A three-dimensional point cloud curvature mutation feature extraction method based on normal vector secondary discrimination

Abstract

The application discloses a three-dimensional point cloud curvature mutation feature extraction method based on normal vector secondary discrimination, which comprises the following steps: pre-processing three-dimensional point cloud data; constructing a covariance matrix based on normal vector consistency weighting, which is used for enhancing the representation of local geometric features; calculating a multi-scale relative curvature, and extracting candidate feature points through double threshold screening; performing normal vector depth analysis on the candidate points, including the calculation of normal vector consistency, angle variance and anisotropy degree; based on the relative curvature and the normal vector analysis result, performing feature classification to distinguish high curvature feature points, low curvature feature points and edge points; and the application can effectively distinguish high curvature features, low curvature features and edge features.

Description

Technical Field

[0001] This invention relates to the field of point cloud data processing technology, specifically to a method for extracting curvature abrupt change features of three-dimensional point clouds based on normal vector secondary analysis. Background Technology

[0002] Currently, obtaining information on rock mass structural surfaces mainly relies on traditional methods such as manual geological logging, contact surveying, and photogrammetry. Manual logging methods depend on geologists measuring the rock mass's attitude on-site, which is not only labor-intensive and inefficient but also difficult to implement in dangerous areas such as steep slopes and underground caverns, posing safety hazards. Contact surveying equipment, such as geological compasses and inclinometers, requires close contact with the rock mass surface, making it impossible to measure high and steep rock walls. Furthermore, the measurement accuracy is greatly affected by the operator's skill level, resulting in poor data reproducibility. With the rapid development of 3D sensing technology, devices such as lidar, structured light scanning, and stereo vision can quickly acquire massive amounts of 3D point cloud data. Point cloud data is increasingly widely used in fields such as rock mass detection. 3D point cloud curvature abrupt change feature extraction can extract regions of curvature abrupt changes from disordered 3D point clouds, helping to obtain rock mass information. Traditional point cloud curvature feature extraction methods are mainly based on local neighborhood PCA analysis, which identifies boundaries and high curvature regions by calculating the distribution of feature values. This leads to problems such as sensitivity to noise, easy false detection, inability to effectively distinguish between real boundaries and surface irregularities caused by noise, and lack of full utilization of normal vector information. Summary of the Invention

[0003] Purpose of the invention: The purpose of this invention is to provide a method for extracting the curvature abrupt change feature of three-dimensional point clouds based on normal vector secondary discrimination, in order to solve the problems of existing traditional point cloud curvature feature extraction methods, which are mainly based on local neighborhood PCA analysis and identify boundaries and high curvature regions by calculating the distribution of feature values. These methods are sensitive to noise, prone to false detection, unable to effectively distinguish between real boundaries and surface irregularities caused by noise, and lack full utilization of normal vector information.

[0004] Technical solution: The present invention provides a method for extracting curvature abrupt change features of three-dimensional point clouds based on normal vector secondary analysis, comprising the following steps:

[0005] (1) Preprocess the 3D point cloud data;

[0006] (2) Construct a covariance matrix based on normal vector consistency weighting to enhance the representation of local geometric features;

[0007] (3) Calculate the relative curvature at multiple scales and extract candidate feature points by screening with dual thresholds;

[0008] (4) Perform in-depth analysis of the normal vectors of the candidate points, including the calculation of normal vector consistency, angle variance and anisotropy.

[0009] (5) Based on the analysis results of relative curvature and normal vector, perform feature classification to distinguish high curvature feature points, low curvature feature points and edge points.

[0010] Furthermore, in step (1), the preprocessing includes: point cloud clipping, calculation of normal vectors and normalization; specifically: using the kd-tree algorithm to perform neighbor search, calculating the normal vectors of the 3D point cloud data, and obtaining the normal vector of each point cloud; and normalizing the point cloud normal vectors to obtain standard normal vectors.

[0011] Furthermore, in step (2), the weighted covariance matrix is ​​constructed as follows: In the covariance moment, the normal vector consistency weight is introduced, and the expression of local surface geometric features is optimized by calculating the weighted centroid and weighted covariance.

[0012] Furthermore, in step (3), the multi-scale relative curvature is calculated, including the ratio of background curvature to local curvature, and candidate points are screened by upper and lower thresholds to highlight curvature change regions.

[0013] Furthermore, in step (4), the depth analysis of the normal vector includes calculating the normal vector consistency coefficient, the normal vector angle variance, and the normal vector distribution anisotropy, which are used to evaluate the stability and directionality of local normal vector changes.

[0014] Furthermore, in step (5), feature classification uses multi-condition decision rules to achieve accurate classification of feature points and noise filtering by introducing relative curvature, normal vector consistency coefficient, angle variance and anisotropy degree.

[0015] The present invention discloses a three-dimensional point cloud curvature abrupt change feature extraction system based on normal vector second-order parsing, comprising:

[0016] Preprocessing module: used to preprocess 3D point cloud data;

[0017] Covariance matrix module: Used to construct a covariance matrix based on normal vector consistency weighting, which enhances the representation of local geometric features;

[0018] Candidate feature point module: used to calculate multi-scale relative curvature and extract candidate feature points through dual threshold filtering;

[0019] Normal vector depth analysis module: used to perform normal vector depth analysis on candidate points, including calculation of normal vector consistency, angle variance and anisotropy degree;

[0020] Classification module: Used to classify features based on the analysis results of relative curvature and normal vector, and to distinguish between high curvature feature points, low curvature feature points and edge points.

[0021] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the methods described herein.

[0022] An electronic device according to the present invention includes a memory and a processor, wherein the memory stores a computer program, and when the program is executed by the processor, it implements any of the methods described herein.

[0023] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: The present invention proposes a highly accurate and robust method for extracting three-dimensional point cloud curvature abrupt change features; through multi-scale curvature calculation with normal vector weighting, relative curvature nonlinear enhancement, dual-threshold candidate point extraction, normal vector consistency analysis, angle variance calculation and anisotropy evaluation, it achieves high-precision feature classification and noise filtering, and can effectively distinguish between high curvature features, low curvature features and edge features. Attached Figure Description

[0024] Figure 1 This is a flowchart of the present invention;

[0025] Figure 2 This is the preprocessed 3D point cloud model of the present invention;

[0026] Figure 3 This is the final three-dimensional point cloud model obtained in this invention. Detailed Implementation

[0027] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0028] like Figure 1 As shown, this embodiment of the invention provides a method for extracting three-dimensional point cloud curvature abrupt change features based on normal vector secondary analysis, including the following steps: The research case of this embodiment is a highway slope in Colorado, USA, which has a relatively obvious curvature abrupt change region.

[0029] S1. Preprocess the 3D point cloud data, obtaining the required research segment through cropping. A total of 293,857 points are involved. The processed point cloud is shown below. Figure 2 As shown. For 3D point cloud data processing, neighbor search is required. A kd-tree algorithm is used to build a kd-tree for the point cloud coordinates. For calculating the point cloud normal vectors, a local surface fitting method can be used to directly obtain the geometric features of the point cloud from the point cloud model. Based on the calculated point cloud normal vectors, the normal vectors of each point cloud are normalized using the following formula:

[0030] ;

[0031] Obtain the standard normal vector for each point cloud for subsequent point cloud processing.

[0032] S2. Construct the weighted covariance matrix. This invention improves upon traditional covariance matrix calculation by using normal vector consistency weights, adding normal vector weight coefficients when calculating the centroid of the point cloud within the distance range. The formula is as follows:

[0033] ɑ ;

[0034] ;

[0035] ;

[0036] in, For the normal vector weights, Let be the normal vector of this point cloud. Here, α is the normal vector of the point cloud, and α is the weight exponent parameter with a value of 1.5. Let k be the weighted centroids of the point clouds. For the j-th range of point cloud; It is the weighted covariance matrix.

[0037] A neighborhood search with a search range of k is performed on a single point cloud. The covariance matrix is ​​calculated using the weighted covariance matrix method. This covariance matrix is ​​then decomposed into eigenvalues, resulting in eigenvalues ​​sorted in descending order. The curvature of the point cloud is calculated using the curvature calculation formula based on three eigenvalues:

[0038] ;

[0039] S3. Calculate multi-scale relative curvature and extract candidate points. This invention uses relative curvature for curvature calculation. This method can highlight abrupt changes in point cloud curvature as much as possible, and can adapt the value range according to different point cloud conditions. It includes the following steps:

[0040] S31. Using a weighted covariance matrix, calculate the background curvature and local curvature of the point cloud based on the set range of neighboring points.

[0041] ;

[0042] ;

[0043] in, For background curvature, The number of surrounding points selected for calculating the background curvature is set to 100. For local curvature, The number of surrounding points selected for calculating the local curvature is 15.

[0044] S32. Calculate the background curvature and local curvature obtained above to obtain the relative curvature.

[0045] ;

[0046] S33. Calculate the relative curvature for each point and extract candidate points according to the set range.

[0047] ;

[0048] in, For the candidate point set, The upper value of relative curvature, This is the value of relative curvature. The value is 1.5. The value is 0.5.

[0049] S4. Depth Analysis of Normal Vectors. This invention calculates the normal vector consistency coefficient, normal vector angle variance, and normal vector distribution anisotropy by utilizing normal vector information. The process is as follows:

[0050] ;

[0051] ;

[0052] ;

[0053] in, The normal vector consistency coefficient. Let be the normal vector of this point cloud. For the range of point cloud normal vectors, The value is 25; The variance of the normal vector angle; The anisotropy degree of the normal vector distribution. These are the eigenvalues ​​calculated from the weighted covariance matrix.

[0054] S5. Intelligent Feature Classification. This invention achieves point cloud type classification through two-level analysis of point cloud relative curvature, normal vector consistency coefficient, normal vector angular variance, and normal vector distribution anisotropy. The implementation process is as follows:

[0055] The formula for selecting high curvature feature points is as follows:

[0056] ;

[0057] The formula for selecting feature points with low curvature is as follows:

[0058] ;

[0059] The formula for filtering edge points is as follows:

[0060] ;

[0061] Besides the three feature points mentioned above, all other points are considered noise and are removed. Relative curvature; The upper value of relative curvature, This is the value of relative curvature. The value is 1.5. The value is 0.5; The normal vector consistency coefficient. The normal vector consistency threshold is set to 0.85. The variance of the normal vector angle. The angle variance threshold is set to 15. The anisotropy degree of the normal vector distribution. The threshold value for the normal vector distribution is 0.7.

[0062] like Figure 3 As shown, the point cloud data after curvature abrupt change feature extraction yields a total of 31,389 points. This demonstrates that the curvature abrupt change region in the original model was successfully extracted with good accuracy. The method of this invention can extract point cloud curvature abrupt change regions with high accuracy and robustness, facilitating subsequent point cloud processing.

[0063] The parameters are explained in Table 1.

[0064] Table 1 Parameter Explanation

[0065] .

Claims

1. A method for extracting a curvature discontinuity feature of a three-dimensional point cloud based on a normal vector second-order discrimination, characterized in that, The method comprises the following steps: (1) preprocessing the three-dimensional point cloud data; (2) constructing a covariance matrix based on the consistency of the normal vector weighting, for enhancing the representation of local geometric features; (3) calculating the multi-scale relative curvature, and extracting candidate feature points through double-threshold screening; the calculation of the multi-scale relative curvature comprises the calculation of the ratio of the background curvature and the local curvature, and the candidate points are screened through upper and lower thresholds to highlight the curvature mutation area; the method comprises the following steps: S31, using the weighted covariance matrix, calculating the point cloud background curvature and the local curvature according to the set point cloud neighbor point selection range: ； ； wherein, is the background curvature, is the number of surrounding points selected for calculating the background curvature; is the local curvature, is the number of surrounding points selected for calculating the local curvature; S32, calculating the background curvature and the local curvature obtained to obtain the relative curvature: ； S33, calculating the relative curvature for each point, and extracting candidate points according to the set range: ； wherein, is a candidate point set, is a relative curvature upper value, is a relative curvature lower value; (4) performing normal vector depth analysis on the candidate points, including the calculation of the normal vector consistency, the angle variance and the anisotropy degree; the normal vector depth analysis comprises the calculation of the normal vector consistency coefficient, the normal vector angle variance and the normal vector distribution anisotropy degree, for evaluating the stability and directionality of the local normal vector change; the implementation process is as follows: ； ； ； wherein, is a normal vector consistency coefficient, is the point cloud normal vector, is the range point cloud normal vector; is a normal vector angle variance; is a normal vector distribution anisotropy degree, is an eigenvalue calculated according to the weighted covariance matrix; (5) based on the relative curvature and the normal vector analysis result, performing feature classification to distinguish high-curvature feature points, low-curvature feature points and edge points; the feature classification uses multi-condition decision rules to realize accurate classification of the feature points and noise filtering by introducing the relative curvature, the normal vector consistency coefficient, the angle variance and the anisotropy degree; the implementation process is as follows: The formula for screening high-curvature feature points is as follows: ； The formula for screening low-curvature feature points is as follows: ； The formula for screening edge points is as follows: ; in addition to the three feature points, the rest of the points are considered as noise and are removed; wherein, is a relative curvature; is a relative curvature upper value, is a relative curvature lower value, is a normal vector consistency coefficient, is a normal vector consistency threshold value, is a normal vector angle variance, is an angle variance threshold value; is a normal vector distribution anisotropy degree, is a normal vector distribution threshold value.

2. The method according to claim 1, wherein, In step (1), the preprocessing comprises point cloud clipping, normal vector calculation and normalization processing; specifically, the kd-tree algorithm is used for neighbor point search, the normal vector of each point cloud is calculated by performing normal vector calculation on the three-dimensional point cloud data, and the standard normal vector is obtained by performing normalization processing on the point cloud normal vector.

3. The method according to claim 1, wherein, In step (2), the weighted covariance matrix is constructed as follows: in the covariance matrix, the normal vector consistency weight is introduced, and the weighted centroid and weighted covariance calculation are used to optimize the expression of the local surface geometric feature.

4. A three-dimensional point cloud curvature mutation feature extraction system based on normal vector secondary resolution, characterized in that, The method is implemented by using any one of claims 1-3, comprising: a preprocessing module for preprocessing the three-dimensional point cloud data; a covariance matrix module for constructing a covariance matrix based on the consistency of the normal vector weighting, for enhancing the representation of local geometric features; a candidate feature point module for calculating the multi-scale relative curvature, and extracting candidate feature points through double-threshold screening; a normal vector depth analysis module for performing normal vector depth analysis on the candidate points, including the calculation of the normal vector consistency, the angle variance and the anisotropy degree; a classification module for performing feature classification based on the relative curvature and the normal vector analysis result, to distinguish high-curvature feature points, low-curvature feature points and edge points.

5. A computer-readable storage medium having stored thereon a computer program, characterized in that, When the program is executed by the processor, the method as claimed in any one of claims 1-3 is implemented.

6. An electronic device comprising a memory and a processor, characterized in that The memory stores a computer program, and when the program is executed by the processor, the method as claimed in any one of claims 1-3 is implemented.

Images