Splicing method for point cloud data of object under multiple measurement stations

Through the method of coarse registration first and then precise registration, combined with principal component analysis method and ICP algorithm, the problems of large-distance easy divergence and close-distance easy local convergence in the splicing of complex shape point cloud data such as aircraft parts are solved, achieving high accuracy The point cloud data splicing is suitable for digital measurement and reverse modeling of complex shape parts.

CN118967436APending Publication Date: 2024-11-15AVIC XIAN AIRCRAFT IND GRP CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410995002.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

In the field of reverse engineering, point cloud scanning data of complex shapes such as aircraft parts cannot obtain complete surface data in one scan due to measurement environment and equipment limitations. It requires multiple measurements of data splicing at the station, but it is difficult to achieve high quality in the existing technology. Accurate splicing, especially when splicing at large distances, is prone to divergence, while close splicing is prone to fall into the problem of local convergence.

Method used

The method of coarse registration first and then fine registration is adopted. The bounding box coordinate system is generated through principal component analysis method for coarse registration, and the characteristic parts with large curvature value of point clouds are extracted for fine registration. The ICP algorithm and quaternary method are used to further further Improve the coordinate conversion relationship and realize the precise splicing of point cloud data under two measurement stations.

Benefits of technology

It realizes accurate splicing of measured point cloud data under multiple measurement stations on the outer surface of large-sized parts, avoids local convergence problems, improves point cloud data registration efficiency and accuracy, and is suitable for digital measurement and reverse modeling of complex shape parts. .

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118967436A_ABST
    Figure CN118967436A_ABST
Patent Text Reader

Abstract

A method for splicing point cloud data of an object under multiple measurement stations mainly comprises the steps that (1) at least two measurement stations are arranged on the periphery of the object, and repeated parts in actual measurement point cloud data under the two different stations are intercepted; 2) respectively generating directional bounding boxes for the two to-be-registered point clouds of the repeated part, and realizing point cloud coarse registration by adopting a bounding box coordinate system coincidence mode; 3) extracting local internal feature parts of the point cloud to be registered based on the surface curvature distribution of the point cloud; and 4) carrying out nearest point iteration by using the local internal features of the point cloud to realize precise registration of the repeated part of the point cloud. According to the method, accurate splicing of dense and disordered actually-measured point cloud data under different measurement stations can be realized, the problem of inaccurate splicing of the actually-measured data caused by local convergence due to independent use of nearest point iteration registration can be avoided, the data calculation amount can be reduced through feature sampling, and the registration and splicing efficiency and precision of the point cloud data are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a high-precision point cloud stitching technology, in particular to a method for fast and high-precision stitching and registration of dense and disordered measured point clouds at different measurement stations. Background Art

[0002] Point cloud scanning, as the current mainstream means of acquiring physical surface data for reverse engineering, plays a very important role in the field of reverse engineering. Its fast scanning speed, high accuracy and non-contact characteristics make it widely adopted in the industry. Aircraft parts are characterized by large size and complex shape. When collecting point cloud scanning data, they are limited by the measurement environment and the measurement range of the measuring equipment. One scan cannot obtain the complete outer surface point cloud data of the measured part, and it is necessary to merge multi-station scanning data to obtain a complete surface feature point cloud. Due to the misalignment of translation and rotation of different scanning data, it is necessary to use point cloud registration technology to accurately splice each part of the point cloud to obtain a complete and unified surface point cloud of the measured part. Based on the above requirements, the present invention proposes a fast and high-precision point cloud splicing method that combines coarse registration and fine registration. By utilizing the common part of the point cloud at the two stations and accurately aligning the point cloud of the common part, high-precision stitching of the data at the two measurement stations can be achieved. This technology has broad application prospects in the fields of digital measurement and 3D reverse modeling, especially in the alignment of parts. For example, in the digital measurement and analysis of complex-shaped parts such as aircraft or automobile bodies. Summary of the invention

[0003] The present invention aims to provide a method for splicing point cloud data of an object under multiple measurement positions. The strategy of coarse registration followed by fine registration is used to solve the problems that fine registration of point clouds measured at large distances and multiple positions is prone to divergence and that close-range registration is prone to local convergence. The method also extracts feature part point clouds with large curvature values ​​for fine registration to reduce the amount of calculated data and improve the registration efficiency.

[0004] In order to achieve the above object, the present invention adopts the following technical solution:

[0005] A method for splicing point cloud data of an object at multiple measurement positions includes the following main steps:

[0006] 1. At least two measuring stations are set around the object, and the object is digitally measured at each measuring station to obtain point cloud data of each measuring station respectively;

[0007] 2. When fitting and splicing the point cloud data of two measuring stations, the point cloud data of one measuring station is named as the target point cloud, and the point cloud data of the other measuring station is named as the point cloud to be registered. The point cloud data of the same area of ​​the object in the target point cloud and the point cloud to be registered are respectively intercepted and named as overlapping target point cloud and overlapping point cloud to be registered, and the oriented bounding boxes of the overlapping target point cloud and the overlapping point cloud to be registered are generated; the bounding box solution is solved by principal component analysis, and the specific process includes: for point cloud data, the three coordinate values ​​of each point can be regarded as variables, and the principal component decomposition method (Pr incipal Component Analysis: PCA) is used to solve the main direction and basic principal element according to the direction in the data set, and the normalized eigenvector corresponding to each eigenvalue is solved as the coordinate system of each bounding box.

[0008] 3. The coarse registration of the overlapping target point cloud and the overlapping point cloud to be registered is realized by overlapping the bounding box coordinate system, so as to form the conversion relationship between the coordinates of the two measuring stations;

[0009] Specific analysis process:

[0010] The two bounding box coordinate systems are overlapped, and the oriented bounding box coordinate system of the overlapping point cloud to be registered is recorded as {O1-X1Y1Z1}, and the oriented bounding box coordinate system of the overlapping target point cloud is recorded as {O2-X2Y2Z2}. The coordinate system transformation relationship can be expressed by the unique translation vector T and the rotation matrix R, as follows:

[0011]

[0012] The translation vector T is obtained from the origin of the two coordinate systems as follows:

[0013]

[0014] The angle φ between the two coordinate systems in the X, Y, and Z directions, θ can get the rotation matrix R as follows:

[0015]

[0016] After the calculation is completed, the oriented bounding box coordinate system {O1-X1Y1Z1} of the point cloud to be registered is moved according to the translation vector T and the rotation matrix R to complete the rough registration.

[0017] 4. Based on the surface curvature distribution of the overlapping target point cloud and the overlapping point cloud to be registered, the feature point cloud parts with large curvature values ​​of the overlapping target point cloud and the overlapping point cloud to be registered are respectively extracted;

[0018] Specific analysis process:

[0019] 1) The point cloud normal vector is estimated by local surface fitting. The specific process includes: selecting a data point P from the data sets of overlapping target point cloud and overlapping point cloud to be registered respectively. i (i=1,2,…,n), its neighboring points are defined as the K points closest to P, and the least squares method is used to fit the local plane P to the neighboring points. s , plane P s Defined as:

[0020]

[0021] Perform eigenvalue decomposition on M, and the eigenvector corresponding to the minimum eigenvalue is P i Normal vector, and stipulate that the outer normal vector is positive;

[0022] 2) After obtaining the normal vector of each point, the normal section circle method is used to calculate the point cloud curvature. The specific process includes: Set the data point P i The unit normal vector is N i , distance P i The nearest K points form a neighboring point set Q = [q1, q2, ..., q k ], where q i =(x i ,y i ,z i )(i=1,2,…,k) is P i The i-th neighbor point, q i The normal vector is P i The local orthogonal coordinate system is L = {p i ,X,Y,N i}, where X and Y are orthogonal unit vectors, and p, q i The normal curvature is estimated by the osculating circle;

[0023] 3) After obtaining the curvature of each point, set the sampling threshold based on the curvature size to perform point cloud feature sampling to extract regional data with large curvature values, delete points that do not meet the threshold, and complete the point cloud extraction of local features.

[0024] 5. Based on the rough registration in step 3, the nearest point iteration is performed on the feature point cloud data with large curvature values ​​of the overlapping target point cloud and the overlapping point cloud to be registered, so as to achieve the precise registration of the overlapping target point cloud and the overlapping point cloud to be registered, and further improve the conversion relationship between the coordinates of the two measurement stations;

[0025] Specific analysis process:

[0026] 1) Use the ICP algorithm to calculate the least squares approximation transformation matrix between the point cloud to be registered and the template point cloud, and use the quaternion method to solve it. Let the overlapping target point cloud X tOverlapping point cloud to be registered X s , let the unit quaternion q R =[q0q1q2q3] T , q0, q1, q2, q3 are rotation transformation vectors, where q0 ≥ 0 and q0 2 +q1 2 +q2 2 +q3 2 =1, the unit quaternion corresponds to the rotation matrix R 3×3 (q R ), and then let the translation transformation vector q T =[q4q5q6] T , then the coordinate transformation vector of the two sets of point clouds is q=[q R |q T ] T , then this transformation can be expressed as the following minimization problem:

[0027]

[0028] Among them, x ti Overlapping target point cloud X t The data points in x si X is the overlapping point cloud to be registered s Data points from X t With X s The method of claim 3 is used to obtain a local feature point cloud P with a large curvature value in an overlapping target point cloud with a point number N. t and the local feature point cloud P with large curvature value in the overlapping cloud to be registered s , calculate P t With P s Centroid:

[0029]

[0030] The covariance matrix C is constructed as follows:

[0031]

[0032] The symmetric matrix Q constructed by C is calculated as follows:

[0033]

[0034] Where I3 is the 3×3 identity matrix, tr(C) is the trace of the covariance matrix C,

[0035] The eigenvalue and corresponding eigenvector of Q, it is easy to get the eigenvector corresponding to the maximum eigenvalue is the optimal rotation vector q R =[q0q1q2q3] T , the optimal translation vector is calculated by the following formula:

[0036] q T =μ s -R(q R )μ t

[0037] Complete the least squares approximation transformation matrix solution;

[0038] 2) The point cloud to be registered is transformed according to the obtained optimal rotation vector and optimal translation vector to achieve precise registration of overlapping target point clouds and overlapping point clouds to be registered, further improving the conversion relationship between the coordinates of the two measurement stations.

[0039] 6. Through the coordinate transformation relationship of the two measurement stations, the target point cloud and the point cloud data to be registered under two different measurement stations can be accurately spliced.

[0040] Beneficial Effects

[0041] (1) A method for stitching point cloud data from multiple measurement stations of an object, which realizes the precise stitching of measured point cloud data from multiple measurement stations on the outer surface of a large-sized part, thereby obtaining complete and unified point cloud data of the surface of the measured part;

[0042] (2) It can avoid the problem of local convergence and inaccurate splicing of measured data that may occur when using the nearest point iterative registration alone. It can also reduce the amount of data calculation through feature sampling and improve the efficiency and accuracy of point cloud data registration and splicing.

[0043] (3) The high-precision spliced ​​point cloud can be used for subsequent measured model reconstruction and pre-assembly analysis, which not only improves the assembly efficiency but also enhances the assembly accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 This is a schematic diagram of the measured point cloud of a large wing rib and its two measurement stations;

[0045] Figure 2 This is a partial schematic diagram of the point cloud obtained for the wing rib at the first measuring station;

[0046] Figure 3 This is a partial schematic diagram of the point cloud obtained for the wing rib at the second measuring station;

[0047] Figure 4 Schematic diagram of bounding box calculation and rough registration involved in step 2 and step 3 of the present invention;

[0048] Figure 5 A schematic diagram of curvature feature extraction involved in step 4 of the present invention;

[0049] Figure 6 Schematic diagram of the point cloud features extracted from the overlapping part of the first measurement station;

[0050] Figure 7 Schematic diagram of the feature extraction of the overlapping point cloud at the second measurement station;

[0051] Figure 8 Schematic diagram of the iterative precision registration of the closest point involved in step 5 of the present invention;

[0052] Fig. 9 This is a schematic diagram of the splicing of point cloud data from two measurement stations of a large wing rib;

[0053] Explanation of numbers in the figure: 1-large-size parts, 2-the first measuring station, 3-the second measuring station, 4-the corresponding measurement area of ​​the first measuring station, 5-the corresponding measurement area of ​​the second measuring station, 6-the overlapping part of the measurement areas of the two measuring stations, 7-the measured point cloud data under measuring station 1, 8-overlapping target point cloud, 9-the measured point cloud data under measuring station 2, 10-overlapping point cloud to be registered, 11-overlapping target point cloud bounding box, 12-overlapping point cloud bounding box to be registered, 13-rough registration, 14-normal vector of each point, 15 large curvature point, 16-large curvature point and its neighborhood points, 17-overlapping local feature point cloud with large curvature value in the target point cloud, 18-overlapping local feature point cloud with large curvature value in the cloud to be registered, 19-fine registration, 20-stitched point cloud. DETAILED DESCRIPTION

[0054] The following will be combined with the attached Figure 1 The present invention is further described with reference to a specific case and a specific implementation method.

[0055] like Figure 1 As shown in the figure, a large-size part 1 is divided into two stations during measurement, namely the first measuring station 2 and the second measuring station 3, wherein the first measuring station corresponds to the measuring area 4, the second measuring station corresponds to the measuring area 5, and there is an overlapping part 6 of the measuring areas of the two measuring stations.

[0056] like Figure 2 As shown, the measured point cloud data 7 at the measuring station 1 is used as the target point cloud, wherein the overlapping point cloud is defined as the overlapping target point cloud 8.

[0057] like Figure 3 As shown, the measured point cloud data 9 at the measuring station 2 is defined as the point cloud to be registered, wherein the overlapping part of the point cloud is defined as the overlapping band registration point cloud 10.

[0058] like Figure 4 As shown, a point cloud bounding box of the overlapping target point cloud 8 is constructed to form an overlapping target point cloud bounding box 11. Similarly, a point cloud bounding box of the overlapping point cloud to be registered 10 is constructed to form an overlapping point cloud bounding box 12. The overlapping point cloud to be registered 10 and the overlapping target point cloud are overlapped by overlapping the coordinate systems of the two bounding boxes, thereby completing the rough alignment 13 of the two sets of point clouds.

[0059] like Figure 5 As shown, local feature point clouds with large curvature values ​​are extracted from overlapping target point clouds and overlapping point clouds to be registered based on the curvature of the point cloud surface. Taking the overlapping target point cloud as an example, the normal vectors of each point in the point cloud are calculated one by one 14, and on this basis, the normal section circle method is used to calculate the curvature of each point, and the large curvature points that meet the curvature requirements are obtained according to the given threshold range 15. The sampling threshold is set based on the curvature size to perform point cloud feature sampling to extract large curvature points and their field points 16, and the data points that do not meet the requirements are deleted. The entire point cloud is traversed to complete the extraction of the feature part.

[0060] like Figure 6 As shown, based on Figure 5 The large curvature value local point cloud extraction process shown in the figure obtains the local feature point cloud 17 with large curvature value in the overlapping target point cloud.

[0061] like Figure 7 As shown, based on Figure 5 The large curvature value local point cloud extraction process shown in the figure obtains the local feature point cloud 18 with large curvature value in the overlapping cloud to be registered.

[0062] like Figure 8 As shown, the local feature point cloud 17 with large curvature values ​​in the overlapping target point cloud and the local feature point cloud 18 with large curvature values ​​in the overlapping cloud to be registered are precisely registered 19 based on the nearest point iteration on the basis of the rough registration 13 .

[0063] like Fig. 9 As shown, according to the coarse registration 13 and the fine registration 19, the spatial pose conversion relationship of the point cloud to be registered at the second measuring station to the target point cloud data at the first measuring station is obtained, and the target point cloud and the point cloud data to be registered at two different measuring stations are accurately spliced ​​together through the coordinate conversion relationship of the two measuring stations to form a spliced ​​point cloud 20.

Claims

1. A method for splicing point cloud data from multiple measurement positions of an object, characterized in that The method comprises the following contents: 1) at least two measuring stations are arranged around the object, and digital measurement is performed on the object at each measuring station to obtain point cloud data of each measuring station respectively; 2) when fitting and splicing the point cloud data of the two measuring stations, the point cloud data of one measuring station is named as the target point cloud, and the point cloud data of the other measuring station is named as the point cloud to be registered, and the point cloud data of the same area of ​​the object in the target point cloud and the point cloud to be registered are respectively intercepted and named as the overlapping target point cloud and the overlapping point cloud to be registered, and the oriented bounding boxes of the overlapping target point cloud and the overlapping point cloud to be registered are generated; 3) The overlapping target point cloud and the overlapping point cloud to be registered are roughly registered by overlapping the bounding box coordinate system, and the conversion relationship between the coordinates of the two measuring stations is formed; 4) The feature point cloud parts with large curvature values ​​of the overlapping target point cloud and the overlapping point cloud to be registered are respectively extracted based on the surface curvature distribution of the overlapping target point cloud and the overlapping point cloud to be registered; 5) Based on the rough alignment in step 3), the nearest point iteration is performed on the feature part point cloud data with large curvature values ​​of the overlapping target point cloud and the overlapping point cloud to be aligned, so as to achieve precise alignment of the overlapping target point cloud and the overlapping point cloud to be aligned, and further improve the conversion relationship of the coordinates of the two measurement stations; 6) The target point cloud and the point cloud data to be aligned under two different measurement stations are accurately spliced ​​through the conversion relationship of the coordinates of the two measurement stations.

2. The method for splicing point cloud data of an object at multiple measurement positions according to claim 1, characterized in that: In step 2), when generating oriented bounding boxes of overlapping target point clouds and overlapping point clouds to be registered, oriented bounding boxes are established for the overlapping target point clouds and the overlapping point clouds to be registered respectively, and the bounding box solution is solved using the principal component analysis method. The specific process includes: for point cloud data, the three coordinate values ​​of each point can be regarded as variables, and PCA decomposition is performed according to the direction in the data set to solve the main direction and the basic principal component, and the normalized eigenvector corresponding to each eigenvalue is solved as the coordinate system of each bounding box.

3. The method for splicing point cloud data of an object at multiple measurement positions according to claim 1, characterized in that: In step 3), when the conversion relationship between the coordinates of the two measurement stations is formed, the two bounding box coordinate systems are overlapped, and the oriented bounding box coordinate system of the overlapping point cloud to be registered is recorded as {O1-X1Y1Z1}, and the oriented bounding box coordinate system of the overlapping target point cloud is recorded as {O2-X2Y2Z2}. The coordinate system conversion relationship can be expressed by a unique translation vector T and a rotation matrix R, as follows: The translation vector T is obtained from the origin of the two coordinate systems as follows: The angle φ between the two coordinate systems in the X, Y, and Z directions, θ can get the rotation matrix R as follows: After the calculation is completed, the oriented bounding box coordinate system {O1-X1Y1Z1} of the point cloud to be registered is moved according to the translation vector T and the rotation matrix R to complete the rough registration.

4. The method for splicing point cloud data of an object at multiple measurement positions according to claim 1, characterized in that: Step 4) extracting feature point cloud parts with large curvature values ​​of the overlapping target point cloud and the overlapping point cloud to be registered based on the surface curvature distribution of the overlapping target point cloud and the overlapping point cloud to be registered, respectively, including the following steps: 1) The point cloud normal vector is estimated by local surface fitting. The specific process includes: selecting a data point P from the data sets of overlapping target point cloud and overlapping point cloud to be registered respectively. i (i=1,2,…,n), its neighboring points are defined as the K points closest to P, and the least squares method is used to fit the local plane P to the neighboring points. s , plane P s Defined as: Perform eigenvalue decomposition on M, and the eigenvector corresponding to the minimum eigenvalue is P i Normal vector, and stipulate that the outer normal vector is positive; 2) After obtaining the normal vector of each point, the normal section circle method is used to calculate the point cloud curvature. The specific process includes: Set the data point P i The unit normal vector is N i , distance P i The nearest K points form a neighboring point set Q = [q1, q2, ..., q k ], where q i =(x i ,y i ,z i )(i=1,2,...,k) is P i The i-th neighbor point, q i The normal vector is M qi =(n x ,n y ,n z ), P i The local orthogonal coordinate system is L = {p i ,X,Y,N i }, where X and Y are orthogonal unit vectors, and p, q i The normal curvature is estimated by the osculating circle; 3) After obtaining the curvature of each point, set the sampling threshold based on the curvature size to perform point cloud feature sampling to extract regional data with large curvature values, delete points that do not meet the threshold, and complete the point cloud extraction of local features.

5. The method for splicing point cloud data of an object at multiple measurement positions according to claim 1, characterized in that: In step 5), the local feature point cloud data with large curvature value is used to perform iterative precision registration of the nearest point, including the following steps: 1) Use the ICP algorithm to calculate the least squares approximation transformation matrix between the point cloud to be registered and the template point cloud, and use the quaternion method to solve it. Let the overlapping target point cloud X t Overlapping point cloud to be registered X s , let the unit quaternion q R =[q0q1q2q3] T , q0, q1, q2, q3 are rotation transformation vectors, where q0 ≥ 0 and q0 2 +q1 2 +q2 2 +q3 2 =1, the unit quaternion corresponds to the rotation matrix R 3×3 (q R ), and then let the translation transformation vector q T =[q4q5q6] T , then the coordinate transformation vector of the two sets of point clouds is q=[q R |q T ] T , then this transformation can be expressed as the following minimization problem: Among them, x ti Overlapping target point cloud X t The data points in x si X is the overlapping point cloud to be registered s Data points from X t With X s The method of claim 3 is used to obtain a local feature point cloud P with a large curvature value in an overlapping target point cloud with a point number N. t and the local feature point cloud P with large curvature value in the overlapping cloud to be registered s , calculate P t With P s Centroid: The covariance matrix C is constructed as follows: The symmetric matrix Q constructed by C is calculated as follows: Where I3 is the 3×3 identity matrix, tr(C) is the trace of the covariance matrix C, The eigenvalue and corresponding eigenvector of Q, it is easy to get the eigenvector corresponding to the maximum eigenvalue is the optimal rotation vector q R =[q0q1q2q3] T , the optimal translation vector is calculated by the following formula: q T =μ s -R(q R )m t Complete the least squares approximation transformation matrix solution; 2) The point cloud to be registered is transformed according to the obtained optimal rotation vector and optimal translation vector to achieve precise registration of overlapping target point clouds and overlapping point clouds to be registered, further improving the conversion relationship between the coordinates of the two measurement stations.

Citation Information

Cited By

  • Over-range spherical point cloud splicing method and system based on symmetry axis

    CN119784588A