Three-dimensional reconstruction method of sheet metal part

Abstract

The invention discloses a three-dimensional reconstruction method for sheet metal parts, and relates to the technical field of three-dimensional model construction. In order to solve the problems that in the prior art, an existing digital detection system often depends on a high-precision scanner and expensive detection equipment and is difficult to adapt to the requirements of small and medium-sized enterprises on cost and efficiency, the equipment is complex, the measurement period is long, and the detection precision is limited, the invention provides a digital detection system. According to the technical scheme provided by the invention, the three-dimensional reconstruction method for the sheet metal part comprises the following steps: acquiring point cloud data of the sheet metal part; a step of smoothing the point cloud data; and performing three-dimensional reconstruction according to the smoothed point cloud data. And performing three-dimensional reconstruction through a greedy projection triangulation reconstruction method and a Poisson reconstruction method. The greedy projection triangulation reconstruction method comprises the following steps: carrying out two-dimensional projection on point cloud data, and then carrying out triangulation on projection points. The Poisson reconstruction method comprises the step of converting the point cloud data into an implicit function. The method is suitable for being applied to three-dimensional reconstruction work of sheet metal parts.

Description

Technical Field

[0001] The invention relates to the technical field of three-dimensional model construction, and in particular to three-dimensional reconstruction of sheet metal parts. Background Art

[0002] In the field of sheet metal parts manufacturing and processing, 3D reconstruction and precise detection are important links to ensure product quality. At present, 3D reconstruction and digital detection technology of sheet metal parts have been widely used in various manufacturing industries, but still face many challenges.

[0003] In the existing technology, the acquisition and processing of point cloud data is one of the key links in 3D reconstruction. Laser scanning, structured light scanning and other technical means are usually used to obtain 3D point cloud data of sheet metal parts. However, due to the complex surface of sheet metal parts, there are often problems such as strong reflection on smooth surfaces, and difficulty in scanning sharp corners and complex shapes. This leads to noise and missing in point cloud data, which in turn affects the accuracy and reconstruction effect of subsequent 3D reconstruction.

[0004] For example, traditional 3D reconstruction methods include the greedy projection algorithm based on triangulation and the Poisson reconstruction algorithm. The greedy projection triangulation algorithm can efficiently triangulate the point cloud, but it is prone to errors when facing sparse point cloud data, and is sensitive to noise, and is prone to overfitting or underfitting. The Poisson reconstruction method generates a smooth 3D surface by constructing an implicit function. Although it can ensure the overall consistency of the model, it often cannot reconstruct local details well when dealing with complex shapes and multiple interconnected parts, and the generated surface model may not be accurate enough.

[0005] In addition, existing digital inspection systems often rely on high-precision scanners and expensive inspection equipment, which are difficult to adapt to the cost and efficiency requirements of small and medium-sized enterprises. The inspection process in the existing technology has problems such as complex equipment, long measurement cycle, and limited inspection accuracy. Especially when measuring sheet metal parts with complex geometric features, it is difficult to ensure sufficient accuracy. Summary of the invention

[0006] In order to solve the problems existing in the prior art, the existing digital detection system often relies on high-precision scanners and expensive detection equipment, which is difficult to adapt to the cost and efficiency requirements of small and medium-sized enterprises, and has the problems of complex equipment, long measurement cycle and limited detection accuracy, the technical solution provided by the present invention is:

[0007] A three-dimensional reconstruction method for a sheet metal part, the method comprising:

[0008] Steps to collect point cloud data of sheet metal parts;

[0009] A step of preprocessing the point cloud data;

[0010] A step of smoothing the pre-processed point cloud data;

[0011] A step of performing three-dimensional reconstruction based on the point cloud data after smoothing.

[0012] Furthermore, a preferred embodiment is provided, wherein the preprocessing includes denoising and registration.

[0013] Furthermore, a preferred embodiment is provided, wherein the preprocessing further comprises using a KD tree algorithm to perform a neighborhood search on each point in the point cloud data to find a preset number of neighboring points.

[0014] Furthermore, a preferred embodiment is provided for performing three-dimensional reconstruction by using a greedy projection triangulation reconstruction method and a Poisson reconstruction method.

[0015] Furthermore, a preferred implementation is provided, in which the greedy projection triangulation reconstruction method is: two-dimensionally projecting the point cloud data and then triangulating the projection points.

[0016] Furthermore, a preferred implementation is provided, where the Poisson reconstruction method is: converting the point cloud data into an implicit function.

[0017] Based on the same inventive concept, the present invention also provides a three-dimensional reconstruction device for sheet metal parts, the device comprising:

[0018] Module for collecting point cloud data of sheet metal parts;

[0019] A module for preprocessing the point cloud data;

[0020] A module for smoothing the pre-processed point cloud data;

[0021] A module for performing three-dimensional reconstruction based on the point cloud data after smoothing.

[0022] Based on the same inventive concept, the present invention also provides a computer storage medium for storing a computer program, and when the computer reads the computer program, the computer executes the described method.

[0023] Based on the same inventive concept, the present invention also provides a computer, including a processor and a storage medium. When the processor reads a computer program stored in the storage medium, the computer executes the method described.

[0024] Based on the same inventive concept, the present invention also provides a computer program product, which is a computer program. When the computer program is executed, the method described above is implemented.

[0025] Compared with the prior art, the technical solution provided by the present invention is beneficial in that:

[0026] The improved greedy projection triangulation algorithm significantly improves the surface smoothness and detail performance of the reconstructed model by combining pixel grid filtering and MLS smoothing. This method effectively solves the problem that the greedy projection triangulation algorithm is sensitive to noise and outliers, making the reconstructed model closer to the real part. Compared with the traditional greedy projection triangulation algorithm, this improved scheme shows better smoothness and detail restoration capabilities when processing sparse point cloud data.

[0027] The Poisson reconstruction algorithm converts point cloud data into a smooth surface model by solving partial differential equations. Its advantage is that it can ensure the tightness and consistency of the overall model. In other studies, Poisson reconstruction is often used to construct more regular and continuous surface models. This paper also observed its advantages in reconstructing the overall consistency when using Poisson reconstruction, especially when the point cloud density is large, its smoothing effect is better than other methods.

[0028] The digital detection system combined with Hough transform is used for line and circle feature detection of two-dimensional images, which can accurately extract key geometric parameters. The advantage of this method lies in its high robustness and anti-interference ability. Compared with other edge detection methods, Hough transform performs well in processing images containing noise, ensuring the accurate recognition of the characteristic dimensions of sheet metal parts, thereby improving the reliability of measurement.

[0029] Compared with the traditional direct measurement method based on 3D model, the dimensional measurement through 2D projection of 3D model combined with digital detection system has higher efficiency and lower computing resource consumption. This method makes up for the low computing efficiency caused by large data volume in traditional measurement methods, and can quickly obtain reliable measurement results, especially in an environment with limited hardware resources.

[0030] Moving least squares (MLS) smoothing solves the problem of irregular point cloud data and uneven surface caused by scanning equipment limitations by fitting the local surface of the point cloud data. Compared with the point cloud data without filtering, the point cloud data after MLS smoothing shows higher smoothness and consistency in the 3D reconstruction process, especially in the presence of surface defects or holes, which significantly improves the quality of the final reconstructed model.

[0031] The improved algorithm combined with voxel grid filtering also brings significant results in the detail performance of model reconstruction. Voxel grid filtering removes small noise and interference by smoothing the details of the model surface, making the surface of the reconstructed model more uniform. Compared with traditional greedy projection triangulation, this method can maintain higher detail fidelity when processing areas with large changes in point cloud density, while reducing redundant faces or extra lines that may appear during the reconstruction process.

[0032] When comparing and analyzing the three reconstruction algorithms, the improved greedy projection triangulation combines the advantages of Poisson reconstruction and traditional greedy projection, which can ensure the global consistency of the model while retaining the detailed features, thus providing a more balanced reconstruction effect. While ensuring overall smoothness, Poisson reconstruction often cannot accurately handle more complex geometric structures, and the improved scheme makes up for this deficiency through reasonable smoothing and filtering, making it perform well in the reconstruction of complex parts.

[0033] For digital detection, a universal tool microscope was used as a comparison benchmark to verify the reliability and accuracy of the digital detection system. This method combined with experimental verification effectively improves the credibility of the measurement system, and by comparing the actual measurement value with the digital detection result, it ensures that the error is controlled within the technical indicators. Compared with the traditional single detection method, this method greatly enhances the credibility and repeatability of the experimental results.

[0034] The present invention proposes an improved 3D reconstruction and digital detection method, which improves the smoothness and accuracy of the reconstructed model by improving the greedy projection triangulation algorithm and combining it with the moving least squares method (MLS). At the same time, the reconstructed model is accurately verified in size through a digital detection system, thereby solving the problems of noise sensitivity, difficulty in reconstructing complex shapes, and low detection efficiency existing in the prior art.

[0035] Suitable for 3D reconstruction of sheet metal parts. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of local plane fitting and point cloud projection;

[0037] Figure 2 is the local projection graph of the three-dimensional k neighboring points;

[0038] Figure 3 It is the steps of Poisson reconstruction algorithm;

[0039] Figure 4 Schematic diagram of loose reconstruction, where (a) is the directed point cloud, (b) is the vector gradient, (c) is the indicator function, and (d) is the reconstructed surface;

[0040] Figure 5 Schematic diagram of the MLS projection process. DETAILED DESCRIPTION

[0041] In order to make the advantages and benefits of the technical solution provided by the present invention more clearly reflected, the technical solution provided by the present invention is now further described in detail with reference to the accompanying drawings, specifically:

[0042] Embodiment 1: This embodiment provides a three-dimensional reconstruction method for a sheet metal part, the method comprising:

[0043] Steps to collect point cloud data of sheet metal parts;

[0044] A step of preprocessing the point cloud data;

[0045] A step of smoothing the pre-processed point cloud data;

[0046] A step of performing three-dimensional reconstruction based on the point cloud data after smoothing.

[0047] Specifically, the 3D reconstruction and digital detection technology of sheet metal parts is divided into the following main steps:

[0048] 1. Acquisition and preprocessing of point cloud data.

[0049] 2. Application of 3D reconstruction algorithms, including greedy projection triangulation and Poisson reconstruction.

[0050] 3. Use the improved greedy projection triangulation algorithm for 3D reconstruction.

[0051] 4. Use the digital inspection system to measure the dimensions of the reconstructed sheet metal parts.

[0052] Embodiment 2: This embodiment further limits the three-dimensional reconstruction method of a sheet metal part provided in Embodiment 1, and the preprocessing includes denoising and registration.

[0053] Implementation method three: This implementation method further limits the three-dimensional reconstruction method of a sheet metal part provided in implementation method two. The preprocessing also includes using a KD tree algorithm to perform a neighborhood search on each point in the point cloud data to find a preset number of neighboring points.

[0054] Implementation method 4: This implementation method further limits the three-dimensional reconstruction method of a sheet metal part provided in implementation method 1, and performs three-dimensional reconstruction through greedy projection triangulation reconstruction method and Poisson reconstruction method.

[0055] Implementation method five: This implementation method further limits the three-dimensional reconstruction method of a sheet metal part provided in implementation method four. The greedy projection triangulation reconstruction method is: two-dimensionally project the point cloud data, and then triangulate the projection points.

[0056] Implementation method six: This implementation method further limits the three-dimensional reconstruction method of a sheet metal part provided in implementation method five. The Poisson reconstruction method is: converting point cloud data into an implicit function.

[0057] Specific:

[0058] First, the point cloud data of the sheet metal parts is obtained through 3D scanning. After preprocessing and registration, the point cloud data that meets the requirements is obtained. Next, a variety of 3D reconstruction algorithms are used to reconstruct the model to generate a 3D model of the sheet metal part. In order to improve the smoothness of the reconstructed model, an improved greedy projection triangulation algorithm is used for reconstruction, and the moving least squares method (MLS) is combined to smooth the point cloud. Finally, the reconstructed sheet metal model is inspected and verified using a digital inspection system to ensure that it meets the accuracy requirements.

[0059] Detailed step description

[0060] Acquisition and preprocessing of point cloud data

[0061] 1. Point cloud data collection: Use a 3D scanner to scan sheet metal parts to obtain preliminary point cloud data.

[0062] 2. Point cloud data preprocessing: denoise, reduce noise and align the collected point cloud data to ensure the consistency and integrity of the data.

[0063] 3. Neighborhood search: Use the KD tree algorithm to perform neighborhood search on each point to find a specified number of neighboring points to provide data support for subsequent plane fitting and projection.

[0064] Specifically, the registration of point cloud data is achieved through the registration method in the invention patent with application number 202411497927.0.

[0065] Application of 3D reconstruction algorithms

[0066] 1. Greedy projection triangulation reconstruction:

[0067] Perform a local planar two-dimensional projection on the point cloud data, projecting each point and its neighborhood points into the tangent plane.

[0068] The projection points are triangulated using the greedy projection triangulation algorithm to construct the topological relationship of the three-dimensional space.

[0069] The advantage of this algorithm is that it can avoid unnecessary lines and clearly display the detailed features of parts, but it is more sensitive to point cloud density and noise.

[0070] 2. Poisson reconstruction:

[0071] The point cloud data is converted into an implicit function, and a smooth three-dimensional surface model is generated by solving the Poisson equation.

[0072] This method can ensure the airtightness and smoothness of the model, but it has certain limitations when dealing with complex objects with multiple interconnected parts.

[0073] Improved 3D reconstruction algorithm

[0074] 1. Improved greedy projection triangulation algorithm: Based on the original greedy projection triangulation algorithm, MLS smoothing is combined to improve the smoothness of the reconstructed surface.

[0075] 2. Moving least squares (MLS) smoothing:

[0076] Smooth and resample point cloud data to fill in missing parts of the surface and eliminate surface overlap during reconstruction.

[0077] Each sampling point is fitted by high-order polynomial interpolation to ensure that the surface of the entire model is smooth enough.

[0078] 3. Comparison of model reconstruction effects: The three reconstruction methods are compared in terms of effect and time, and it is concluded that the improved greedy projection triangulation algorithm has obvious advantages in reconstruction quality.

[0079] Digital inspection of sheet metal parts

[0080] 1. Planar projection of the 3D model: Project the 3D reconstructed sheet metal part model onto a plane to obtain a 2D image for subsequent dimensional measurement.

[0081] 2. Dimension detection and accuracy verification: Use a digital detection system to measure the straight line and circle features of the two-dimensional projection, specifically using Hough transform for detection.

[0082] 3. Analysis of test results:

[0083] The actual dimensions of sheet metal parts are measured using a universal tool microscope and compared with the measurements of the digital inspection system.

[0084] The measurement results show that the error is within the range of ±0.3mm, which meets the accuracy requirements and verifies the reliability of the reconstruction and detection system.

[0085] Embodiment 7: This embodiment provides a three-dimensional reconstruction device for sheet metal parts, the device comprising:

[0086] Module for collecting point cloud data of sheet metal parts;

[0087] A module for preprocessing the point cloud data;

[0088] A module for smoothing the pre-processed point cloud data;

[0089] A module for performing three-dimensional reconstruction based on the point cloud data after smoothing.

[0090] Embodiment 8: This embodiment provides a computer storage medium for storing a computer program. When the computer reads the computer program, the computer executes the method provided in embodiment 1.

[0091] Embodiment 9: This embodiment provides a computer, including a processor and a storage medium. When the processor reads a computer program stored in the storage medium, the computer executes the method provided in embodiment 1.

[0092] Embodiment 10: This embodiment provides a computer program product, which is a computer program. When the computer program is executed, the method provided in embodiment 1 is implemented.

[0093] Implementation Method XI: Combination Figure 1-5 This embodiment is described in detail by using specific examples to further describe the technical solution provided above. Specifically:

[0094] This embodiment effectively solves the problem of insufficient reconstruction accuracy caused by point cloud noise and complex shapes in the prior art by using an improved greedy projection triangulation algorithm combined with moving least squares smoothing. The MLS smoothing process significantly improves the smoothness and consistency of the model surface, so that the final three-dimensional model is closer to the geometric shape of the real object. By reducing the impact of noise on the model, the improved reconstruction algorithm shows stronger robustness and detail retention when dealing with complex geometric features. In addition, the model is dimensionally verified by a digital detection system, achieving a faster and more accurate detection process. Compared with traditional expensive and complex detection equipment, the detection method adopted in this embodiment not only reduces hardware requirements and usage costs, but also greatly shortens the measurement cycle, and enhances the flexibility and ease of operation of the detection process. This makes the production process of sheet metal parts more adaptable to industrial application needs, especially in small and medium-sized enterprises, improves quality assurance capabilities, and provides higher production efficiency and economic benefits, which has significant comprehensive advantages compared with the prior art.

[0095] Greedy projection triangulation:

[0096] The greedy projection triangulation algorithm is mainly used for 3D reconstruction of point cloud data. Its core principle is to use the local optimal solution to approximate the overall optimal surface structure. The specific process is as follows:

[0097] (1) Nearest neighbor search: For any point P in the point cloud data, the KDtree nearest neighbor search method is used to determine its neighborhood containing k nearest neighbor points.

[0098] (2) Local plane two-dimensional projection. Determine the projection tangent plane of a point P and all points in its k-neighborhood, and project all points in the neighborhood onto this two-dimensional plane, which is also the normal plane of the data points. Figure 1 shown.

[0099] ① Assume that P0(x0, y0, z0) is a selected point and the normal vector of this point is n(A, B, C). If point P(x, y, z) is located on the tangent plane passing through point P0, then this tangent plane can be expressed by equation (1).

[0100] A(x-x0)+B(y-y0)+C(z-z0)=0 (1)

[0101] ② Map the points in space to the two-dimensional tangent plane π by rotation and translation. The corresponding projection transformation matrix is As shown in formula (2).

[0102]

[0103] Where: T C ——Translation transformation matrix:

[0104] R x ——The point cloud is rotated α degrees around the x-axis:

[0105] R y ——The point cloud is rotated by θ degrees around the y-axis:

[0106] By using formulas (1) and (2), we can calculate any point p(x i ,y i ,z i ) is projected onto its tangent plane Π, as shown in formula (3).

[0107]

[0108] The local projection of the three-dimensional k neighboring points is as follows Figure 2 shown.

[0109] (3) Constructing a spatial network. The greedy triangle projection algorithm is used to perform plane triangulation on the points on the projection plane to determine the topological relationship. The algorithm randomly selects a point from the point cloud data as the initial growth point, selects the local optimal point as the new growth point each time, and then maps the growth point on the two-dimensional plane back to the three-dimensional space based on the projection relationship, thereby constructing the topological relationship of the three-dimensional space.

[0110] Poisson surface reconstruction processing:

[0111] Poisson reconstruction is a process of converting three-dimensional point cloud data into an implicit function expression. This method constructs a smooth surface model from point cloud data by solving partial differential equations. The core idea of ​​the algorithm is to construct a densified function field that can characterize the topological structure and geometric information of the point cloud data set P. By using the invisible fitting method, the discrete Poisson equation is solved with the point cloud data as the boundary condition to solve the value of the surface function. Then, the function field is extracted by isosurface extraction to construct a model surface containing the complete set entity information, which more accurately reflects the extremely small differences in the surface features of the set. Poisson reconstruction algorithm such as Figure 3 shown.

[0112] If an implicit function f(x, y, z) is used to represent a surface model, the location of the point cloud data is determined by comparing the function f(x, y, z) with 0. The solution set where f(x, y, z) = 0 is the boundary of the model surface. Figure 4 shown.

[0113] The advantage of this algorithm is that it can ensure the sealing of the reconstructed surface of the model while accurately reflecting the geometric characteristics and subtle features of the model surface.

[0114] Assuming that the object to be measured is known to be M, its surface boundary can be expressed as The indicator function is χ M . Indicator function χ M is a function that is 1 inside the model and less than 0 outside the model. By solving the χ M The model surface can be determined by the value, as shown in formula (4).

[0115]

[0116] The collected point cloud data can be approximated by summing discrete points to approximate the surface integral, but due to the indicator function χ M Discontinuous, it is impossible to obtain the boundary value of the point cloud data by direct calculation, so a smoothing filter function is introduced To smooth χ M , that is, for the indicator function χ M Approximation. According to the Gaussian divergence theorem, the smoothed indicator function χ M The gradient is equal to the smoothed normal vector field As shown in formula (5).

[0117]

[0118] Where: *——convolution, that is, smoothing operation;

[0119] p – point on the model surface;

[0120] ——The normal vector of the model surface pointing inwards.

[0121] Due to the discreteness of the sample points, the smoothed normal vector field is calculated hour, Not all points on the surface are known and cannot be directly calculated Therefore, it is necessary to perform segmented calculations to segment the point cloud data and approximate the overall surface by performing approximate integral summation on all segmented areas, as shown in formula (6).

[0122]

[0123] Where: s——a ​​point in the point cloud data set S;

[0124] sp——location information of the point;

[0125] ——the normal vector of the point;

[0126] P s ——The neighborhood of this point.

[0127] Since the normal vector field is usually is not integrable, so the indicator function χ of the calculation model is M Transformed into a gradient operator, that is, finding the normal vector domain where the gradient value best approximates the sample point The indicator function of . Let χ=(χ M *F), then The relationship between the indicator function and the gradient field can be obtained, as shown in formula (7).

[0128]

[0129] By citing the Euler-Lagrange theorem to make formula (7) valid, the problem is transformed into making the Laplace operator △ equal to the normal vector field The divergence of is shown in formula (8).

[0130]

[0131] Where: ——divergence operator;

[0132] Δ——Laplacian operator.

[0133] Based on the basic principle of Poisson algorithm, the algorithm mainly transforms the problem of surface reconstruction into solving Poisson equation, that is, determining the indicator function χ M , the specific steps are as follows:

[0134] (1) Construct an octree. Using the characteristics of the octree, the implicit function is represented. For the point cloud data S, the depth D of the octree is adjusted according to the actual situation such as the density of the point cloud to ensure that every point in the point cloud data is in a child node with a depth of D.

[0135] (2) Set up the function space. For any child node o∈O, use the description function F o To solve the indicator function, as shown in formula (9).

[0136]

[0137] Where: q——a point in the point cloud data;

[0138] oc——the center position of node o;

[0139] ow – the depth of node o.

[0140] Define the basis function F(q) so that the vector domain With the node function F o Establish the connection, as shown in formula (10).

[0141]

[0142] Then, by increasing the number of convolutions of the basis function, the basis function can be gradually approached to the Gaussian filter function, as shown in formula (11).

[0143]

[0144] (3) Create a vector field. Under the condition of uniform sampling, the optimal position of the sampling points is found by trilinear interpolation. The sampling points are distributed on the eight surrounding nodes and cubic interpolation is performed to make the vector field To approximate the indicator function. As shown in formula (12).

[0145]

[0146] Where: s——neighborhood point set of the sample point;

[0147] S——point cloud dataset;

[0148] Ngb D (S)——Eight nodes with a depth of D around the sample point q;

[0149] α o,s ——weight coefficient;

[0150] ——Normal vector of point q.

[0151] (4) Solving the Poisson equation. In the process of solving the indicator function χ, its gradient is usually approximated to Both can be represented by basis functions, but due to Δχ and It is not necessarily possible to solve it through basis functions. So we can use Δχ and Minimize on the basis function, that is, project them onto the basis function to solve, as shown in formula (13).

[0152]

[0153] Where: ‖·‖ is the integral of all points in the point cloud data.

[0154] By defining the matrix U and the vector v, and assuming that there is a specific point The spatial second-order derivative Δχ of χ can be described by the matrix U. As shown in formula (14).

[0155]

[0156] This converts the solution of the indicator function χ into the calculation

[0157] (5) Extract isosurfaces. Isosurface extraction is to extract the surface corresponding to the sampling points on the surface of the object being tested. To make the generated surface as similar as possible to the input sample point set position, a threshold of the isosurface can be set to make the Poisson surface and the surface of the object to be measured highly consistent. The threshold of the isosurface is set to the mean of χ to approximate the sample point set, as shown in formula (15).

[0158]

[0159] MLS Smoothing:

[0160] When using measuring equipment to measure sheet metal parts to obtain point cloud data, irregular point cloud data is generated because the scanning light cannot completely cover the surface of the smooth object, which manifests as an uneven surface or even holes, thereby affecting the integrity of the part surface information. In the point cloud data downsampling stage, if the data has a rough surface, the three-dimensional model generated during the surface reconstruction process may not accurately reflect the fine geometric features of the object surface. Therefore, in order to obtain better reconstruction results, the point cloud data must be smoothed before performing three-dimensional reconstruction. This paper uses the Moving Lest Squares (MLS) method to perform smoothing and resampling operations on the point cloud data, fill in the missing parts of the surface through high-order polynomial interpolation, and eliminate the influence of surface overlap of the point cloud data after registration. The core of this algorithm is to map the sampling points to the local surface obtained by the MLS method through projection, such as Figure 5shown.

[0161] The core idea of ​​the moving least squares (MLS) method is to first select a certain number of neighboring points around each point p in the point cloud data and establish the relationship between the coordinates of these neighboring points and the fitting surface g as a set of constraint equations. Then, the constraint equations are solved to obtain the optimal fitting surface at point p, and the point p in the original point cloud data is replaced by the point on the surface. The above steps are then repeated until the entire point cloud data is processed to achieve smoothing of the surface. The key to this method lies in the determination of the fitting function and the weight function. The main process is as follows:

[0162] (1) Determine the fitting function.

[0163] The point cloud data is regarded as a fitting data region. On a local subdomain U of the region, the fitting function f(x) is shown in formula (16).

[0164]

[0165] In the formula: α(x)=[α1(x),α2(x),…α m (x)] T ——The coefficient to be determined is a function of x;

[0166] m——the number of terms in the basis function;

[0167] p(x)=[p1(x),p2(x),…p m (x)] T - Basis function, which is a complete polynomial of order k;

[0168] m – the number of basis function terms.

[0169] For a one-dimensional problem, the basis function is p(x)=[1,x,x 2 ,…,x m ] T For two-dimensional problems, the basis functions usually have two forms: the linear basis is p(x) = [1, x, y] T ,m=6; the quadratic basis is p(x)=[1,u,v,u 2 ,uv,v 2 ] T , m = 6. In actual situations, quadratic basis is often used. In summary, the fitting function f(x) in formula (16) can be expressed as:

[0170] f(x)=a0(x)+a1(x)u+a2(x)v+a3(x)u 2 +a4(x)v 2 +a5(x)uv (17)

[0171] In order to obtain a more accurate local fit, it is necessary to minimize f(x i ) and node value y i The weighted sum of the squares of the local fit differences of . Therefore, the discrete weighted paradigm can be described as:

[0172]

[0173] Where: n – the number of nodes in the affected area;

[0174] f(x)——fitting function;

[0175] y i =y(x i )——Node x=x i The node value of

[0176] w(xx i )——Node x=x i The weight function of .

[0177] In order to determine the coefficient a(x), equation (18) should take the minimum value and take the derivative of a(x) so that Available

[0178] a=(BWB) -1 BWy (19)

[0179] Where:

[0180] W——diagonal matrix of order n,

[0181] y=[y(x1),y(x2),…y(x n )] T .

[0182] (2) Determine the weight function.

[0183] For the moving least squares method, the determination of the weight function is crucial to the smoothing of point cloud data. A circular area is usually selected as the weight function support domain, and its radius is defined as r. The weight function w(xx i ) has the following characteristics:

[0184] ①Compact support: weight function w(xx i ) is not 0 within a specific subdomain and is 0 outside the subdomain.

[0185] ② Non-negativity: The value of the weight function is always non-negative. In the support domain, the value of the weight function is always greater than zero, which ensures that the weight function has a positive impact on the data points in the selected area.

[0186] ③ Smoothness: Weight function w(xx i) is a smooth and continuous function, which helps to ensure a smooth fitting surface during point cloud data processing and improves the fitting accuracy of local data.

[0187] ④ Monotonically decreasing: In the support domain, the weight function w(xx i ) value changes with ||xx i It decreases monotonically with the increase of ||.

[0188] The commonly used weight function is the cubic spline weight function, which is expressed as shown in formula (20).

[0189]

[0190] Where: r i =‖xx i ‖;

[0191] h i ——The support domain size of the weight function of the i-th node;

[0192] β - influence coefficient, usually ranging from 1.2 to 2.5.

[0193] The technical solution provided by the present invention is further described in detail above through several specific implementation modes in order to highlight the advantages and benefits of the technical solution provided by the present invention. However, the several specific implementation modes described above are not intended to be used as limitations on the present invention. Any reasonable modification and improvement of the present invention, combination of implementation modes and equivalent substitution within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A three-dimensional reconstruction method for sheet metal parts, characterized in that: Methods include: Steps to collect point cloud data of sheet metal parts; A step of preprocessing the point cloud data; A step of smoothing the pre-processed point cloud data; A step of performing three-dimensional reconstruction based on the point cloud data after smoothing.

2. The three-dimensional reconstruction method of a sheet metal part according to claim 1, characterized in that: The preprocessing includes denoising and registration.

3. The three-dimensional reconstruction method of a sheet metal part according to claim 2, characterized in that: The preprocessing also includes using a KD tree algorithm to perform a neighborhood search on each point in the point cloud data to find a preset number of neighboring points.

4. The three-dimensional reconstruction method of a sheet metal part according to claim 1, characterized in that: Three-dimensional reconstruction is performed through greedy projection triangulation reconstruction method and Poisson reconstruction method.

5. The three-dimensional reconstruction method of a sheet metal part according to claim 4, characterized in that: The greedy projection triangulation reconstruction method is to perform two-dimensional projection on the point cloud data and then triangulate the projection points.

6. The three-dimensional reconstruction method of a sheet metal part according to claim 5, characterized in that: The Poisson reconstruction method is to convert the point cloud data into an implicit function.

7. A three-dimensional reconstruction device for sheet metal parts, characterized in that: The device includes: Module for collecting point cloud data of sheet metal parts; A module for preprocessing the point cloud data; A module for smoothing the pre-processed point cloud data; A module for performing three-dimensional reconstruction based on the point cloud data after smoothing.

8. A computer storage medium for storing a computing program, characterized in that: When the computer reads the computer program, the computer executes the method of claim 1.

9. A computer, comprising a processor and a storage medium, characterized in that: When the processor reads the computer program stored in the storage medium, the computer executes the method of claim 1 .

10. A computer program product, being a computer program, characterized in that When the computer program is executed, the method of claim 1 is implemented.

Images