A boundary distortion point cloud optimization extraction method for curved surface parts

Abstract

A kind of boundary distortion point cloud optimization extraction method for curved surface parts, comprising the following steps: 1) first, the scanning point cloud of the curved surface part containing boundary features is established grid structure, and the point cloud is divided into boundary feature area and non-boundary feature area according to the curvature of grid center and is simplified;2) normal statistics is carried out on the simplified boundary feature points, the normal is smoothed using weighted average and a plane is fitted, and the points with too large distance to the fitted plane are position optimized, so as to complete the smoothing of boundary features;3) the optimized point cloud patch is fitted into a surface using a parameterization method, and the reconstructed surface is intersected and cut to obtain a measured model with clear boundary features.The present application improves the accuracy of the boundary features of the measured model according to the characteristics of the boundary distortion of the scanning point cloud of the curved surface part.

Description

Technical Field

[0001] The present invention relates to the field of measured model construction, and more specifically to a method for accurately constructing a measured model with clear sharp boundaries by optimizing the boundary point cloud of a curved surface part with clear sharp boundaries (without chamfers) to solve the boundary distortion problem. Background Art

[0002] 3D scanning technology is widely used to obtain surface geometry information for curved parts with well-defined sharp edges (no chamfers). 3D scanning equipment can quickly generate point cloud data of the target object, thereby constructing an accurate 3D model. However, existing scanning technologies and point cloud processing methods often face challenges when processing structures with well-defined sharp edges (no chamfers).

[0003] First, due to the operating principles and measurement errors of 3D scanning equipment, the scanned point cloud data often exhibits distortion in areas with sharp boundaries. This distortion can manifest as blurred boundaries, unclear edges, or the generation of additional noise points, thus affecting the accuracy and quality of the final model. Second, while traditional point cloud processing algorithms, such as smoothing filters and surface fitting methods, can improve the overall smoothness of the point cloud, they often result in the loss or deformation of boundary features when processing sharp boundaries, making the reconstructed model unable to accurately reflect the details of the original structure.

[0004] Furthermore, maintaining sharp edges is crucial for applications requiring high-precision models, such as industrial design, reverse engineering, and cultural heritage preservation. Therefore, effectively optimizing and restoring sharp edges in scanned point clouds while maintaining overall model quality has become a pressing technical challenge.

[0005] In summary, existing technologies have obvious shortcomings in processing structures with clear sharp boundaries. A new method is urgently needed to optimize the boundary point cloud to obtain a measured model with clear sharp boundaries. Summary of the Invention

[0006] The purpose of the present invention is to solve the problem of boundary distortion of scanning point clouds of parts with clear and sharp boundaries, and to invent a boundary distortion point cloud optimization extraction method for surface parts. It optimizes the boundary point cloud to construct a high-precision measured model with clear and sharp boundaries.

[0007] The technical solution of the present invention is:

[0008] A method for optimizing and extracting boundary distortion point clouds for curved surface parts is characterized in that the method for constructing a measured model of a complex structure based on boundary optimization comprises the following steps:

[0009] S1: First, a grid structure is established for the scanned point cloud. The point cloud is divided into regions with boundary features and regions without boundary features according to the grid center curvature and then simplified.

[0010] S2: Perform normal statistics on the simplified boundary feature points, use weighted average to smooth the normals and fit the plane, and optimize the positions of points that are too far from the fitting plane to complete the smoothing of the boundary features;

[0011] S3: Use parametric methods to perform surface fitting on the optimized point cloud patches, make topological consistency judgment on the reconstructed surface, and crop it to obtain a measured model with clear boundary features.

[0012] In step S1, a spatial grid is established for the scanned point cloud, and the point cloud is divided into regions with boundary features and regions without boundary features according to the center curvature of the grid. The following steps are simplified:

[0013] S1.1: Create a spatial grid and set the boundary value of the point cloud data x, y, and z to x max 、x min 、y max 、y min 、z max 、z min , determine the grid size d according to the resolution of the scanning device, using (x min -d / 2,y min -d / 2,z min -d / 2) as the grid initial point, and extend the grid to the maximum point (x max ,y max , z max ) including the termination, dividing the point cloud into N x ×N y ×N z In the grid

[0014]

[0015] Among them, ┌﹒ ┐ is rounded up. Use n x ∈(0, Nx-1), n x ∈(0, Nx-1), n x ∈(0, Nx-1), p∈{0,1} is used to number each grid, and the grid coordinates (n x , n y , n z , p) indicates the starting point is (n x d, n y d, n z d) A grid with a side length of d. Place the point cloud data into the corresponding grid. If the point exists in the grid, p = 1, otherwise p = 0.

[0016] S1.2: Use the quadratic surface fitting method to calculate the average curvature of the grid structure center point and set the curvature threshold to determine whether further subdivision is needed. The specific subdivision method is as follows:

[0017] ① Calculate the average curvature of all point clouds in the grid and determine the threshold H based on the curvature distribution t ;

[0018] ② The mean curvature H i >H t The grid is further subdivided, and the length of the grid structure after subdivision is d / 2;

[0019] ③ Re-divide the point cloud in the original grid into the subdivided grid according to the spatial distribution.

[0020] S1.3: In step S1.2, there are grids of length d and d / 2. The point cloud in the grid of length d does not contain boundary regions, while the point cloud in the spatial grid of length d / 2 contains boundary regions. Different simplification strategies are used for the two grid point clouds:

[0021] ① For points in a large grid, the grid center is used as the simplified point cloud;

[0022] ② The points in the small grid use a curvature-based streamlined method to calculate the average curvature of the grid center point in the neighborhood, and sample the grid at a rate ρ based on the curvature. i Random sampling of j =H j / H max .

[0023] In step S2, normal statistics are performed on the simplified boundary feature points, the normals are smoothed using weighted average and a plane is fitted, and the positions of points that are too far from the fitting plane are optimized, thereby completing the smoothing of the boundary features. The steps include:

[0024] S2.1: Use the gradient method to detect clear boundary lines from the point cloud containing the boundary area. The steps are as follows:

[0025] ①Define the neighborhood radius: Use a fixed radius spherical neighborhood and define the neighborhood of point p as the set of all points whose distance to p is less than or equal to a fixed radius r.

[0026] N(p)={q|||pq||≤r} (2)

[0027] ②Calculate the gradient size G(p) of point p

[0028]

[0029] Where ω(q) is the weight function, n p、n q are the normal vectors of point p and point q respectively.

[0030] ③Set threshold G T , the gradient is greater than G T The points are retained in set C

[0031] C={pi|G p >G T} (4)

[0032] ④ Extract points greater than the gradient threshold, use non-maximum suppression, retain the points with the largest gradient in the neighborhood, and connect them into boundary feature lines

[0033] G(p)=max{G(q)|q∈N(p)∩C} (5)

[0034] S2.2: Use the normal-guided method to optimize the boundary and nearby point clouds. The optimization process is divided into two steps:

[0035] ① Use the weighted average normal to smooth the normal, for the vertex p in the boundary area i , calculate the weighted average of the normals of other points in the neighborhood to get the smooth normal n i ˊ

[0036]

[0037] Vertex p i , select the angle consistency threshold θ t and the distance consistency threshold r t , check whether the smooth normal of the neighboring point is consistent with its own smooth normal n i The angle between ˊ is less than θ t And the distance is less than r t :

[0038]

[0039] ② For the point cloud data after normal smoothing, use An i ˊ+b represents the fitted plane, and the fitted plane P is obtained by solving the least squares problem. m

[0040] n′ T x+d m =0 (8)

[0041] where d m For the offset, calculate the point to plane P m Distance D im , for the value exceeding the threshold d t Point, translate it along the smoothed normal to the surface

[0042]

[0043] The step S3 uses a parameterization method to perform surface fitting on the optimized point cloud patches, and the obtained reconstructed surface is intersected and cropped to obtain a measured model with clear boundary features, including the following steps:

[0044] S3.1: After obtaining the optimized boundary, accurate surface features can be obtained by segmentation according to the boundary, and a parameterization fitting method is adopted for the surface.

[0045] S3.2: The surface features obtained by fitting and reconstructing from the point cloud are only single surface features, and there are intersections, which need to be post-processed to obtain the measured model. The post-processing mainly includes intersection processing according to the topological relationship of the surface features, and the obtained intersection line is the boundary feature of the measured model. It is also necessary to make reservation and removal judgment according to the direction of the angle line segmentation face piece, and finally to ensure that the generated measured model is a complete geometric model, the steps are as follows:

[0046] ①Firstly, the topological relationship of two intersecting reconstructed surfaces S1 and S2 is judged and intersected, and the obtained intersection line L1 is the boundary line of the measured model;

[0047] ②According to the boundary line L1, S1 is divided into two parts after intersection, and S2 is divided into two parts after intersection. The reference points of S1 and S2 are projected to the nearest reconstructed surface, the projection area is reserved, and the other area is deleted, and the surface pruning work is completed.

[0048] ③Finally, the integrity of the measured model is judged, and each reconstructed uncropped face piece is judged. If the boundary feature of the cropped face piece is closed, it means that the measured model is a complete geometric model.

[0049] The beneficial effects of the present application are:

[0050] 1. Improve model accuracy: By optimizing the boundary point cloud, the boundary accuracy of the measured model can be significantly improved, ensuring the clarity and accuracy of sharp edges, thereby providing higher quality three-dimensional models.

[0051] 2. Enhance detail retention: The method of the present application can accurately retain the sharp boundary features of the structure while maintaining overall smoothness, avoiding the boundary blurring or deformation phenomenon caused by traditional smoothing algorithms.

[0052] 3. Improve application universality: Since the optimized model has higher accuracy and better detail retention capability, it is suitable for various high-precision application scenarios, such as industrial design, reverse engineering, cultural heritage protection, etc. BRIEF DESCRIPTION OF DRAWINGS

[0053] Figure 1 This is a flow chart of the method for optimizing the extraction of boundary distortion point clouds for curved surface parts in the present invention.

[0054] Figure 2 Schematic diagram of the extraction of boundary point clouds for parts containing boundary feature surfaces in the present invention.

[0055] Figure 3 This is a schematic diagram of the boundary optimization results of parts containing boundary feature surfaces in the present invention.

[0056] Figure 4 This is a schematic diagram of the surface fitting, cutting and splicing of parts containing boundary feature surfaces in the present invention. DETAILED DESCRIPTION

[0057] The present invention will be further described below with reference to the accompanying drawings and implementation methods.

[0058] like Figure 1-4 shown.

[0059] A boundary distortion point cloud optimization extraction method for surface parts, the process is as follows Figure 1 The specific steps are as follows:

[0060] S1: First, a grid structure is established for the scanned point cloud. The point cloud is divided into regions with boundary features and regions without boundary features according to the grid center curvature and then simplified.

[0061] S2: Perform normal statistics on the simplified boundary feature points, use weighted average to smooth the normals and fit the plane, and optimize the positions of points that are too far from the fitting plane to complete the smoothing of the boundary features;

[0062] S3: Use parametric methods to perform surface fitting on the optimized point cloud patches, make topological consistency judgment on the reconstructed surface, and crop it to obtain a measured model with clear boundary features.

[0063] The detailed description is as follows:

[0064] In step S1, a spatial grid is established for the scanned point cloud, and the point cloud is divided into regions with boundary features and regions without boundary features according to the center curvature of the grid. The following steps are simplified:

[0065] S1.1: Create a spatial grid and set the boundary value of the point cloud data x, y, and z to x max 、x min 、y max 、y min 、z max 、z min , determine the grid size d according to the resolution of the scanning device, using (xmin -d / 2,y min -d / 2,z min -d / 2) as the grid initial point, and extend the grid to the maximum point (x max ,y max , z max ) including the termination, dividing the point cloud into N x ×N y ×N z In the grid

[0066]

[0067] Among them, ┌﹒ ┐ is rounded up. Use n x ∈(0, Nx-1), n x ∈(0, Nx-1), n x ∈(0, Nx-1), p∈{0,1} is used to number each grid, and the grid coordinates (n x , n y , n z , p) indicates the starting point is (n x d, n y d, n z d) A grid with a side length of d. Place the point cloud data into the corresponding grid. If the point exists in the grid, p = 1, otherwise p = 0.

[0068] S1.2: Use the quadratic surface fitting method to calculate the average curvature of the grid structure center point and set the curvature threshold to determine whether further subdivision is needed. The specific subdivision method is as follows:

[0069] ① Calculate the average curvature of all point clouds in the grid and determine the threshold H based on the curvature distribution t ;

[0070] ② The mean curvature H i >H t The grid is further subdivided, and the length of the grid structure after subdivision is d / 2;

[0071] ③ Re-divide the point cloud in the original grid into the subdivided grid according to the spatial distribution.

[0072] The extracted boundary point cloud is divided into small voxel grids. The specific divided point cloud is as follows Figure 2 shown.

[0073] S1.3: In step S1.2, there are grids of length d and d / 2. The point cloud in the grid of length d does not contain boundary regions, while the point cloud in the spatial grid of length d / 2 contains boundary regions. Different simplification strategies are used for the two grid point clouds:

[0074] ① For points in a large grid, the grid center is used as the simplified point cloud;

[0075] ② The points in the small grid use a curvature-based streamlined method to calculate the average curvature of the grid center point in the neighborhood, and sample the grid at a rate ρ based on the curvature. i Random sampling of j =H j / H max .

[0076] In step S2, normal statistics are performed on the simplified boundary feature points, the normals are smoothed using weighted average and a plane is fitted, and the positions of points that are too far from the fitting plane are optimized, thereby completing the smoothing of the boundary features. The steps include:

[0077] S2.1: Use the gradient method to detect clear boundary lines from the point cloud containing the boundary area. The steps are as follows:

[0078] ①Define the neighborhood radius: Use a fixed radius spherical neighborhood and define the neighborhood of point p as the set of all points whose distance to p is less than or equal to a fixed radius r.

[0079] N(p)={q|||pq||≤r} (2)

[0080] ②Calculate the gradient size G(p) of point p

[0081]

[0082] Where ω(q) is the weight function, n p 、n q are the normal vectors of point p and point q respectively.

[0083] ③Set threshold G T , the gradient is greater than G T The points are retained in set C

[0084] C={pi|G p >G T} (4)

[0085] ④ Extract points greater than the gradient threshold, use non-maximum suppression, retain the points with the largest gradient in the neighborhood, and connect them into boundary feature lines

[0086] G(p)=max{G(q)|q∈N(p)∩C} (5)

[0087] S2.2: Use the normal-guided method to optimize the boundary and nearby point clouds. The optimization process is divided into two steps:

[0088] ① Use the weighted average normal to smooth the normal, for the vertex p in the boundary areai , calculate the weighted average of the normals of other points in the neighborhood to get the smooth normal n i ˊ

[0089]

[0090] Vertex p i , select the angle consistency threshold θ t and the distance consistency threshold r t , check whether the smooth normal of the neighboring point is consistent with its own smooth normal n i The angle between ˊ is less than θ t And the distance is less than r t :

[0091]

[0092] ② For the point cloud data after normal smoothing, use An i ˊ+b represents the fitted plane, and the fitted plane P is obtained by solving the least squares problem. m

[0093] n′ T x+d m =0 (8)

[0094] where d m For the offset, calculate the point to plane P m Distance D im , for the value exceeding the threshold d t Point, translate it along the smoothed normal to the surface

[0095]

[0096] The point cloud mesh obtained by boundary optimization is as follows Figure 3 shown.

[0097] In step S3, the optimized point cloud patches are subjected to surface fitting using a parametric method, and the obtained reconstructed surfaces are intersected and clipped to obtain a measured model with clear boundary features, including the following steps:

[0098] S3.1: After obtaining the optimized boundary, segmentation based on the boundary can obtain accurate surface features, and a parametric fitting method is adopted for the surface.

[0099] S3.2: The surface features obtained by point cloud fitting and reconstruction are only single surface features and may intersect, requiring post-processing to obtain the measured model. Post-processing mainly involves finding intersections based on the topological relationships of the surface features. The resulting intersections are the boundary features of the measured model. It is also necessary to determine whether to retain or remove the facets that are split diagonally based on their directions. Finally, to ensure that the generated measured model is a complete geometric model, it is checked for closure. The steps are as follows:

[0100] ① First, Figure 4 The topological relationship of the two intersecting reconstructed surfaces S1 and S2 is judged and the intersection is calculated. The obtained intersection line L1 is the edge line of the measured model.

[0101] ② According to the edge line L1, S1 is divided into two parts ① and ② after the intersection, and S2 is divided into two parts ③ and ④ after the intersection. Project the reference points of S1 and S2 onto the nearest reconstructed surface, retain the projected area, delete other areas, and complete the surface trimming.

[0102] ③Finally, the integrity of the measured model is judged by judging each reconstructed untrimmed face. If the edge features after trimming are closed, it means that the measured model is a complete geometric model.

[0103] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

[0104] The parts not involved in the present invention are the same as the existing technology or can be implemented by using the existing technology.

Claims

1. A method for optimizing and extracting boundary distortion point clouds for curved surface parts, characterized by: The steps of optimizing, extracting and reconstructing the boundary distorted point cloud of the surface part are as follows: S1: First, a grid structure is established for the scanned point cloud. The point cloud is divided into regions with boundary features and regions without boundary features according to the grid center curvature and then simplified. S2: Perform normal statistics on the simplified boundary feature points, use weighted average to smooth the normals and fit the plane, and optimize the positions of points that are too far from the fitting plane to complete the smoothing of the boundary features; S3: Use parametric methods to perform surface fitting on the optimized point cloud patches, make topological consistency judgment on the reconstructed surface, and crop it to obtain a measured model with clear boundary features; In step S1, a spatial grid is established for the scanned point cloud, and the point cloud is divided into regions with boundary features and regions without boundary features according to the center curvature of the grid. The following steps are simplified: S1.1: Create a spatial grid and set the boundary value of the point cloud data x, y, and z to x max 、x min 、y max 、y min 、z max 、z min , determine the grid size d according to the resolution of the scanning device, using (x min -d / 2,y min -d / 2,z min -d / 2) as the grid initial point, and extend the grid to the maximum point (x max ,y max , z max ) including the termination, dividing the point cloud into N x ×N y ×N z In the grid Where ┌﹒ ┐ is rounded up, N x , N y , N z Represents the number of grids in the x, y, and z directions respectively; use n x ∈[0,N x -1]、n y ∈[0,N y -1]、n z ∈[0,N z -1], u∈{0,1} is used to number each grid, and the grid coordinates (n x , n y , n z , u) indicates that the x, y, and z directions are numbered n respectively x , n y , n z The point cloud data is placed in the corresponding grid. If there is a point in the grid, u = 1, otherwise u = 0; S1.2: Use the quadratic surface fitting method to calculate the average curvature of the grid structure center point and set the curvature threshold to determine whether further subdivision is needed. The specific subdivision method is as follows: ① Calculate the average curvature of all point clouds in the grid and determine the threshold H based on the curvature distribution t ; ② The mean curvature H i >H t The grid is further subdivided, and the length of the grid structure after subdivision is d / 2; ③ Re-divide the point cloud in the original grid into the subdivided grid according to the spatial distribution; S1.3: In step S1.2, there are grids of length d and d / 2. The point cloud in the grid of length d does not contain boundary areas; the point cloud in the spatial grid of length d / 2 contains boundary areas. Different simplification strategies are used for the two grid point clouds: ① For points in a large grid, the grid center is used as the simplified point cloud; ② The points in the small grid use a curvature-based streamlined method to calculate the average curvature of the grid center point in the neighborhood, and sample the grid at a rate ρ based on the curvature. i Random sampling of i =H i / H max ; In step S2, the normal statistics of the simplified boundary feature points are performed, the normals are smoothed using weighted average and fitted to the plane, and the positions of the points that are too far from the fitting plane are optimized to complete the smoothing of the boundary features. The following steps are involved: S2.1: Use the gradient method to detect clear boundary lines from the point cloud containing the boundary area. The steps are as follows: ① Define the neighborhood radius: Use a fixed radius spherical neighborhood and define the neighborhood of point p as the set of all points whose distance to p is less than or equal to a fixed radius r: N(p)={q|||pq||≤r} (2) ②Calculate the gradient size G(p) of point p: Where ω(q) is the weight function, n p 、n q are the normal vectors of point p and point q respectively; ③Set threshold G T , the gradient is greater than G T The points are retained in set C: C={pi|G p >G T } (4) ④ Extract points greater than the gradient threshold, use non-maximum suppression, retain the points with the largest gradient in the neighborhood, and connect them into boundary feature lines: G(p)=max{G(q)|q∈N(p)∩C} (5) S2.2: Use the normal-guided method to optimize the boundary and nearby point clouds. The optimization process is divided into two steps: ① Use the weighted average method to smooth its normals. For the vertex p in the boundary area i , calculate the weighted average of the normals of other points in the neighborhood to get the smooth normal n i ˊ: Vertex p i , select the angle consistency threshold θ t and the distance consistency threshold r t , check whether the smooth normal of the neighboring point is consistent with its own smooth normal n i The angle between ˊ is less than θ t And the distance is less than r t : ② For the point cloud data after normal smoothing, use An i ˊ+b represents the fitted plane, and the fitted plane P is obtained by solving the least squares problem. m : n′ T x+d m =0 (8) where d m For the offset, calculate the point to plane P m Distance D im , for the value exceeding the threshold d t , and translate it onto the surface along the smoothed normal: In step S3, the optimized point cloud patches are subjected to surface fitting using a parametric method, and the obtained reconstructed surfaces are intersected and clipped to obtain a measured model with clear boundary features, including the following steps: S3.1: After obtaining the optimized boundary, segmentation based on the boundary can obtain accurate surface features, and a parametric fitting method is adopted for the surface; S3.2: The surface features reconstructed from point cloud fitting are only single surface features and may intersect. Post-processing is required to obtain the measured model. Post-processing mainly involves finding intersections based on the topological relationships of the surface features. The resulting intersections are the boundary features of the measured model. Diagonal facets must be retained or removed based on their orientation. Finally, to ensure that the generated measured model is a complete geometric model, a closedness check is performed. The steps are as follows: ① First, the topological relationship of the two intersecting reconstructed surfaces S1 and S2 is judged and the intersection is calculated. The obtained intersection line L1 is the edge line of the measured model; ② According to the edge line L1, S1 is divided into two parts after the intersection, and S2 is divided into two parts after the intersection. Project the reference points of S1 and S2 onto the nearest reconstructed surface, retain the projected area, delete the other areas, and complete the surface trimming; ③Finally, the integrity of the measured model is judged by judging each reconstructed untrimmed face. If the edge features after trimming are closed, it means that the measured model is a complete geometric model.

Images