Muck pile point cloud under-segmentation extraction method

Abstract

The invention belongs to the technical field of underground metal mine blasting mining engineering, and relates to a muck pile point cloud under-segmentation extraction method. The method comprises the following steps: acquiring original three-dimensional point cloud data of a muck pile; preprocessing the original three-dimensional point cloud data of the muck pile, and reconstructing a three-dimensional point cloud model of the muck pile; constructing a neighborhood search structure based on the KD tree and determining the resolution; segmenting the muck pile three-dimensional point cloud model to generate segmentation result point cloud; constructing an oriented bounding box (OBB) through principal component analysis, and obtaining an orthogonal direction vector and a side length of the OBB; taking the OBB direction vector and the plane as a projection direction and a plane to obtain three orthogonal projection point clouds; a neighborhood radius is set according to projection point cloud parameters, and local density is calculated; and counting the maximum density value, judging the point cloud greater than the threshold value as an under-segmentation point cloud, and carrying out re-segmentation. The method has the beneficial effects that the under-segmentation point cloud is accurately identified through plane projection point cloud local density analysis, a basis is provided for adjusting a segmentation strategy, and the segmentation accuracy and precision are improved.

Description

Technical Field

[0001] This invention relates to the field of underground metal mine blasting mining engineering technology, and in particular to a method for under-segmentation extraction of blast pile point clouds. Background Technology

[0002] In underground metal mining blasting, 3D scanning technology has become an important tool for reconstructing the 3D morphology of blast piles. Based on the point cloud data of blast piles obtained by 3D scanning, efficient and accurate point cloud segmentation methods can be used to segment the point cloud data, providing precise information for measuring the block size of the blast pile. Due to significant differences in the shape and size of the rocks in the blast pile, the obtained point cloud data of the blast pile usually exhibits irregular shapes, large density differences, and complex stacking characteristics.

[0003] Common point cloud segmentation methods, such as region growing algorithms, DBSCAN algorithms, and VCCS+LCCP algorithms, typically rely on geometric features or predefined rules for segmentation. The high complexity of bursty point clouds leads to problems such as low accuracy and poor completeness in these segmentation methods, especially when dealing with bursty point clouds with significant regional density differences, where under-segmentation is prone to occur.

[0004] Furthermore, the multi-scale characteristics of blasted rocks make it difficult for single-scale segmentation methods to comprehensively and accurately identify blasted rock blocks. Summary of the Invention

[0005] Technical problems to be solved In view of the above-mentioned shortcomings and deficiencies of the existing technology, the present invention provides a method for under-segmentation extraction of burst point clouds, which aims to solve the problem of low accuracy and poor integrity of existing segmentation methods caused by the high complexity of burst point clouds. Especially when processing burst point clouds with large regional density differences, the method can effectively avoid under-segmentation by extracting under-segmented point clouds at multiple scales for subsequent re-segmentation.

[0006] Technical solution To achieve the above objectives, the main technical solutions adopted by the present invention include: In a first aspect, the present invention provides a method for under-segmentation extraction of bursty point clouds, comprising the following steps: Step 1: Obtain the original 3D point cloud data of the burst pile; Step 2: Preprocess the original 3D point cloud data of the blast pile and reconstruct the 3D point cloud model of the blast pile; Step 3: Construct a point cloud spatial neighborhood search structure based on the KD tree and determine the point cloud resolution; Step 4: Segment the 3D point cloud model of the burst pile and generate the segmentation result point cloud; Step 5: For each segmented point cloud, principal component analysis is used to determine the principal direction of the segmented point cloud, and an oriented bounding box (OBB) is constructed based on the principal direction to obtain the three mutually orthogonal direction vectors of the oriented bounding box and their side lengths in the corresponding directions. Step 6: For each segmented point cloud, take the three orthogonal direction vectors of its bounding box and the corresponding bounding box plane as the projection direction and projection plane, respectively, and perform orthogonal projection to obtain three orthogonal projected point clouds. ; Step 7: Based on the projected point cloud The neighborhood search radius is set by the side length of the bounding box plane and the point cloud resolution, and the local density of the point cloud is calculated based on the neighborhood search radius; Step 8: Statistical segmentation result point cloud Planar projection point cloud Local density maximum R max Compare it with the set density threshold R th The comparison is used to determine whether the segmentation is complete.

[0007] As a further improvement to the method of the present invention, step 2 specifically includes the following method: Step 2.1: Use the SOR algorithm to denoise the original 3D point cloud data of the burst pile, remove outliers and edge noise in the original 3D point cloud data of the burst pile, and obtain the denoised 3D point cloud data of the burst pile. Step 2.2: Based on the voxel downsampling algorithm, downsample the denoised 3D point cloud data of the burst pile to reconstruct the 3D point cloud model of the burst pile.

[0008] As a further improvement to the method of the present invention, step 3 involves constructing a KD tree and determining the point cloud resolution using the average neighborhood distance, specifically including the following methods: Step 3.1: Based on the three-dimensional spatial coordinates of each point in the point cloud, the KD tree algorithm is used to spatially divide the point cloud and establish the spatial neighborhood search structure of the point cloud, i.e., the KD tree structure. Step 3.2: Using any point in the point cloud p i For the query point, perform a neighborhood search in the KD-tree structure to obtain the point. p i The set of neighboring points within a preset spatial range; Step 3.3: Based on the spatial distance between the neighborhood point set point and the query point, calculate the average neighborhood distance of the point cloud, and use this average neighborhood distance as the point cloud resolution. .

[0009] As a further improvement to the method of the present invention, step 5 specifically involves: processing each segmentation result point cloud... First, its covariance matrix is ​​calculated, and then eigenvalue decomposition is performed on the covariance matrix to obtain a set of eigenvectors. and the corresponding eigenvalues : Among them, the feature vector Corresponding to the principal directions of the point cloud, the feature vectors constitute the principal direction group of the point cloud in three-dimensional space. The principal direction group consists of three pairwise orthogonal direction vectors. composition; Based on the main direction Construct a direction matrix for a directed bounding box, using the centroid of the segmented point cloud as the origin of the local coordinate system; map the coordinates of all points in the segmented point cloud to the local coordinate system defined by the direction matrix and the origin; For any principal direction Segmentation result point cloud any point in the middle The projection in this principal direction is Determined in this coordinate system Extreme range in each principal direction This allows us to obtain the side lengths of the directed bounding box in the three orthogonal directions. .

[0010] As a further improvement to the method of the present invention, step 7 specifically includes the following method: Step 7.1: Based on the projected point cloud The neighborhood search radius is set by the side length of the bounding box plane and the point cloud resolution. ; Step 7.2: Based on the projected point cloud neighborhood search radius Calculate the projected point cloud Local density.

[0011] As a further improvement to the method of the present invention, step 7.1 specifically involves: based on the projected point cloud... Side length of the bounding box plane and ( ), calculate the projected point cloud The length of the diagonal of the bounding box plane At the same time, based on the determined point cloud resolution Determine the projected point cloud neighborhood search radius .

[0012] As a further improvement to the method of the present invention, step 7.2 specifically involves: projecting the point cloud. any point in space ,by Centered on the neighborhood search radius Determine the midpoint of the point cloud. neighborhood point set Calculate the neighboring points in the neighborhood point set. Between the current spatial point European distance The Gaussian kernel function is used to perform distance-weighted accumulation on neighboring points to calculate the projected point cloud. Local density .

[0013] As a further improvement to the method of the present invention, in step 8, if R max >R th If R is not segmented, it is determined to be an under-segmented point cloud, which needs to be extracted and re-segmented; max ≤R th If the segmentation is complete, there are no unsegmented point clouds.

[0014] In a second aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed, implements the method for undersegmentation extraction of burst point clouds as described in any of the first aspects above.

[0015] Thirdly, the present invention provides a storage device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements the method for undersegmentation extraction of burst point clouds as described in any of the first aspects above.

[0016] The beneficial effects of this invention are: 1. Achieving Objective and Automated Determination of Segmentation Integrity. This invention overcomes the subjective limitations of traditional methods that rely on human experience, and innovatively proposes an analysis framework based on multi-orthogonal projection density fusion. First, principal component analysis is used to determine the principal direction of the segmented point cloud and construct a directed bounding box. Then, a two-dimensional point cloud is obtained by projection along three orthogonal directions. Finally, a comprehensive judgment is made based on the local density distribution characteristics of the projected point cloud. This method avoids the one-sidedness of a single criterion through multi-view complementarity, and by combining an adaptive parameter adjustment mechanism with a quantitative point cloud density standard, it achieves objective and automated determination of segmentation integrity.

[0017] 2. Clear physical meaning and computational efficiency. By using projection dimensionality reduction, the complex 3D under-segmentation problem is transformed into 2D density analysis, achieving a balance between clear physical meaning and computational efficiency. The local density of the projected point cloud directly reflects the geometric morphology of the rock blocks: the projected density distribution of a single, intact rock block is uniform, while under-segmented rock blocks show anomalous clustering. This physical correspondence ensures strong interpretability of the results. Simultaneously, converting 3D neighborhood search into 2D computation effectively reduces time complexity, improves processing efficiency, and meets the real-time requirements of engineering projects.

[0018] 3. Broad compatibility and strong adaptability with existing technologies. As a general-purpose post-processing module for segmentation, it can be seamlessly integrated with various point cloud segmentation algorithms. Through an adaptive parameter mechanism, it can adapt to different acquisition conditions and rock block sizes, possessing good engineering applicability and promotional value. Attached Figure Description

[0019] Figure 1 A flowchart of a method for under-segmentation extraction of burst point clouds provided in an embodiment of the present invention; Figure 2 This refers to the original three-dimensional point cloud data of the burst pile in this embodiment of the invention; Figure 3 This is a three-dimensional point cloud model of a burst pile in an embodiment of the present invention; Figure 4 This is a result image of the three-dimensional point cloud model of the burst pile after segmentation using the DBSCAN algorithm in an embodiment of the present invention; Figure 5 This is a projection point cloud image of a certain segmentation result of the three-dimensional point cloud model of the burst pile on its three orthogonal planes OBB in an embodiment of the present invention; Figure 6 This is a statistical diagram of the projection point cloud density distribution of the segmentation result of the three-dimensional point cloud model of the blast pile in an embodiment of the present invention; Figure 7 This refers to the under-segmentation result point cloud of the three-dimensional point cloud model of the burst pile extracted according to the point cloud density threshold in this embodiment of the invention; Figure 8 The point cloud values ​​are the undersegmented results of the burst pile 3D point cloud model in this embodiment of the invention. Detailed Implementation

[0020] To better explain and facilitate understanding of the present invention, it will be described in detail below with reference to the accompanying drawings and specific embodiments. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a clearer and more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0021] Firstly, such as Figure 1As shown, the present invention provides a method for under-segmentation extraction of burst point clouds, comprising the following steps: Step 1: Use a 3D scanner to acquire the original 3D point cloud data of the burst pile; Specifically, in step 1, a handheld 3D scanner is used. Scanning parameters are set according to the scanner's characteristics and the morphology of the rock pile, a scanning path is planned, and multi-angle, multi-position scanning is performed. During the scanning process, the various scan data are automatically stitched together to finally obtain the original 3D point cloud data of the rock pile, such as... Figure 2 As shown.

[0022] Step 2: Preprocess the original 3D point cloud data of the blast pile to reconstruct the 3D point cloud model of the blast pile; the preprocessing includes point cloud denoising and downsampling; Step 2.1: Use the SOR algorithm to denoise the original 3D point cloud data of the burst pile, remove outliers and edge noise in the original 3D point cloud data of the burst pile, and obtain the denoised 3D point cloud data of the burst pile. Step 2.2: Based on the voxel downsampling algorithm, downsample the denoised 3D point cloud data of the burst pile to reconstruct the 3D point cloud model of the burst pile;

[0023] The three-dimensional point cloud model of the bursting pile in this embodiment is as follows: Figure 3 As shown.

[0024] Step 3: Construct a point cloud spatial neighborhood search structure based on a KD tree to support fast neighborhood search operations for any point in the point cloud and determine the point cloud resolution; Step 3.1: Based on the three-dimensional spatial coordinates of each point in the point cloud, the KD tree algorithm is used to spatially divide the point cloud and establish the spatial neighborhood search structure of the point cloud, i.e., the KD tree structure. Step 3.2: Using any point in the point cloud p i For the query point, perform a neighborhood search in the KD-tree structure to obtain the point. p i The set of neighboring points within a preset spatial range; Step 3.3: Based on the spatial distance between the neighborhood point set point and the query point, calculate the average neighborhood distance of the point cloud, and use this average neighborhood distance as the point cloud resolution. .

[0025] Step 4: Use the DBSCAN segmentation algorithm to segment the reconstructed 3D point cloud model of the burst pile and generate the segmentation result point cloud.

[0026] In this embodiment, a total of 77 segmentation result point clouds were generated, and the segmentation results are as follows: Figure 4 As shown.

[0027] Step 5: For each segmented point cloud, Principal Component Analysis (PCA) is used to determine the principal direction of the segmented point cloud, and a directed bounding box is constructed based on the principal direction to obtain the three mutually orthogonal direction vectors of the directed bounding box and their side lengths in the corresponding directions. The directed bounding box consists of three pairwise orthogonal direction vectors, which correspond to the first principal direction, the second principal direction, and the third principal direction of the point cloud in space, respectively. The three direction vectors together define the local coordinate system of the point cloud.

[0028] Specifically, for each segmentation result point cloud First, its covariance matrix is ​​calculated, and then eigenvalue decomposition is performed on the covariance matrix to obtain a set of eigenvectors. and the corresponding eigenvalues : Among them, the feature vector Corresponding to the principal directions of the point cloud, the feature vectors constitute the principal direction group of the point cloud in three-dimensional space. The principal direction group consists of three pairwise orthogonal direction vectors. composition.

[0029] Based on the main direction Construct a direction matrix for a directed bounding box, using the centroid of the segmented point cloud as the origin of the local coordinate system; map the coordinates of all points in the segmented point cloud to the local coordinate system defined by the direction matrix and the origin; For any principal direction Segmentation result point cloud any point in the middle The projection in this principal direction is Determine the extreme range of the point cloud in each principal direction within this coordinate system. This allows us to obtain the side lengths of the directed bounding box in the three orthogonal directions. .

[0030] Step 6: For each segmented point cloud, take the three orthogonal direction vectors of its bounding box and the corresponding bounding box plane as the projection direction and projection plane, respectively, and perform orthogonal projection to obtain three orthogonal projected point clouds. .

[0031] In this embodiment, steps 5 and 6 are executed on any segmentation result generated in step 4 to obtain three orthogonal planar projection point clouds of the segmentation result point cloud, namely planar projection point cloud 1, planar projection point cloud 2, and planar projection point cloud 3. The obtained planar projection point clouds are as follows: Figure 5 As shown.

[0032] Step 7: Based on the projected point cloud The neighborhood search radius is set by the side length of the bounding box plane and the point cloud resolution, and the local density of the point cloud is calculated based on the neighborhood search radius.

[0033] Step 7.1: Based on the projected point cloud The neighborhood search radius is set by the side length of the bounding box plane and the point cloud resolution. ; Specifically, based on the projected point cloud Side length of the bounding box plane and ( ), calculate the projected point cloud The length of the diagonal of the bounding box plane At the same time, based on the determined point cloud resolution Determine the projected point cloud neighborhood search radius .

[0034] Step 7.2: Based on the projected point cloud neighborhood search radius Calculate the projected point cloud Local density.

[0035] Specifically, for projected point clouds any point in space ,by Centered on the neighborhood search radius Determine the midpoint of the point cloud. neighborhood point set Calculate the neighboring points in the neighborhood point set. Between the current spatial point European distance The Gaussian kernel function is used to perform distance-weighted accumulation on neighboring points to calculate the projected point cloud. Local density .

[0036] Step 8: Calculate the local density maximum R of the planar projection point cloud of the segmented point cloud. max Compare it with the set density threshold R th Compare the results to determine whether the segmentation is complete; If R max >R th If the value is not specified, it is determined to be an under-segmented point cloud, which needs to be extracted and re-segmented. If R max ≤R th If the segmentation is complete, there are no unsegmented point clouds.

[0037] Specifically, according to steps 6 and 7, the local density features of the projected point cloud of the segmented point cloud on its three oriented bounding box planes are calculated, and the maximum local density is used as the density characterization parameter of the segmented point cloud.

[0038] like Figure 6 As shown, the above calculation process is repeated for all segmented point clouds to obtain the distribution of local density maximum values ​​for each segmented point cloud. Based on this distribution, a local density threshold R is set. th =220. When the maximum local density of the segmented point cloud is lower than this threshold, the segmented point cloud is determined to be an under-segmented point cloud.

[0039] The segmentation point cloud extracted based on the above threshold determination rules is as follows: Figure 7 As shown.

[0040] In this embodiment, taking one segmented point cloud as an example, the maximum local density of its three projected point clouds is 1649.10, and this is compared with the actual under-segmented point cloud (such as...). Figure 8 As shown in the figure, the undersegmentation extraction method for the burst pile point cloud achieves an accuracy rate of 93% in identifying undersegmented point clouds. The accuracy rate of the undersegmented point cloud identification is calculated as the percentage of the number of undersegmented result point clouds identified by the present invention to the actual number of undersegmented result point clouds.

[0041] In this embodiment, the DBSCAN segmentation algorithm used, the number of point clouds generated from the segmentation results, the calculation method of local density, and the density threshold, etc., are only exemplary settings used to illustrate the technical solution of the present invention and do not constitute a limitation of the present invention.

[0042] In a second aspect, embodiments of the present invention provide a computer-readable storage medium having a computer program stored thereon, wherein the program, when executed, implements the method for undersegmentation extraction of burst point clouds as described in any of the first aspects above.

[0043] Thirdly, embodiments of the present invention provide a storage device, including a storage medium and a processor, wherein the storage medium stores a computer program, and when the program is executed by the processor, it implements the method for undersegmentation extraction of burst point clouds as described in any of the first aspects above.

[0044] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0045] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, then this invention should also include these modifications and variations.

[0046] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for under-segmentation extraction of bursty point clouds, characterized in that, Includes the following steps: Step 1: Obtain the original 3D point cloud data of the burst pile; Step 2: Preprocess the original 3D point cloud data of the blast pile and reconstruct the 3D point cloud model of the blast pile; Step 3: Construct a point cloud spatial neighborhood search structure based on the KD tree and determine the point cloud resolution; Step 4: Segment the 3D point cloud model of the burst pile and generate the segmentation result point cloud; Step 5: For each segmented point cloud, principal component analysis is used to determine the principal direction of the segmented point cloud, and a directed bounding box (OBB) is constructed based on the principal direction to obtain the three mutually orthogonal direction vectors of the directed bounding box and their side lengths in the corresponding directions. Step 6: For each segmented point cloud, take the three orthogonal direction vectors of its bounding box and the corresponding bounding box plane as the projection direction and projection plane, respectively, and perform orthogonal projection to obtain three orthogonal projected point clouds. ; Step 7: Based on the projected point cloud The neighborhood search radius is set by the side length of the bounding box plane and the point cloud resolution, and the local density of the point cloud is calculated based on the neighborhood search radius; Step 8: Statistical segmentation result point cloud Planar projection point cloud Local density maximum R max Compare it with the set density threshold R th The comparison is used to determine whether the segmentation is complete.

2. The method for under-segmentation extraction of burst point clouds according to claim 1, characterized in that, Step 2 specifically includes the following methods: Step 2.1: Use the SOR algorithm to denoise the original 3D point cloud data of the burst pile, remove outliers and edge noise in the original 3D point cloud data of the burst pile, and obtain the denoised 3D point cloud data of the burst pile. Step 2.2: Based on the voxel downsampling algorithm, downsample the denoised 3D point cloud data of the burst pile to reconstruct the 3D point cloud model of the burst pile.

3. The method for under-segmentation extraction of burst point clouds according to claim 2, characterized in that, Step 3 involves constructing a KD tree and determining the point cloud resolution using the average neighborhood distance. This process includes the following methods: Step 3.1: Based on the three-dimensional spatial coordinates of each point in the point cloud, the KD tree algorithm is used to spatially divide the point cloud and establish the spatial neighborhood search structure of the point cloud, i.e., the KD tree structure. Step 3.2: Using any point in the point cloud p i For the query point, perform a neighborhood search in the KD-tree structure to obtain the point. p i The set of neighboring points within a preset spatial range; Step 3.3: Based on the spatial distance between the neighborhood point set point and the query point, calculate the average neighborhood distance of the point cloud, and use this average neighborhood distance as the point cloud resolution. .

4. The method for under-segmentation extraction of burst point clouds according to claim 3, characterized in that, In step 5, specifically: for each segmentation result point cloud First, its covariance matrix is ​​calculated, and then eigenvalue decomposition is performed on the covariance matrix to obtain a set of eigenvectors. and the corresponding eigenvalues : Among them, the feature vector Corresponding to the principal directions of the point cloud, the feature vectors constitute the principal direction group of the point cloud in three-dimensional space. The principal direction group consists of three pairwise orthogonal direction vectors. composition; Based on the main direction Construct a direction matrix for a directed bounding box, and map the segmented point cloud to the local coordinate system defined by the direction matrix; For any principal direction Segmentation result point cloud any point in the middle The projection in this direction is Determine the extreme range of the point cloud in each principal direction within this coordinate system. This allows us to obtain the side lengths of the directed bounding box in the three orthogonal directions. .

5. The method for under-segmentation extraction of burst point clouds according to claim 4, characterized in that, Step 7 specifically includes the following methods: Step 7.1: Based on the projected point cloud The neighborhood search radius is set by the side length of the bounding box plane and the point cloud resolution. ; Step 7.2: Based on the projected point cloud neighborhood search radius Calculate the projected point cloud Local density.

6. The method for under-segmentation extraction of burst point clouds according to claim 5, characterized in that, In step 7.1, specifically: based on the projected point cloud... Side length of the bounding box plane and ( ), calculate the projected point cloud The length of the diagonal of the bounding box plane At the same time, based on the determined point cloud resolution Determine the projected point cloud neighborhood search radius .

7. The method for under-segmentation extraction of burst point clouds according to claim 6, characterized in that, In step 7.2, specifically: projecting the point cloud. any point in space ,by Centered on the neighborhood search radius Determine the midpoint of the point cloud. neighborhood point set Calculate the neighboring points in the neighborhood point set. Between the current spatial point European distance The Gaussian kernel function is used to perform distance-weighted accumulation on neighboring points to calculate the projected point cloud. Local density .

8. The method for under-segmentation extraction of burst point clouds according to claim 7, characterized in that, In step 8, if R max >R th If R is not segmented, it is determined to be an under-segmented point cloud, which needs to be extracted and re-segmented; max ≤R th If the segmentation is complete, there are no unsegmented point clouds.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the method for undersegmentation extraction of burst point clouds as described in any one of claims 1 to 8.

10. A storage device comprising a storage medium and a processor, the storage medium storing a computer program, characterized in that, When the processor executes the computer program, it implements the method for under-segmentation extraction of burst point clouds as described in any one of claims 1 to 8.

Images