset threshold, the enhanced curvature and the effective edge area are double screened to ensure the accuracy and reliability of the segmentation results. The method effectively improves the extraction precision and accuracy of point cloud edge features, avoids the omission and redundancy of edge features, improves the reliability of edge extraction results, can process parts with complex geometric structures, and is suitable for industrial inspection, reverse engineering and other fields.

[0018] Furthermore, this method effectively reduces data density and computational burden through voxel grid downsampling while retaining key geometric features. In the process of dividing voxels and calculating centroid points, it achieves simplification and feature enhancement of point cloud data, providing a high-quality data foundation for subsequent processing. For the downsampled point cloud data set, the point cloud neighborhood point estimated normal vector of each centroid point is accurately extracted, the recognition ability of curvature features is enhanced, and more accurate directional information is provided for subsequent edge detection and segmentation, which ensures processing speed while significantly improving the accuracy and robustness of edge segmentation.

[0019] Furthermore, this method first determines the set of neighborhood points of each centroid point, then calculates the covariance matrix and decomposes it to obtain eigenvalues ​​and eigenvectors, accurately estimates the normal vector, and effectively captures the local geometric features. Further analysis and processing based on the eigenvalues ​​and normal vectors enhances the curvature, makes the edge features more prominent, improves the accuracy of edge segmentation, and enhances the robustness of the algorithm, significantly improving the efficiency and quality of point cloud data processing.

[0020] Furthermore, this method effectively captures the change in the direction of the normal vector by calculating the difference angle between the two, and combines the curvature obtained by eigenvalue calculation, comprehensively considering the intrinsic properties of the surface, and calculates the enhanced curvature based on the comprehensive analysis of the difference angle and curvature, which improves the accuracy of curvature estimation and enhances the algorithm's ability to recognize complex surface structures.

[0021] Furthermore, this method constructs a projection plane, accurately projects the point cloud data into a two-dimensional image, and effectively smoothes the image through Gaussian filtering to reduce noise interference. The Sobel operator is used to accurately calculate the gradient amplitude and gradient direction, and the significant edge features in the image are retained through non-maximum suppression. Gradient amplitude detection is performed based on the set threshold to further purify the edge information and obtain a clear edge image. Through connected component screening, isolated edge points are effectively removed, and continuous and meaningful edge areas are retained.

[0022] Furthermore, this method uses connected components to analyze edge images, identifies complete edge regions, effectively removes isolated points and broken edges, retains continuous and meaningful edge features, and further screens out edge regions with significant geometric significance by calculating the perimeter of each connected region and comparing it with a preset perimeter threshold, while excluding small or insignificant edges.

[0023] Furthermore, this method effectively balances the difference in curvature values ​​in different regions by calculating the mean of the neighborhood curvature of each point corresponding to the enhanced curvature and performing normalization processing. By setting the first screening threshold, points with significant curvature characteristics, namely the first edge points, are retained, thereby enhancing the accuracy of edge recognition. At the same time, the edge points in the effective edge region are accurately matched with the projection points in the two-dimensional view, and the two-dimensional Euclidean distance is calculated. By setting the second screening threshold, the edge points are further purified and the second edge points are obtained, thereby improving the accuracy of edge segmentation and ensuring the stability and reliability of the segmentation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 A schematic diagram of a flow chart of a point cloud edge segmentation method applicable to complex industrial parts proposed by the present invention; Figure 2 It is one of the parts drawings of the automobile flange blank of the embodiment; Figure 3 This is the second example of the automotive flange blank parts diagram; Figure 4 This is the third example of the automotive flange blank parts diagram; Figure 5 It is a map before enhancing curvature in a point cloud edge segmentation method applicable to complex industrial parts proposed by the present invention; Figure 6 This is a graph after enhancing curvature in a point cloud edge segmentation method applicable to complex industrial parts proposed by the present invention; Figure 7 This is one of the schematic diagrams of connected region analysis and edge screening in a point cloud edge segmentation method applicable to complex industrial parts proposed by the present invention; Figure 8 This is the second schematic diagram of connected region analysis and edge screening in a point cloud edge segmentation method applicable to complex industrial parts proposed by the present invention; Fig. 9 It is one of the edge point cloud images after matching in a point cloud edge segmentation method applicable to complex industrial parts proposed by the present invention; Fig.10 This is the second edge point cloud image after matching in the point cloud edge segmentation method applicable to complex industrial parts proposed by the present invention; Fig.11 This is the third edge point cloud image after matching in the point cloud edge segmentation method applicable to complex industrial parts proposed by the present invention; Fig.12 This is one of the edge point result images after segmentation in a point cloud edge segmentation method applicable to complex industrial parts proposed by the present invention; Fig.13 This is the second result diagram of edge points after segmentation in the point cloud edge segmentation method suitable for complex industrial parts proposed by the present invention. DETAILED DESCRIPTION

[0025] In the following, only some exemplary embodiments are briefly described. As those skilled in the art will appreciate, the described embodiments may be modified in various ways without departing from the spirit or scope of the present invention. Therefore, the drawings and descriptions are considered to be exemplary and non-restrictive in nature.

[0026] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0027] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0028] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, an electrical connection, or a communication; it can be a direct connection, or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0029] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may include that the first and second features are in direct contact, or may include that the first and second features are not in direct contact but are in contact through another feature between them. Moreover, a first feature being "above", "above" and "above" a second feature includes that the first feature is directly above and obliquely above the second feature, or simply indicates that the first feature is higher in level than the second feature. A first feature being "below", "below" and "below" a second feature includes that the first feature is directly above and obliquely above the second feature, or simply indicates that the first feature is lower in level than the second feature.

[0030] The embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0031] The present invention proposes a point cloud edge segmentation method suitable for complex industrial parts, such as Figure 1 As shown, the following steps are included: Extract the estimated normal vector of the point cloud neighborhood point of each point in the point cloud data after the voxel grid downsampling processing, and perform analysis and processing based on the estimated normal vector of the point cloud neighborhood point to obtain enhanced curvature; Specifically, the NimbleTrack smart 3D scanning system is used to collect 3D point cloud data of complex workpieces. In this embodiment, an automobile flange blank part is taken as an example. Figure 2~Figure 4 Among them, NimbleTrack has high-precision measurement capabilities and can obtain detailed surface information of objects with an accuracy of 0.025mm. For example, use a handheld scanner to start scanning from one side of the workpiece, move the scanner slowly and evenly, and ensure that the scanner always maintains an appropriate distance from the surface of the object. The scanner will capture the surface information in real time and convert it into point cloud data. The number of 3D point cloud data obtained by the scanning system is 449,730 points; Set a voxel The size s is used to divide the 3D space where the 3D point cloud data is located into A cube grid is formed to obtain multiple voxels , at each voxel In the above example, each voxel is counted All points within , select each voxel The centroid point in As a representative point, use the centroid point Replace each voxel All points within , the centroid point The points are retained in the downsampled point cloud to obtain a downsampled point cloud data set; the 3D point cloud data is processed by voxel grid downsampling, retaining the basis of the point cloud geometry, and completing the simplification of the 3D point cloud data. For example, the number of 3D point cloud data is 449,730 points, and the points processed by voxel grid downsampling are 52,447, which is about 10% of the original point cloud.

[0032] Select each voxel The centroid point in The process is:

[0033] Where n is the voxel The number of points in Voxel The point in .

[0034] Based on centroid , determine the centroid through K nearest neighbor search The set of neighboring points ; Neighborhood point set Contains point The geometric neighborhood information can provide reliable local support for curvature and normal vector calculation; for the neighborhood point set Perform centralization and calculate the covariance matrix ; Covariance matrix The structural information of the neighborhood point in the neighborhood is provided to facilitate further calculation of curvature and normal vector.

[0035] Covariance matrix The calculation process is:

[0036] in, Neighborhood point set The number of midpoints, Neighborhood point set The mean of the midpoints, is the centroid value.

[0037] Covariance matrix Perform eigenvalue decomposition to obtain the eigenvalue and the eigenvector corresponding to the eigenvalue, that is, to obtain the estimated normal vector of the point cloud neighborhood point; Obtain the normal vector of the first neighborhood point of the point corresponding to the estimated normal vector of the neighborhood point of the point cloud; Calculate the difference angle between the estimated normal vector of the point cloud neighborhood point and the normal vector of the domain point ; Difference Angle The calculation process is:

[0038] Seek the middle, The centroid The normal vector of the point cloud neighborhood point is The centroid The normal vector of the first domain point, that is, the domain point normal vector; The curvature is calculated based on the eigenvalue , curvature When the value is large, it means that the point is located at the surface mutation point, which is more likely to be an edge point. and curvature, the enhanced curvature is calculated ; Enhanced curvature It can more sensitively capture tiny edge changes, improve the accuracy of edge segmentation, and enhance curvature. The before and after graphs are as follows Figure 5 and Figure 6 shown.

[0039] Enhanced curvature The calculation process is:

[0040] in, is the curvature, is the difference angle.

[0041] The Canny edge detection algorithm is used to extract the edge area in the two-dimensional view after the point cloud data is processed in two dimensions, and the connected components are used to filter and obtain the effective edge area; Specifically, set the resolution of the two-dimensional projection plane , that is, the size of each unit area, to construct the projection plane, namely the XOZ plane; where the width of the two-dimensional projection plane is , the height of the two-dimensional projection plane is ; The acquired 3D point cloud data is projected onto the XOZ plane to obtain a 2D image; wherein, for each point in the 3D point cloud data , projecting it to the coordinates on the XOZ plane Defined as: , ,neglect The two-dimensional coordinate point set formed after the coordinates are formed constitutes the plane data after projection. Mapping to image coordinates The relationship is:

[0042]

[0043] Gaussian filtering is performed on the two-dimensional image to smooth the image and reduce noise to obtain a smoothed image; wherein the filter kernel of the preset Gaussian filter is , then the standard deviation Controls the degree of smoothing, the pixel value after smoothing for:

[0044] in, represents the radius of the nucleus, and are the offsets of the filter kernel in the horizontal and vertical directions respectively.

[0045] The gradient amplitude of the smoothed image is calculated using the Sobel operator and the gradient direction , in calculating the gradient amplitude and the gradient direction Then, based on the gradient direction Non-maximum suppression is performed on the gradient amplitude, along the gradient direction , check the pixels For the adjacent pixels on both sides of the gradient amplitude, the pixel value with the largest gradient amplitude is retained, and the remaining pixels are set to zero, which can effectively remove the gradient response of non-edge points, further optimize the accuracy of edge detection, and retain the local maximum value; Gradient Amplitude The calculation process is:

[0046] Gradient direction The calculation process is:

[0047] in, and The images are and Direction gradient.

[0048] Then, set the high threshold and low threshold , based on the set threshold, the suppressed gradient amplitude is Perform detection, if the gradient amplitude Greater than , it is marked as a strong edge; if , then it is marked as a weak edge and the edge image is obtained ; Edge Image It can be expressed in the following way:

[0049] Among them, 1 represents the previous pixel It is an edge point, including two cases: (1) Gradient amplitude Greater than high threshold , directly marked as a strong edge point.

[0050] (2) Gradient amplitude In If the point is within the range of , and is directly or indirectly connected to a strong edge point, it is marked as a weak edge point and retained as an edge.

[0051] 0 represents the current pixel Not an edge point, including two cases: gradient amplitude Less than low threshold , gradient amplitude In The others refer to all other pixel points except the above edge points, which are finally classified as non-edge points.

[0052] Using connected components to map edge images Perform analysis and filter out complete edge areas, such as Figure 7 and Figure 8 , get the perimeter of the connected area in the entire edge area , based on a preset perimeter threshold Perimeter of connected region Compare and retain the perimeter of the connected area Greater than perimeter threshold The edge area is obtained by calculating the effective edge area.

[0053] Based on preset perimeter thresholds Perimeter of connected region Compare and retain the perimeter of the connected area Greater than perimeter threshold The edge area of ​​​​the effective edge area is simplified as follows:

[0054] in, Indicates connected regions are from the edge image The independent areas separated from Indicates that all conditions are met The set of connected regions, that is, the set of valid edge regions.

[0055] The enhanced curvature and the effective edge area are screened based on a preset threshold to obtain a segmentation result.

[0056] Specifically, the enhanced curvature obtained The second area point of the corresponding point is statistically obtained as a set of neighborhood points , calculate the neighborhood point set Medium Enhanced Curvature The mean curvature ;

[0057] in, To enhance the curvature The number of .

[0058] In order to make data of different scales comparable and facilitate subsequent edge extraction, the curvature mean Perform minimum-maximum normalization to make the curvature mean Go to the same interval, the same interval is [0, 1];

[0059] in, and The enhanced curvature The minimum and maximum values ​​of the normalized curvature value It will be between 0 and 1. 0 corresponds to the minimum value of the original curvature, 1 corresponds to the maximum value of the original curvature, and the edge feature point appears as a point with a larger curvature.

[0060] The curvature value The preset first screening threshold Compare and retain those greater than the first screening threshold The curvature value , extract the preserved curvature value The corresponding point is the first edge point; and less than the first screening threshold The curvature value The corresponding points are regarded as non-edge points and discarded.

[0061] After the 3D point cloud data is projected onto the XOZ plane, each edge point obtained forms a 2D edge point set. The edge points in the effective edge area are matched with the projection points of the 2D edge point set, such as Figure 9~Figure 11 As shown; Calculate the two-dimensional Euclidean distance between two points that can match ; 2D Euclidean distance The calculation process is:

[0062] in, It is an edge point With the projection point The Euclidean distance between are the coordinates of the edge points, are the coordinates of the projection point.

[0063] The two-dimensional Euclidean distance is compared with the preset second screening threshold In contrast, keep the two-dimensional Euclidean distance less than the second screening threshold The edge points in the effective edge area are obtained by obtaining the second edge points; that is, a second screening threshold is set , represents the maximum distance between the edge point in the effective edge area and the projection point of the two-dimensional edge point set. If the two-dimensional Euclidean distance Less than the second screening threshold , then it is considered that the edge point in the effective edge area matches the projection point of the two-dimensional edge point set successfully, and the second edge point is obtained; The first edge point and the second edge point are collected to construct an edge point set, and the segmentation result is obtained, such as Fig.11 and Fig.13 shown.

[0064] The present invention proposes a point cloud edge segmentation system suitable for complex industrial parts, which is used to implement the above method, including: The first processing unit is configured as follows: Used to extract the estimated normal vector of the point cloud neighborhood point of each point in the point cloud data after voxel grid downsampling processing, perform analysis and processing based on the estimated normal vector of the point cloud neighborhood point to obtain enhanced curvature, and extract edge feature points based on the enhanced curvature; The second processing unit is configured as follows: It is used to extract the edge area in the two-dimensional view after two-dimensional processing of point cloud data using the Canny edge detection algorithm, and obtain the effective edge area by filtering using connected components; The third processing unit is configured as follows: Used to filter the edge feature points and the effective edge area based on a preset threshold to obtain a segmentation result; Output unit, configured as: Used to output segmentation results.

[0065] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention. Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive. The scope of the present invention is defined by the attached claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

[0066] In addition, it should be understood that although this specification is described in accordance with the implementation modes, not every implementation mode contains only one independent technical solution. This description of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation modes that can be understood by those skilled in the art. The above content is only to illustrate the technical idea of ​​the present invention, and cannot be used to limit the protection scope of the present invention. Any changes made on the basis of the technical solution according to the technical idea proposed by the present invention shall fall within the protection scope of the claims of the present invention.

Claims

1. A point cloud edge segmentation method suitable for complex industrial parts, characterized in that: The following steps are involved: Extract the estimated normal vector of the point cloud neighborhood point of each point in the point cloud data after the voxel grid downsampling processing, and perform analysis and processing based on the estimated normal vector of the point cloud neighborhood point to obtain enhanced curvature; The Canny edge detection algorithm is used to extract the edge area in the two-dimensional view after the point cloud data is processed in two dimensions, and the connected components are used to filter and obtain the effective edge area; The enhanced curvature and the effective edge area are screened based on a preset threshold to obtain a segmentation result.

2. A point cloud edge segmentation method suitable for complex industrial parts according to claim 1, characterized in that: The extracting of the point cloud data after the voxel grid downsampling process, and estimating the normal vector of the point cloud neighborhood point of each point, comprises: Collect 3D point cloud data of complex workpieces; Divide the 3D space where the 3D point cloud data is located into multiple voxels , count each voxel All points within , select each voxel The centroid point in , using the centroid point Replace each voxel All points within , the centroid point Retain in the downsampled point cloud to obtain a downsampled point cloud dataset; Extract the downsampled point cloud data set, each centroid point Estimated normal vector of the point cloud neighborhood point.

3. A point cloud edge segmentation method suitable for complex industrial parts according to claim 2, characterized in that: The selection of each voxel The centroid point in The process is: Where n is the voxel The number of points in Voxel The point in .

4. The point cloud edge segmentation method applicable to complex industrial parts according to claim 2, characterized in that: The extracted downsampled point cloud data set has each centroid point The estimated normal vector of the point cloud neighborhood point includes: Based on the centroid point , determine the centroid point The neighborhood point set ; For the neighborhood point set Perform centralization and calculate the covariance matrix ; For the covariance matrix Perform eigenvalue decomposition to obtain eigenvalues ​​and eigenvectors corresponding to the eigenvalues, that is, obtain estimated normal vectors of point cloud neighborhood points; The enhanced curvature is obtained by performing analysis based on the eigenvalue and the estimated normal vector of the point cloud neighborhood point.

5. The point cloud edge segmentation method applicable to complex industrial parts according to claim 4, characterized in that: The covariance matrix The calculation process is: in, Neighborhood point set The number of midpoints, Neighborhood point set The mean of the midpoints, is the centroid value.

6. The point cloud edge segmentation method applicable to complex industrial parts according to claim 1, characterized in that: The step of analyzing and processing based on the eigenvalue and the estimated normal vector of the point cloud neighborhood point to obtain the enhanced curvature includes: Obtaining a neighborhood point normal vector of a first neighborhood point corresponding to a point whose estimated normal vector is a neighborhood point of the point cloud; Calculate the difference angle between the estimated normal vector of the point cloud neighborhood point and the normal vector of the domain point ; Calculate the curvature based on the eigenvalue; Based on the difference angle and the curvature, and the enhanced curvature is calculated.

7. The point cloud edge segmentation method applicable to complex industrial parts according to claim 1, characterized in that: The method of using the Canny edge detection algorithm to extract edge regions in a two-dimensional view after two-dimensional processing of the point cloud data includes: Based on the set resolution and the size of the plane, construct the projection plane; Projecting the point cloud data onto the projection plane to obtain a two-dimensional image; Performing Gaussian filtering on the two-dimensional image to obtain a smoothed image; The gradient amplitude of the smoothed image is calculated using the Sobel operator and the gradient direction , and the gradient amplitude along the gradient direction Perform non-maximum suppression and retain local maxima; Based on the set threshold, the suppressed gradient amplitude is Detect and get edge image ; The edge image is constructed by using connected components Screening is performed to obtain the effective edge area.

8. The point cloud edge segmentation method applicable to complex industrial parts according to claim 7, characterized in that: The edge image is processed by using connected components Screening is performed to obtain effective edge areas, including: Using connected components to map edge images Perform analysis and screen the complete edge area; Get the perimeter of the connected area in the entire edge region , based on a preset perimeter threshold The perimeter of the connected region Compare and retain the perimeter of the connected area Greater than the perimeter threshold The edge area is obtained by calculating the effective edge area.

9. The point cloud edge segmentation method applicable to complex industrial parts according to claim 1, characterized in that: The step of screening the enhanced curvature and the effective edge area based on a preset threshold to obtain a segmentation result includes: Obtain the second area point corresponding to the point of enhanced curvature, and obtain the neighborhood point set by statistics , calculate the neighborhood point set Mean curvature of the medium enhanced curvature ; For the mean curvature Normalization is performed so that the curvature mean Go to the same interval; The transformed curvature mean Compare with the preset first screening threshold respectively, and retain the converted curvature mean value greater than the first screening threshold , extract the retained curvature mean The corresponding point is the first edge point; Matching edge points in the effective edge area with projection points of the edge area in the corresponding two-dimensional view; Calculate the two-dimensional Euclidean distance between two points that can be matched; Comparing the two-dimensional Euclidean distance with a preset second screening threshold, retaining edge points in the valid edge region whose two-dimensional Euclidean distance is less than the second screening threshold, and obtaining second edge points; The first edge points and the second edge points are collected to construct an edge point set to obtain the segmentation result.

10. A point cloud edge segmentation system suitable for complex industrial parts, used to implement the method described in any one of claims 1 to 9, characterized in that: include: The first processing unit is configured as follows: Used to extract the estimated normal vector of the point cloud neighborhood point of each point in the point cloud data after voxel grid downsampling processing, perform analysis and processing based on the estimated normal vector of the point cloud neighborhood point to obtain enhanced curvature, and extract edge feature points based on the enhanced curvature; The second processing unit is configured as follows: It is used to extract the edge area in the two-dimensional view after two-dimensional processing of point cloud data using the Canny edge detection algorithm, and obtain the effective edge area by filtering using connected components; The third processing unit is configured as follows: Used to filter the edge feature points and the effective edge area based on a preset threshold to obtain a segmentation result; Output unit, configured as: Used to output segmentation results.

Images