Point cloud feature extraction method and device, equipment and storage medium

Through the Gaussian curvature and average curvature typing of point cloud data, the extraction error problem caused by the loss of surface details of the fitted surface is solved, and more accurate and stable point cloud feature extraction is achieved, suitable for irregular surfaces.

CN118968134APending Publication Date: 2024-11-15WUHAN UNIV
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202410986437.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

The prior art is prone to loss of details when fitting the surface of the curved surface, resulting in point cloud feature extraction errors and the inability to accurately determine the registration relationship between point cloud data.

Method used

By obtaining the Gaussian curvature and average curvature of each point in the point cloud data, it is divided into four types: peak/well, ridge/grough, saddle, and plane. Based on these curvature characteristics statistics, feature symbols are obtained and points are determined using formulas. The feature symbol of cloud data has scale immutability.

Benefits of technology

It improves the accuracy and stability of point cloud feature extraction, has a wider scope of application, especially when there are irregular surfaces, has a smaller error, a wider scope of application, and can more accurately extract point cloud features.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118968134A_ABST
    Figure CN118968134A_ABST
Patent Text Reader

Abstract

Disclosed are a point cloud feature extraction method, apparatus and device, and a storage medium, which belong to the technical field of point cloud processing, the method comprising: acquiring point cloud data of a target area, the point cloud data comprising a plurality of points; acquiring the Gaussian curvature and the average curvature of each point in the point cloud data; based on the Gaussian curvature and the average curvature of each point, dividing the plurality of points into four point cloud types of peak / trap, ridge / valley, saddle and plane; and determining a feature symbol of the point cloud data based on the number of points in different point cloud types in each trigram. According to the method, the point cloud features of the point cloud data can be accurately extracted.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the technical field of point cloud processing, and in particular to a point cloud feature extraction method, device, equipment and storage medium. Background Art

[0002] In order to realize the automated processing of point cloud products, it is usually necessary to extract point cloud features of different point cloud data, compare the similarities between the extracted features, and thus determine whether different point cloud data can be registered.

[0003] In the related art, the point cloud feature extraction method includes: obtaining geometric parameters by fitting the surface of the curved surface, and extracting the point cloud features of the point cloud data through the geometric parameters.

[0004] However, when fitting a curved surface, the fitted curved surface is prone to detail loss, resulting in errors in the point cloud features of the point cloud data finally extracted. Summary of the invention

[0005] The present disclosure provides a point cloud feature extraction method, device, equipment and storage medium, which can accurately extract the point cloud features of point cloud data. The technical solution at least includes the following solutions: In a first aspect, a point cloud feature extraction method is provided, comprising: acquiring point cloud data of a target area, the point cloud data comprising a plurality of points; acquiring the Gaussian curvature and the average curvature of each point in the point cloud data; dividing the plurality of points into four point cloud types, namely, peak / well, ridge / valley, saddle, and plane, based on the Gaussian curvature and the average curvature of each point; and determining a feature symbol of the point cloud data based on the number of points in the different point cloud types in each quadrant.

[0006] Optionally, based on the Gaussian curvature and the average curvature of each of the points, the multiple points are divided into four point cloud categories: peak / well, ridge / valley, saddle, and plane, including: obtaining the principal curvature of the first point cloud; based on the principal curvature of the first point cloud, obtaining the threshold of the Gaussian curvature and the threshold of the average curvature; based on the threshold of the Gaussian curvature and the threshold of the average curvature, as well as the Gaussian curvature and the average curvature of each of the points, determining the point cloud type of the first point cloud.

[0007] Optionally, the principal curvature of the first point cloud includes a maximum principal curvature and a minimum principal curvature, and the obtaining of the threshold of the Gaussian curvature and the threshold of the average curvature based on the principal curvature of the first point cloud includes: using the following formula to obtain the threshold of the Gaussian curvature and the threshold of the average curvature based on the principal curvature of the first point cloud:

[0008] in, is the threshold of the Gaussian curvature, is the threshold of the mean curvature, is the true value of the maximum principal curvature, is the true value of the minimum principal curvature, It is the probability that the value in the normal distribution is outside the 3sigma range.

[0009] Optionally, the target area includes 8 quadrants, and the method further includes: counting the number of points in the different point cloud types in each of the 8 quadrants, and performing normalization processing to obtain a feature array, wherein the feature array includes 32 components, and each component is used to indicate the number of points of one point cloud type in one quadrant; determining the feature symbol of the point cloud data based on the number of points in the different point cloud types of the point cloud data in each quadrant includes: using the following formula to implement the determination of the feature symbol of the point cloud data based on the number of points in the different point cloud types of the point cloud data in each quadrant:

[0010] in, is the feature symbol of the point cloud data, is the feature array, is the i-th component.

[0011] Optionally, obtaining the Gaussian curvature and the average curvature of each point in the point cloud data includes: obtaining multiple first neighborhood points of a first point cloud, where the first point cloud is any point in the point cloud data; determining the normal vectors of the multiple first neighborhood points; and determining the Gaussian curvature and the average curvature of the first point cloud based on the normal vectors of the multiple first neighborhood points.

[0012] Optionally, determining the Gaussian curvature and the mean curvature of the first point cloud based on the normal vectors of the multiple first field points includes: determining the normal curvature of each first field point based on the normal vectors of the multiple first field points; determining the principal curvature of the first point cloud based on the normal curvature of each first field point; determining the Gaussian curvature and the mean curvature of the first point cloud based on the principal curvature of the first point cloud.

[0013] Optionally, the obtaining of multiple first neighborhood points of the first point cloud includes: constructing a KD tree of the point cloud data; and obtaining multiple first neighborhood points of the first point cloud using a K-nearest neighbor search algorithm based on the KD tree of the point cloud data.

[0014] In the second aspect, a point cloud feature extraction device is also provided, including: a first acquisition module, used to acquire point cloud data of a target area, the point cloud data including a plurality of points; a second acquisition module, used to acquire the Gaussian curvature and the average curvature of each point in the point cloud data; a classification module, used to classify the plurality of points into four point cloud types, namely peak / well, ridge / valley, saddle, and plane, based on the Gaussian curvature and the average curvature of each point; and a feature extraction module, used to determine the feature symbol of the point cloud data based on the number of points in the different point cloud types in each quadrant.

[0015] Optionally, the classification module is also used to obtain the principal curvature of the first point cloud; based on the principal curvature of the first point cloud, obtain the threshold of the Gaussian curvature and the threshold of the average curvature; based on the threshold of the Gaussian curvature and the threshold of the average curvature, as well as the Gaussian curvature and the average curvature of each point, determine the point cloud type of the first point cloud.

[0016] Optionally, the principal curvature of the first point cloud includes a maximum principal curvature and a minimum principal curvature, and the classification module is further used to implement the principal curvature based on the first point cloud by adopting the following formula to obtain the threshold of the Gaussian curvature and the threshold of the average curvature:

[0017] in, is the threshold of the Gaussian curvature, is the threshold of the mean curvature, is the true value of the maximum principal curvature, is the true value of the minimum principal curvature, It is the probability that the value in the normal distribution is outside the 3sigma range.

[0018] Optionally, the target area includes 8 quadrants, and the device further includes: a statistical module, used to count the number of points in different point cloud types in each of the 8 quadrants, and perform normalization processing to obtain a feature array, wherein the feature array includes 32 components, and each component is used to indicate the number of points of one point cloud type in one quadrant; the feature extraction module is also used to implement the determination of the feature symbol of the point cloud data based on the number of points in different point cloud types in each quadrant of the point cloud data using the following formula:

[0019] in, is the feature symbol of the point cloud data, is the feature array, is the i-th component.

[0020] Optionally, the second acquisition module is also used to acquire multiple first neighborhood points of a first point cloud, where the first point cloud is any point in the point cloud data; determine the normal vectors of the multiple first neighborhood points; and determine the Gaussian curvature and mean curvature of the first point cloud based on the normal vectors of the multiple first neighborhood points.

[0021] Optionally, the second acquisition module is also used to determine the normal curvature of each first field point based on the normal vectors of the multiple first field points; determine the principal curvature of the first point cloud based on the normal curvature of each first field point; and determine the Gaussian curvature and mean curvature of the first point cloud based on the principal curvature of the first point cloud.

[0022] Optionally, the second acquisition module is further used to construct a KD tree of the point cloud data; based on the KD tree of the point cloud data, a K-nearest neighbor search algorithm is used to acquire multiple first domain points of the first point cloud.

[0023] In a third aspect, a computer device is also provided, comprising: a memory and a processor, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor, thereby executing the point cloud feature extraction method described in the above embodiment.

[0024] In a fourth aspect, a computer-readable storage medium is also provided, wherein at least one computer program is stored in the computer-readable storage medium, and the at least one computer program is loaded and executed by a processor, thereby executing the point cloud feature extraction method described in the above embodiment.

[0025] In a fifth aspect, a computer program product is provided, comprising a computer program / instruction, wherein when the computer program / instruction is executed by a processor, the method described in the first aspect is implemented.

[0026] The beneficial effects brought by the technical solution provided by the embodiments of the present disclosure include at least: In the disclosed embodiment, since the final feature symbol is based on curvature, and curvature has scale invariance, the feature symbol obtained also has a certain scale invariance. In addition, compared with classifying multiple points in point cloud data by fitting the surface to obtain geometric parameters, when Gaussian curvature and mean curvature are used to classify multiple points in point cloud data, when encountering irregular surfaces (such as cross sections), the classification results obtained are more stable, with smaller errors and a wider range of applications, thereby effectively improving the accuracy of the extracted point cloud features. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0028] Figure 1 A flow chart of a point cloud feature extraction method provided by an exemplary embodiment of the present disclosure is shown; Figure 2 A flow chart of a point cloud feature extraction method provided by another exemplary embodiment of the present disclosure is shown; Figure 3 A schematic diagram for extracting point cloud data in the target area; Figure 4 A schematic diagram of the orientation of the principal component analysis algorithm; Figure 5 It is a schematic diagram of the distribution characteristics of the curvature of multiple points in the point cloud data in space; Figure 6 A schematic structural diagram of a point cloud feature extraction device provided by an exemplary embodiment of the present disclosure is shown; Figure 7 It is a schematic diagram of the structure of a computer device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] Unless otherwise defined, the technical or scientific terms used herein shall have the usual meanings understood by persons of ordinary skill in the field to which the present disclosure belongs. The words "first", "second", "third" and similar words used in the patent application specification and claims of the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. Similarly, words such as "one" or "one" do not indicate a quantitative limitation, but rather indicate the presence of at least one. Words such as "include" or "comprise" and similar words mean that the elements or objects appearing before "include" or "comprises" cover the elements or objects listed after "include" or "comprises" and their equivalents, and do not exclude other elements or objects.

[0030] In order to make the objectives, technical solutions and advantages of the present disclosure more clear, the embodiments of the present disclosure will be further described in detail below with reference to the accompanying drawings.

[0031] In the related art, geometric parameters are obtained by fitting the surface, and then the point cloud features of the point cloud data are extracted through the geometric parameters. If the surface of a model is irregular and has many surface details, the fitted surface will often ignore these surface details, resulting in a large error in the fitted surface. Geometric parameters are extracted based on the fitted surface, which leads to the point cloud features being extracted based on the geometric parameters, and the extracted point cloud features are not accurate enough.

[0032] Figure 1 A flowchart of a point cloud feature extraction method provided by an exemplary embodiment of the present disclosure is shown. The method can be executed by a computer device. Figure 1 , the method comprising: In step 101, point cloud data of a target area is obtained.

[0033] Point cloud data includes multiple points.

[0034] In a possible implementation, the point cloud data of the target area may be acquired by laser radar scanning.

[0035] In another possible implementation, the point cloud data of the target area may be extracted from a spatial three-dimensional model of the target area, and the spatial three-dimensional model of the target area may be a triangular mesh model.

[0036] Exemplarily, the spatial three-dimensional model can be stored in the form of obj, osgb, etc. Such files support triangular mesh models.

[0037] The spatial three-dimensional model of the target area may be an existing one, or the spatial three-dimensional model of the target area may be a triangular mesh model generated by using an image of the target area collected by a drone and three-dimensional reconstruction processing software.

[0038] In step 102, the Gaussian curvature and the average curvature of each point in the point cloud data are obtained.

[0039] In step 103 , based on the Gaussian curvature and the mean curvature of each point, the multiple points are classified into four point cloud types: peak / well, ridge / valley, saddle, and plane.

[0040] In the disclosed embodiment, when the point cloud type of a certain point is a peak / well, it means that the concavity and convexity of the surface where the point is located are consistent in all directions, that is, the extreme point; when the point cloud type of a certain point is a ridge / valley, it means that the surface where the point is located is locally convex or concave, and is flat in one main direction; when the point cloud type of a certain point is a saddle, it means that the concavity and convexity of the surface where the point is located are inconsistent in all directions; when the point cloud type of a certain point is a plane, it means that the surface where the point is located is a local plane, and here, the local plane is also the plane formed by the domain points of the point.

[0041] In step 104 , a signature of the point cloud data is determined based on the number of points in the different point cloud types in each octant.

[0042] In the disclosed embodiment, since the final feature symbol is based on curvature, and curvature has scale invariance, the feature symbol obtained also has a certain scale invariance. In addition, compared with extracting point cloud features by fitting the surface to obtain geometric parameters, when Gaussian curvature and mean curvature are used to extract point cloud features of point cloud data, when encountering irregular surfaces (such as cross sections), the extracted point cloud features are more accurate, with smaller errors and a wider range of applications.

[0043] Figure 2 A flowchart of a point cloud feature extraction method provided by another exemplary embodiment of the present disclosure is shown. The method can be executed by a computer device. Figure 2 , the method comprising: In step 201, point cloud data of a target area is obtained.

[0044] The relevant contents of step 201 refer to the aforementioned step 101, and the detailed description is omitted here.

[0045] Figure 3 This is a schematic diagram for extracting point cloud data in the target area. Figure 3 As shown, in the case where there is already a spatial three-dimensional model of the target area, point cloud data of the target area can be extracted based on the spatial three-dimensional model.

[0046] In step 202, a principal component analysis algorithm is used to orient the point cloud data.

[0047] The principal component analysis (PCA) algorithm is a statistical analysis method that aims to transform linearly dependent variables in the original feature space into new linearly independent variables, namely principal components, through orthogonal transformation. These principal components can retain most of the information of the original data while reducing the dimension of the data, making data processing and analysis more efficient. PCA projects the original data into a new coordinate system by finding an orthogonal transformation matrix so that the variance of the projected data in each dimension is as large as possible. These new dimensions are principal components, which are independent of each other and can retain as much information of the original data as possible.

[0048] Optionally, step 202 is implemented by the following steps ad.

[0049] Step a, obtaining the mean value of data of multiple points in the point cloud data.

[0050] Assuming there are a total of m points in the point cloud data, step (a) can be expressed by formula (1).

[0051] (1) In formula (1), is the mean of multiple points in the point cloud data, that is, the point cloud centroid of the point cloud data. Among the multiple points in the point cloud data, The coordinates of a point.

[0052] Step b: determining the covariance matrix of the point cloud data based on the mean of the data of the multiple points.

[0053] The covariance matrix of point cloud data can be calculated using formula (2).

[0054] (2) In formula (2), is the covariance matrix of the point cloud data. The meanings of other parameters in formula (2) are the same as those in formula (1), and their detailed description is omitted here.

[0055] Step c: perform eigenvector decomposition on the covariance matrix of the point cloud data to obtain the eigenvalues ​​and eigenvectors of the point cloud data, thereby constructing a new coordinate system.

[0056] Based on formula (3), the covariance matrix of the point cloud data can be decomposed into eigenvectors to obtain the eigenvalues ​​and eigenvectors of the point cloud data.

[0057] (3) In formula (3), Represents the eigenvalue of point cloud data, The meaning of other parameters in formula (3) is the same as that in formula (2), and the detailed description is omitted here.

[0058] Based on formula (3), multiple eigenvalues ​​and multiple eigenvectors can be solved. One eigenvalue uniquely corresponds to one eigenvector. Sort the multiple eigenvalues ​​from large to small. The first three eigenvalues ​​are , and The eigenvectors corresponding to these three eigenvalues ​​can form a matrix, which can be expressed by formula (4). These three eigenvectors are also three principal components, and a new coordinate system can be constructed from these three principal components.

[0059] (4) In formula (4), represents the matrix composed of these three eigenvectors, for The corresponding eigenvector represents the X-axis in the new coordinate system; for The corresponding eigenvector represents the Y axis in the new coordinate system; for The corresponding eigenvector represents the Z axis in the new coordinate system.

[0060] Step d: convert the data of multiple points in the point cloud data into a new coordinate system to obtain the data after dimensionality reduction.

[0061] Step d can be achieved by formula (5) (5) In formula (5), is the coordinate of the i-th point among the multiple points of the point cloud data in the new coordinate system. The meanings of other parameters in formula (5) are the same as those in formula (1) and formula (4), and detailed description is omitted here.

[0062] Figure 4 For a schematic diagram of the principal component analysis algorithm orientation, see Figure 4 ,Based on the principal component analysis algorithm, three eigenvectors can be determined, and these three eigenvectors can construct a spatial coordinate system.

[0063] Through steps ad, the coordinates of multiple points in the point cloud data can be projected into the coordinate system oriented by principal component analysis.

[0064] In the embodiment of the present disclosure, since the principal component analysis algorithm is used to achieve orientation, the feature symbol finally obtained has rotation invariance and a certain anti-interference ability.

[0065] In step 203, the Gaussian curvature and the average curvature of each point in the point cloud data are obtained.

[0066] Optionally, step 203 is implemented by the following steps eg.

[0067] Step e: obtaining a plurality of first neighborhood points of the first point cloud.

[0068] The first point cloud is any point in the point cloud data.

[0069] Optionally, step e includes the following two steps: The first step is to construct a KD tree of point cloud data.

[0070] KD tree (K-Dimensional-Tree) is a tree data structure used for fast retrieval of multidimensional spatial data points. When constructing a KD tree, first select the dimension with the largest variance in the three-dimensional data set, then select the median value on this dimension to divide the data set into two subsets; at the same time, create a tree node for storage. Repeat the first step for the two divided subsets until all subsets cannot be divided any further; if a subset cannot be divided any further, save the data in the subset to the leaf node. In the embodiment of the present disclosure, the three-dimensional data set is also point cloud data.

[0071] There are many methods for implementing the KD tree in the related art, so detailed description is omitted here.

[0072] In the second step, based on the KD tree of the point cloud data, a K-nearest neighbor search algorithm is used to obtain multiple first domain points of the first point cloud.

[0073] K-Nearest Neighbors (KNN) is a machine learning algorithm. Based on the K-nearest neighbor search algorithm, the k points closest to the first point cloud among multiple points in the point cloud data can be found. These k points are the multiple first domain points of the first point cloud.

[0074] There are many methods for implementing the K-nearest neighbor search algorithm in the related art, so detailed description is omitted here.

[0075] Step f, determining normal vectors of a plurality of first domain points.

[0076] Optionally, when determining the normal vectors of the plurality of first domain points, a principal component analysis algorithm may also be used, the principle of which is similar to the aforementioned steps ad. In this case, step f may be implemented by the following formulas (6) to (10).

[0077] Assume that there are k first domain points in total, and formula (6) is used to calculate the mean of the k first domain points.

[0078] (6) In formula (6), is the mean of the k first domain points, that is, the point cloud centroid of the k first domain points, Among the k first domain points, The coordinates of the points, Represents a set of points consisting of k first-region points.

[0079] Formula (7) is used to calculate the covariance matrix of k first domain points.

[0080] (7) In formula (7), is the covariance matrix of k first domain points. The meanings of other parameters in formula (7) are the same as those in formula (6), and detailed description is omitted here.

[0081] Based on formula (8), the covariance matrix of the k first domain points can be decomposed by eigenvectors, thereby solving the eigenvalues ​​and eigenvectors of the k first domain points.

[0082] (8) In formula (8), represents the eigenvalues ​​of k first domain points, represents the feature vector of k first domain points. The meanings of other parameters in formula (8) are the same as those in formula (7), and detailed description is omitted here.

[0083] Based on formula (8), multiple eigenvalues ​​and multiple eigenvectors can be solved. One eigenvalue uniquely corresponds to one eigenvector. Sort the multiple eigenvalues ​​from large to small, and the first three eigenvalues ​​are , and , that is, when Take the eigenvalues ​​of 0, 1, and 2 (here , and represents the three eigenvalues ​​of the k first domain points in formula (8), and the , and Represents the three eigenvalues ​​of the point cloud data in formula (2). Although they have the same symbols, they have different meanings. , and is a common usage in related technologies). The eigenvectors corresponding to these three eigenvalues ​​are , and , then the normal vector of the first domain point can be expressed by formula (9).

[0084] (9) In formula (9), is the normal vector of the first domain point. The meanings of other parameters in formula (9) are the same as those in formula (8), and their detailed description is omitted here.

[0085] When the normal vector of the first point cloud is calculated using step ef, it is necessary to ensure that the normal vector of the first point cloud points in a direction away from the point cloud centroid of the point cloud data, that is, to make formula (10) valid.

[0086] (10) The parameters in formula (10) have the same meanings as those in formula (1) and formula (6), and their detailed description is omitted here. If the calculated normal vector of the first point cloud is substituted into formula (10), and the inner product calculation result in formula (10) is negative, then the normal vector is negated and the negated normal vector is used as the normal vector of the first point cloud.

[0087] Step g: determining the Gaussian curvature and the mean curvature of the first point cloud based on the normal vectors of the plurality of first domain points.

[0088] Optionally, step g includes the following three steps.

[0089] In the first step, based on the normal vectors of the plurality of first domain points, the normal curvature of each first domain point is determined.

[0090] To simplify the calculation, a local coordinate system with the first point cloud as the origin is constructed , the local coordinate system consists of the origin and three coordinate axes The origin is the first point cloud, and the coordinate axis is the normal vector of the first point cloud; coordinate axis Perpendicular to the coordinate axis Two unit vectors, and the coordinate axis They are also perpendicular to each other.

[0091] Then in this local coordinate system, the first point cloud coordinates, The coordinates of the first domain point, the coordinates of the normal vector of the first point cloud, and the The coordinates of the normal vectors of the first domain points can be expressed by formula (11). There are a total of k first domain points, is an integer, Greater than 0 and less than or equal to k.

[0092] (11) In formula (11), is the first point cloud, For the The coordinate components of the first domain point in the local coordinate system; is the normal vector of the first field point, is the coordinate component of the normal vector of the first domain point in the local coordinate system; is the normal vector of the first point cloud, is the coordinate component of the normal vector of the first point cloud in the local coordinate system. The meanings of other parameters in formula (11) are the same as those in formula (6), and their detailed description is omitted here. Then the coordinate axis It can be expressed by formula (12).

[0093] (12) In formula (12), For the coordinate axis The coordinates in this local coordinate system are expressed as, For the coordinate axis The coordinates in this local coordinate system are expressed as, and is an intermediate variable and can be expressed by formula (13).

[0094] (13) The parameters in formula (13) have the same meanings as those in formula (11) and formula (12), and their detailed description is omitted here.

[0095] The first The normal curvature of a first domain point can be expressed by formula (14).

[0096] (14) In formula (14), For the The normal curvature of the first sphere point, For vector and vector The angle between For vector and vector The angle between and is an intermediate variable, which can be expressed by formula (15). The first point cloud and the second point cloud in formula (11) can be The coordinates of the first field point are obtained, and the details are omitted here. (15) The meanings of the parameters in formula (15) are the same as those in formula (11) and formula (14), and their detailed description is omitted here.

[0097] In the second step, the principal curvatures of the first point cloud are determined based on the normal curvature of each first domain point.

[0098] The principal curvatures include the maximum principal curvature and the minimum principal curvature.

[0099] The normal curvature is obtained by calculating the rate of change of the tangent of the surface in a given direction. Through the Euler formula, a least squares problem about the principal curvature can be constructed. This problem attempts to find an optimal principal curvature (that is, the maximum principal curvature and the minimum principal curvature) so that the difference between the estimated value and the actual value of the normal curvature of the surface at this point is minimized in all possible directions.

[0100] Therefore, when implementing the second step, the equation of the least squares problem can be constructed by Euler's formula, and then the principal curvature of the first point cloud can be solved. The constructed equation of the least squares problem is shown in formula (16).

[0101] (16) In formula (16), is the principal curvature of the first point cloud, For vector and vector The angle between The first point cloud corresponds to Unit vectors in the principal directions of ; For vector and vector The angle between For the The coordinates of the projections of the first domain points on the tangent plane through the first point cloud, ,vector Can be The coordinates of and the coordinates of the first point cloud in formula (11) are obtained, and the detailed description is omitted here. The meanings of other parameters in formula (16) are the same as those in formula (14), and the detailed description is omitted here.

[0102] By solving formula (16), we can get and Among them, when Greater than hour, is the maximum principal curvature, is the minimum principal curvature; when Less than hour, is the minimum principal curvature, is the maximum principal curvature.

[0103] There are many related technologies about constructing the equation of the least squares problem through the Euler formula and the implementation method of solving the equation, which will be omitted here.

[0104] In the third step, based on the principal curvature of the first point cloud, the Gaussian curvature and the mean curvature of the first point cloud are determined.

[0105] Based on the principal curvature equations in formula (17), the Gaussian curvature and mean curvature can be solved by formula (16) using Vieta's theorem. and Express it.

[0106] (17) In formula (17), is the principal curvature of the surface where the first point cloud is located at a certain point, is the mean curvature of the first point cloud, is the Gaussian curvature of the first point cloud.

[0107] By using Vieta's theorem, the Gaussian curvature and the mean curvature are solved by formula (16) to obtain and It is expressed as formula (18).

[0108] (18) The meanings of the parameters in formula (18) are the same as those in formula (16) and formula (17), and detailed description is omitted here.

[0109] For points other than the first point cloud in the point cloud data, the Gaussian curvature and the mean curvature of these points can also be calculated through steps eg, which are omitted here for details.

[0110] In step 204 , based on the Gaussian curvature and the mean curvature of each point, the multiple points are classified into four point cloud types: peak / well, ridge / valley, saddle, and plane.

[0111] Optionally, step 204 includes the following three steps: The first step is to obtain the principal curvature of the first point cloud.

[0112] For the relevant content of the principal curvature of the first point cloud, please refer to the aforementioned formula (16), and the detailed description is omitted here.

[0113] In the second step, based on the principal curvature of the first point cloud, the threshold of the Gaussian curvature and the threshold of the mean curvature are obtained.

[0114] Assuming that the calculation result of the principal curvature of a point on the local plane conforms to the Gaussian normal distribution, according to the 3sigma principle, the principal curvature of a point on the local plane is in the interval The probability of being within is 0.997, then the principal curvature of a point on the local plane is in the interval The probability of being outside is 0.003, which is the error in calculating the principal curvature of a point on the local plane. Among them, the local plane is the plane formed by the domain points of a certain point.

[0115] In this case, based on formula (19), it is possible to obtain the threshold of the Gaussian curvature and the threshold of the mean curvature based on the principal curvature of the first point cloud.

[0116] (19) In formula (19), is the threshold of Gaussian curvature, is the threshold of mean curvature, is the true value of the maximum principal curvature, is the true value of the minimum principal curvature, is 0.03.

[0117] In the third step, based on the threshold of the Gaussian curvature and the threshold of the mean curvature, as well as the Gaussian curvature and the mean curvature of each point, the point cloud type of the first point cloud is determined.

[0118] The third step can be represented by the following Table 1. The meanings of the parameters in Table 1 are the same as those in formula (19), and detailed description is omitted here.

[0119] Table 1: The relationship between Gaussian curvature threshold, mean curvature threshold and point cloud type, and the meaning of different point cloud types.

[0120]

[0121] Based on the values ​​of Gaussian curvature and mean curvature of the first point cloud, the point cloud type of the first point cloud can be determined. For points other than the first point cloud among the multiple points in the point cloud data, the point cloud types of these points can also be determined in the same way as the first point cloud.

[0122] In step 205 , a signature of the point cloud data is determined based on the number of points in the different point cloud types in each octant.

[0123] After all point cloud types are determined, the target area can be divided into 8 quadrants according to the coordinate axes of the coordinate system obtained after principal component analysis orientation.

[0124] In this case, step 205 includes the following two steps.

[0125] The first step is to count the number of points in different point cloud types in each of the eight hexagrams and perform normalization to obtain the feature array of the point cloud data.

[0126] Since there are a total of 8 hexagrams and the point cloud types in each hexagram include four categories, a total of 32 statistical values ​​can be obtained. After normalizing these 32 statistical values, the obtained feature array includes 32 components. This feature array can represent the distribution characteristics of the curvature of multiple points in the point cloud data in space.

[0127] For example, taking the specific data from artec3d as an example, the target area includes a 3D mesh model with 249,773 vertices and 499,690 triangle meshes. The 3D mesh model is visualized and point cloud is extracted to obtain the point cloud data of the 3D mesh model. Then, the point cloud data is processed using the method in steps 201-205, and the 32 statistical values ​​obtained are shown in Table 2 and Table 3. Figure 5 shown.

[0128] Table 2: Distribution characteristics of curvature of multiple points in point cloud data in space.

[0129]

[0130] The parameters in Table 2 represent the proportion of the number of points of a certain point cloud type in a certain quadrant to the entire point cloud data. For example, 0.00438466 means that the number of point clouds of peak / well type in quadrant 1 accounts for 0.00438466 of the entire point cloud data.

[0131] Figure 5 This is a schematic diagram of the distribution characteristics of the curvature of multiple points in the point cloud data in space, see Figure 5 ,exist Figure 5 In the figure, blue points indicate points with plane point cloud type; green points indicate points with ridge / valley point cloud type; purple points indicate points with saddle point cloud type (it can be seen from Table 2 that there are no saddle points in the point cloud data, so Figure 5 There are no purple points in the point cloud); red points indicate points whose point cloud type is peak / well.

[0132] In the disclosed embodiment, since statistics are performed based on the scale-invariant features such as the Gaussian curvature and the mean curvature of each point, the feature symbol obtained subsequently also has a certain scale-invariance.

[0133] The second step is to determine the feature symbols of the point cloud data based on the feature array.

[0134] This second step can be achieved by formula (20).

[0135] (20) In formula (20), is the feature symbol of point cloud data, is the feature array of point cloud data, is the i-th component in the feature array.

[0136] Through formula (20), the feature symbol of the point cloud data can be obtained, and the feature symbol is the point cloud feature extracted from the point cloud data.

[0137] Based on the feature, the matching relationship between different point cloud data can be determined. For example, the feature is extracted from different point cloud data, and the matching relationship between different point cloud data is determined by comparing the similarity of the features of different point cloud data, and then point cloud registration is performed based on the features of the point cloud data.

[0138] In some embodiments, the point cloud data in the target area may be divided into multiple sub-point clouds, and the feature symbols of each sub-point cloud may be calculated and then integrated into the feature symbols of the point cloud data in the target area, thereby simplifying the calculation process of the feature symbols of the point cloud data.

[0139] In the embodiment of the present disclosure, since the calculation only involves the three-dimensional coordinates of each point in the point cloud data, and does not involve the color information of the point cloud data and the intensity data of the sensor, etc., there is no need to ensure the accuracy of the color information of the point cloud data and the intensity data of the sensor. Instead, it is only necessary to ensure the accuracy of the three-dimensional coordinates of each point in the point cloud data, thereby reducing the requirements for the accuracy of the point cloud data.

[0140] The point cloud feature extraction method in the disclosed embodiment is versatile and has excellent robustness. Since the final feature symbol is based on curvature, and curvature has scale invariance, the obtained feature symbol also has a certain scale invariance. In addition, compared with extracting point cloud features by fitting the surface to obtain geometric parameters, when using Gaussian curvature and mean curvature to extract point cloud features of point cloud data, when encountering irregular surfaces (such as cross sections), the extracted point cloud features are more accurate, with smaller errors and a wider range of applications.

[0141] The following are device embodiments of the present application. For details not described in detail in the device embodiments, reference may be made to the above method embodiments.

[0142] Figure 6 A schematic diagram of a point cloud feature extraction device provided by an exemplary embodiment of the present disclosure is shown. Figure 6 The point cloud feature extraction device 600 includes: a first acquisition module 601, a second acquisition module 602, a classification module 603 and a feature extraction module 604.

[0143] The first acquisition module 601 is used to acquire point cloud data of the target area, where the point cloud data includes a plurality of points.

[0144] The second acquisition module 602 is used to acquire the Gaussian curvature and the average curvature of each point in the point cloud data.

[0145] The classification module 603 is used to classify the multiple points into four point cloud types: peak / well, ridge / valley, saddle, and plane based on the Gaussian curvature and the mean curvature of each point.

[0146] The feature extraction module 604 is used to determine the feature of the point cloud data based on the number of points in different point cloud types in each octant.

[0147] Optionally, the classification module 603 is also used to obtain the principal curvature of the first point cloud; based on the principal curvature of the first point cloud, obtain the threshold of Gaussian curvature and the threshold of average curvature; based on the threshold of Gaussian curvature and the threshold of average curvature, as well as the Gaussian curvature and the average curvature of each point, determine the point cloud type of the first point cloud.

[0148] Optionally, the principal curvature of the first point cloud includes a maximum principal curvature and a minimum principal curvature, and the classification module 603 is further configured to obtain a Gaussian curvature threshold and a mean curvature threshold based on the principal curvature of the first point cloud using the following formula:

[0149] in, is the threshold of Gaussian curvature, is the threshold of mean curvature, is the true value of the maximum principal curvature, is the true value of the minimum principal curvature, It is the probability that the value in the normal distribution is outside the 3sigma range.

[0150] Optionally, the target area includes 8 quadrants, and the device further includes: a statistical module 605.

[0151] The statistical module 605 is used to count the number of points in different point cloud types in each of the eight hexagrams, and perform normalization processing to obtain a feature array, wherein the feature array includes 32 components, each component is used to indicate the number of points of a point cloud type in a hexagram; The feature extraction module 604 is further used to determine the feature symbol of the point cloud data based on the number of points in different point cloud types in each quadrant of the point cloud data using the following formula:

[0152] in, is the feature symbol of point cloud data, is the feature array, is the ith component.

[0153] Optionally, the second acquisition module 602 is also used to obtain multiple first neighborhood points of the first point cloud, where the first point cloud is any point in the point cloud data; determine the normal vectors of the multiple first neighborhood points; and determine the Gaussian curvature and mean curvature of the first point cloud based on the normal vectors of the multiple first neighborhood points.

[0154] Optionally, the second acquisition module 602 is also used to determine the normal curvature of each first field point based on the normal vectors of multiple first field points; determine the principal curvature of the first point cloud based on the normal curvature of each first field point; and determine the Gaussian curvature and mean curvature of the first point cloud based on the principal curvature of the first point cloud.

[0155] Optionally, the second acquisition module 602 is further used to construct a KD tree of the point cloud data; based on the KD tree of the point cloud data, a K-nearest neighbor point search algorithm is used to acquire multiple first domain points of the first point cloud.

[0156] It should be noted that: when the point cloud feature extraction device provided in the above embodiment extracts point cloud features, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the point cloud feature extraction device provided in the above embodiment and the point cloud feature extraction method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0157] The division of modules in the embodiments of the present disclosure is schematic and is only a logical function division. There may be other division methods in actual implementation. In addition, each functional module in each embodiment of the present disclosure may be integrated into a processor, or may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated modules may be implemented in the form of hardware or in the form of software functional modules.

[0158] If the integrated module is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present disclosure is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a terminal device (which can be a personal computer, mobile phone, or communication device, etc.) or a processor (processor) to execute all or part of the steps of the method of each embodiment of the present disclosure. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.

[0159] Figure 7 Schematic diagram of the structure of the computer device provided by the embodiment of the present disclosure. Figure 7 As shown, the computer device 700 includes: a processor 701 and a memory 702 .

[0160] The processor 701 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 701 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 701 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 701 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 701 may also include an AI (Artificial Intelligence) processor, which is used to process computing operations related to machine learning.

[0161] The memory 702 may include one or more computer-readable storage media, which may be non-transitory. The memory 702 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 702 is used to store at least one instruction, which is used to be executed by the processor 701 to implement the point cloud feature extraction method provided in the embodiments of the present disclosure.

[0162] Those skilled in the art will understand that Figure 7 The structure shown in the figure does not constitute a limitation on the computer device 700, and the computer device 700 may include more or less components than shown in the figure, or combine some components, or adopt a different arrangement of components.

[0163] The embodiment of the present disclosure also provides a non-temporary computer-readable storage medium. When the instructions in the storage medium are executed by a processor of a computer device, the computer device is enabled to execute the point cloud feature extraction method provided in the embodiment of the present disclosure.

[0164] The embodiments of the present disclosure also provide a computer program product, including a computer program / instruction, which implements the point cloud feature extraction method provided in the embodiments of the present disclosure when the computer program / instruction is executed by a processor.

[0165] The above description is only an optional embodiment of the present disclosure and is not intended to limit the present disclosure. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A point cloud feature extraction method, It is characterized in that The method comprises: Acquire point cloud data of a target area, wherein the point cloud data includes a plurality of points; Obtaining the Gaussian curvature and the average curvature of each point in the point cloud data; Based on the Gaussian curvature and the average curvature of each point, the multiple points are divided into four point cloud types: peak / well, ridge / valley, saddle, and plane; A signature of the point cloud data is determined based on the number of the points in the different point cloud types in each octant.

2. The method according to claim 1, It is characterized in that Based on the Gaussian curvature and the average curvature of each point, the multiple points are divided into four point cloud categories: peak / well, ridge / valley, saddle, and plane, including: Obtaining the principal curvature of the first point cloud; Based on the principal curvature of the first point cloud, acquiring a threshold value of the Gaussian curvature and a threshold value of the average curvature; The point cloud type of the first point cloud is determined based on the threshold of the Gaussian curvature and the threshold of the average curvature, and the Gaussian curvature and the average curvature of each of the points.

3. The method according to claim 2, It is characterized in that The principal curvature of the first point cloud includes a maximum principal curvature and a minimum principal curvature, and the step of obtaining the threshold of the Gaussian curvature and the threshold of the average curvature based on the principal curvature of the first point cloud includes: The following formula is used to implement the principal curvature based on the first point cloud to obtain the threshold of the Gaussian curvature and the threshold of the average curvature: in, is the threshold of the Gaussian curvature, is the threshold of the mean curvature, is the true value of the maximum principal curvature, is the true value of the minimum principal curvature, It is the probability that the value in the normal distribution is outside the 3sigma range.

4. The method according to any one of claims 1 to 3, It is characterized in that The target area includes 8 octaves, and the method further includes: Counting the number of points in different point cloud types in each of the eight hexagrams, and performing normalization processing to obtain a feature array, wherein the feature array includes 32 components, each of which is used to indicate the number of points of one point cloud type in one hexagram; The determining the signature of the point cloud data based on the number of points in the different point cloud types in each octant of the point cloud data comprises: The following formula is used to implement the determination of the feature symbol of the point cloud data based on the number of points in the different point cloud types in each octave of the point cloud data: in, is the feature symbol of the point cloud data, is the feature array, is the i-th component.

5. The method according to any one of claims 1 to 3, It is characterized in that The obtaining of the Gaussian curvature and the average curvature of each point in the point cloud data comprises: Acquire a plurality of first neighborhood points of a first point cloud, where the first point cloud is any point in the point cloud data; determining normal vectors of the plurality of first domain points; Based on the normal vectors of the plurality of first domain points, a Gaussian curvature and a mean curvature of the first point cloud are determined.

6. The method according to claim 5, It is characterized in that The determining of the Gaussian curvature and the average curvature of the first point cloud based on the normal vectors of the plurality of first domain points comprises: Determining a normal curvature of each of the first domain points based on the normal vectors of the plurality of first domain points; determining a principal curvature of the first point cloud based on the normal curvature of each of the first field points; Based on the principal curvatures of the first point cloud, a Gaussian curvature and a mean curvature of the first point cloud are determined.

7. The method according to claim 5, It is characterized in that The step of obtaining a plurality of first neighborhood points of the first point cloud includes: Constructing a KD tree of the point cloud data; Based on the KD tree of the point cloud data, a K-nearest neighbor point search algorithm is used to obtain multiple first domain points of the first point cloud.

8. A point cloud feature extraction device, It is characterized in that The device comprises: A first acquisition module is used to acquire point cloud data of a target area, wherein the point cloud data includes a plurality of points; A second acquisition module is used to acquire the Gaussian curvature and the average curvature of each point in the point cloud data; A classification module, for classifying the plurality of points into four point cloud types, namely, peak / well, ridge / valley, saddle, and plane, based on the Gaussian curvature and the mean curvature of each of the points; A feature extraction module is used to determine a feature of the point cloud data based on the number of points in different point cloud types in each octant.

9. A computer device, It is characterized in that The computer device comprises: a memory and a processor, wherein at least one computer program is stored in the memory, and the at least one computer program is loaded and executed by the processor to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores at least one computer program, and the at least one computer program is loaded and executed by a processor to implement the method according to any one of claims 1 to 7.

Citation Information

Patent Citations

  • Simulated scene model establishment method and system for SAR irradiation region

    CN103631990A

  • Method and device for segmenting point cloud data, storage medium and electronic equipment

    CN110706238A

  • Three-dimensional laser point cloud feature point line complete and accurate extraction method

    CN113963138A

  • Point group feature extraction device, point group feature extraction method, and program

    JP2019159940A

  • Method for extracting rivet points in large scale three-dimensional point cloud base on deep learning

    US20220121875A1