A method and device for quickly sampling a three-dimensional point cloud based on a graph structure

Abstract

The application provides a kind of based on graph structure's three-dimensional point cloud fast sampling method and device, it is related to computer simulation and method optimization technical field. Including: using K nearest neighbor algorithm, generate the weighted adjacency matrix of three-dimensional point cloud data;Three-dimensional point cloud data and weighted adjacency matrix are segmented, obtain multiple sub-blocks and multiple sub-adjacency matrix, according to multiple sub-blocks and multiple sub-adjacency matrix, determine the subgraph and graph laplacian matrix corresponding to each sub-block;According to the eigenvalue of graph laplacian matrix, preset spectral truncation threshold and the impulse signal of node in each subgraph, utilize Chebyshev approximation algorithm, determine filter and the local energy of node in each subgraph;According to filter, local energy and preset regularization parameter, sample multiple target nodes in the node of each subgraph, determine the sampling signal set of three-dimensional point cloud data according to multiple target nodes.In processing large-scale point cloud data, reduce the computational complexity and improve sampling efficiency.

Description

Technical Field

[0001] This invention relates to the field of computer simulation and method optimization technology, and in particular to a method and apparatus for rapid sampling of three-dimensional point clouds based on graph structures. Background Technology

[0002] 3D point cloud sampling technology reduces redundant points in point cloud data, achieving data compression while preserving key geometric information. This technology is widely used in fields such as autonomous driving, virtual reality, and industrial inspection. 3D point cloud data is typically collected by devices such as LiDAR or depth cameras. Due to the massive volume and redundancy of the collected 3D point cloud data, directly processing it incurs enormous computational and storage overhead. Therefore, rapid sampling methods have become one of the key technologies for point cloud processing.

[0003] Currently, 3D point cloud sampling methods include curvature-based sampling methods and deep learning-based sampling methods. Curvature-based sampling methods prioritize retaining points with higher curvature, thus better preserving geometric features; however, they suffer from high computational complexity when processing large-scale point cloud data. Deep learning-based sampling methods achieve adaptive sampling by learning the intrinsic structure of the point cloud, but the training of the corresponding models requires a large amount of labeled data, resulting in low sampling efficiency for large-scale 3D point cloud data. Summary of the Invention

[0004] The purpose of this invention is to provide a method and apparatus for rapid sampling of 3D point clouds based on graph structures, which solves the problems of high computational complexity and low sampling efficiency in the prior art when processing large-scale point cloud data.

[0005] To address the aforementioned technical problems, the embodiments of the present invention provide the following technical solutions:

[0006] The first aspect of this invention provides a fast sampling method for 3D point clouds based on graph structures, comprising:

[0007] The weighted adjacency matrix of 3D point cloud data is generated using the K-nearest neighbor algorithm.

[0008] According to a preset number, the 3D point cloud data and the weighted adjacency matrix are divided into multiple sub-blocks and corresponding sub-adjacency matrices. Based on the multiple sub-blocks and sub-adjacency matrices, the subgraph corresponding to each sub-block and the graph Laplacian matrix corresponding to each subgraph are determined.

[0009] Based on the eigenvalues ​​corresponding to the graph Laplacian matrix, the preset spectral truncation threshold, and the pulse signal of each node in each subgraph, the Chebyshev approximation algorithm is used to determine the filter and the local energy of each node in each subgraph.

[0010] Based on the filter, local energy, and preset regularization parameters, multiple target nodes are sampled in each subgraph node, and the sampling signal set corresponding to the 3D point cloud data is determined based on the multiple target nodes.

[0011] A second aspect of the present invention provides a graph-based three-dimensional point cloud rapid sampling device, comprising:

[0012] The generation module is used to generate a weighted adjacency matrix of 3D point cloud data using the K-nearest neighbor algorithm.

[0013] The segmentation module is used to segment the 3D point cloud data and the weighted adjacency matrix according to a preset number, to obtain multiple sub-blocks and corresponding multiple sub-adjacency matrices, and to determine the subgraph corresponding to each sub-block and the graph Laplacian matrix corresponding to each subgraph based on the multiple sub-blocks and multiple sub-adjacency matrices.

[0014] The determination module is used to determine the filter and the local energy of each node in the subgraph based on the eigenvalues ​​corresponding to the graph Laplacian matrix, the preset spectral truncation threshold, and the pulse signal of each node in the subgraph, using the Chebyshev approximation algorithm.

[0015] The sampling module is used to sample multiple target nodes in each subgraph based on the filter, local energy and preset regularization parameters, and determine the sampling signal set corresponding to the 3D point cloud data based on the multiple target nodes.

[0016] Compared to existing technologies, the present invention provides a graph-based method and apparatus for fast sampling of 3D point clouds. It utilizes the K-nearest neighbor algorithm to generate a weighted adjacency matrix of the 3D point cloud data. The 3D point cloud data and the weighted adjacency matrix are segmented according to a preset number of segments, resulting in multiple sub-blocks and corresponding sub-adjacency matrices. Based on the multiple sub-blocks and sub-adjacency matrices, a subgraph corresponding to each sub-block and a graph Laplacian matrix corresponding to each subgraph are determined. Based on the eigenvalues ​​of the graph Laplacian matrix, a preset spectral truncation threshold, and the pulse signals of nodes in each subgraph, a Chebyshev approximation algorithm is used to determine a filter and the local energy of nodes in each subgraph. Based on the filter, local energy, and preset regularization parameters, multiple target nodes are sampled from the nodes in each subgraph, and the sampled signal set corresponding to the 3D point cloud data is determined based on these target nodes. In this way, the 3D point cloud data can be divided into multiple sub-blocks, the weighted adjacency matrix can be divided into multiple sub-adjacency matrices, and multiple subgraphs corresponding to the sub-blocks can be determined based on the multiple sub-blocks and the multiple sub-adjacency matrices. Each subgraph is sampled, and the sampled signal set can be obtained through the multiple target nodes sampled in each subgraph, which makes the computational complexity low when processing large-scale point cloud data. Furthermore, based on the filter, local energy, and preset regularization parameters, multiple target nodes are sampled in the nodes of each subgraph, which maximizes the local energy and minimizes the correlation interference, thereby making the sampling efficiency high. Attached Figure Description

[0017] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, with the same or corresponding reference numerals denoteing the same or corresponding parts, wherein:

[0018] Figure 1 A flowchart illustrating a graph-based method for fast sampling of 3D point clouds is shown.

[0019] Figure 2 A schematic diagram of a graph-based 3D point cloud rapid sampling device is shown. Detailed Implementation

[0020] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. 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 more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0021] It should be noted that, unless otherwise stated, the technical or scientific terms used in this invention should have the ordinary meaning as understood by those skilled in the art.

[0022] The methods in the embodiments of the present invention will be described in detail below.

[0023] Figure 1 A flowchart illustrating a graph-based fast sampling method for 3D point clouds in an embodiment of the present invention is shown schematically. See [link to flowchart illustration]. Figure 1 As shown, this graph-based method for fast sampling of 3D point clouds may include:

[0024] S101. Using the K-nearest neighbor algorithm, generate a weighted adjacency matrix of the 3D point cloud data.

[0025] Specifically, the K-Nearest Neighbors Algorithm (KNN) is used to generate a weighted adjacency matrix for 3D point cloud data, including:

[0026] Step A1: Construct an adjacency graph of the 3D point cloud data using the K-nearest neighbor algorithm.

[0027] In this context, each node in the adjacency graph corresponds to a point in the 3D point cloud data, and each point has K nearest neighboring points.

[0028] The adjacency graph is an adjacency graph based on Euclidean distance.

[0029] Step A2: Calculate the edge weights between each point and its K neighboring points.

[0030] The expression for edge weight is:

[0031] ;

[0032] in, For point and points corresponding Edge weights between adjacent nodes For point The corresponding coordinate vector, For point corresponding The coordinate vectors of the adjacent points To control the scaling parameters that affect distance, For point Corresponding coordinate vectors and points corresponding Euclidean distance between the coordinate vectors of adjacent points It is the square of the L2 norm.

[0033] Step A3: Generate a weighted adjacency matrix based on the edge weights.

[0034] Weighted adjacency matrix The non-zero elements in the equation correspond to the edge weights.

[0035] S102. According to the preset number, the 3D point cloud data and the weighted adjacency matrix are divided into multiple sub-blocks and corresponding multiple sub-adjacency matrices. Based on the multiple sub-blocks and multiple sub-adjacency matrices, the subgraph corresponding to each sub-block and the graph Laplacian matrix corresponding to each subgraph are determined.

[0036] Specifically, according to a preset number, the 3D point cloud data and the weighted adjacency matrix are divided into multiple sub-blocks and corresponding sub-adjacency matrices. Based on the multiple sub-blocks and sub-adjacency matrices, the subgraph corresponding to each sub-block and the graph Laplacian matrix corresponding to each subgraph are determined, including:

[0037] Step B1: Divide the 3D point cloud data into multiple sub-blocks according to the preset number.

[0038] Each sub-block corresponds to a different 3D point cloud data, and each sub-block has a sub-block index.

[0039] The preset number can be , 3D point cloud data Divided into Individual blocks. Each sub-block is a non-overlapping sub-block. Each sub-block is represented as ,in, For the first sub-block, For the first A small block, For the first Individual blocks, 3D point cloud data satisfy That is, the union of all sub-blocks equals the 3D point cloud data. The size of each sub-block is represented as ,in, For set The cardinality is the number of points in a point cloud sub-block. For the first The size of each sub-block. Segmented. Each sub-block is displayed using a small green box.

[0040] Step B2: Divide the weighted adjacency matrix according to the sub-block index to obtain the sub-adjacency matrix corresponding to each sub-block.

[0041] Sub-block index refers to According to the sub-block index, the weighted adjacency matrix is... Divided into Adjacency matrix , ,in, For the first Adjacency matrix It is the first Sub-block The corresponding sub-adjacency matrix.

[0042] The number of all sub-adjacency matrices is the same as the number of all sub-blocks.

[0043] Step B3: Take the set of points of each sub-block as the vertex set, and construct the subgraph corresponding to each sub-block based on the vertex set and the sub-adjacency matrix.

[0044] For example, the first Sub-block The corresponding subgraph is represented as , where subgraph The set of vertices and the first Sub-block The point set corresponds to the subgraph. The subgraph is used to describe the local structural relationships of each point in the corresponding sub-block.

[0045] Step B4: Calculate the corresponding degree matrix based on the elements in the sub-adjacency matrix.

[0046] The degree matrix corresponding to the sub-adjacency matrix is ​​a diagonal matrix.

[0047] Step B5: Determine the graph Laplacian matrix corresponding to each subgraph based on the degree matrix and the subadjacency matrix.

[0048] For example, the first Adjacency matrix The corresponding degree matrix is ​​represented as subgraph The corresponding Graph Laplace matrix .

[0049] S103. Based on the eigenvalues ​​corresponding to the graph Laplacian matrix, the preset spectral truncation threshold, and the pulse signal of each node in the subgraph, the Chebyshev approximation algorithm is used to determine the filter and the local energy of each node in the subgraph.

[0050] Each subgraph contains selected and unselected nodes. The pulse signals for each node in each subgraph include pulse signals for selected and unselected nodes. Local energy includes the first local energy for selected nodes and the second local energy for unselected nodes.

[0051] Specifically, based on the eigenvalues ​​corresponding to the graph Laplacian matrix, the preset spectral truncation threshold, and the impulse signals of nodes in each subgraph, the Chebyshev approximation algorithm is used to determine the filter and the local energy of nodes in each subgraph, including:

[0052] Step C1: Determine the kernel function based on the eigenvalues ​​corresponding to the graph Laplacian matrix and the preset spectral truncation threshold.

[0053] The expression for the kernel function is:

[0054] ;

[0055] in, The kernel function is the eigenvalues ​​corresponding to the graph Laplacian matrix. These are the eigenvalues ​​corresponding to the graph Laplacian matrix. This is the preset spectrum truncation threshold.

[0056] This kernel function is typically a monotonically decreasing function. It uses low-pass filtering to preserve the low-frequency components of the signal in the subgraph and can also progressively suppress the high-frequency components of the signal in the subgraph.

[0057] Step C2: Use the Chebyshev approximation algorithm to process the kernel function and obtain the filter.

[0058] The operation of the kernel function is transformed into a cascade of function variables by the Chebyshev approximation algorithm, and a random Gaussian characteristic matrix is ​​generated. The random Gaussian characteristic matrix is ​​then transformed by the Chebyshev approximation algorithm to obtain the filter.

[0059] Step C3: Determine the local energy of each node in each subgraph based on the filter and the pulse signal of each node in the subgraph.

[0060] The expression for the local energy of a node in each subgraph is:

[0061] ;

[0062] in, For each node in the subgraph The local energy of each node in the subgraph. The square of the L2 norm of the localization filter response, For each node in the subgraph The localization filter response, The filter corresponding to the kernel function. For each node in the subgraph Local energy.

[0063] S104. Based on the filter, local energy, and preset regularization parameters, sample multiple target nodes in each subgraph node, and determine the sampling signal set corresponding to the three-dimensional point cloud data based on the multiple target nodes.

[0064] The object of sampling in this invention is a large-scale three-dimensional point cloud data. The sampling goal is to effectively sample multiple target nodes from the nodes of each sub-graph, reconstruct the color attributes of the three-dimensional point cloud data from the sampled signal set determined by the multiple target nodes.

[0065] Specifically, based on filters, local energy, and preset regularization parameters, multiple target nodes are sampled in each subgraph, and the sampled signal set corresponding to the 3D point cloud data is determined based on these multiple target nodes, including:

[0066] Step D1: Determine the first localization filter response of the selected node based on the pulse signals of the filter and the selected node.

[0067] The expression for the first localization filter response of the selected node is:

[0068] ;

[0069] in, Selected nodes The first localization filter response, For filters, Selected nodes The pulse signal.

[0070] The first node selected from the selected nodes is the one with the largest energy norm.

[0071] Step D2: Determine the second localization filter response of the unselected node based on the pulse signals of the filter and the unselected node.

[0072] ;

[0073] in, Unselected node The second localization filter response, For filters, Unselected node The pulse signal.

[0074] Step D3: Based on the first localization filter response, the second localization filter response, the first local energy, and the preset regularization parameters, determine the weighted cumulative amount of correlation squared and the cumulative amount of energy coupling between the selected and unselected nodes.

[0075] Specifically, the expression for the correlation squared weighted cumulant is:

[0076] ;

[0077] in, Selected nodes and unselected nodes The squared weighted cumulative correlation between them Selected nodes The first localization filter response, Selected nodes The transpose of the first localization filter response, Unselected node The second localization filter response, Selected nodes The first local energy, i.e., the selected node. The square of the L2 norm of the first localization filter response, To preset regularization parameters, This is the sampled signal set corresponding to the 3D point cloud data. Preset regularization parameters. Used to enhance numerical stability.

[0078] The expression for the cumulative energy coupling is:

[0079] ;

[0080] in, Selected nodes and unselected nodes The cumulative amount of energy coupling between them Selected nodes The first localization filter response, Selected nodes The transpose of the first localization filter response, Unselected node The second localization filter response, Selected nodes The first local energy, To preset regularization parameters, This is the set of sampled signals corresponding to the 3D point cloud data.

[0081] Step D4: Based on the correlation squared weighted cumulative amount, energy coupling cumulative amount, second local energy and preset regularization parameters, sample multiple target nodes in the nodes of each subgraph to obtain all target nodes of all subgraphs.

[0082] Among them, multiple target nodes are nodes from the selected nodes.

[0083] Sample multiple target nodes corresponding to each subgraph from the nodes of each subgraph, and determine the set of multiple target nodes corresponding to each subgraph as the local sampling point set. .

[0084] Specifically, the expression for multiple target nodes is as follows:

[0085] ;

[0086] in, For multiple target nodes, Unselected node The second localization filter response, Unselected node The second local energy, i.e., the unselected node. The square of the L2 norm of the second localization filter response, To preset regularization parameters, Selected nodes and unselected nodes The squared weighted cumulative correlation between them Selected nodes and unselected nodes The cumulative amount of energy coupling between them This is the set of unselected sampled signals formed by the unselected nodes.

[0087] The algorithm corresponding to the expressions for the multiple target nodes in step D4 can be called a greedy algorithm. This greedy algorithm maximizes the remaining independent energy and minimizes correlation interference. Maximizing the remaining independent energy corresponds to the numerator of the expressions for the multiple target nodes. Minimizing correlation interference corresponds to the denominator of the expression for multiple target nodes. .

[0088] Step D5: Merge all target nodes of all subgraphs to obtain the sampling signal set corresponding to the 3D point cloud data.

[0089] The operations described in steps D1 to D5 above can be called Kernel-based Sampling using Localized Graph Filter Impulse Responses (KSLF). The core of the KSLF algorithm is to optimize nodes using filter responses to select multiple target nodes.

[0090] During sampling, multiple target nodes are sampled in each subgraph node, which can meet the global sampling rate limit and retain important information as much as possible, while significantly reducing computational overhead.

[0091] To evaluate the effectiveness of the sampling and the representativeness of the sampled point cloud, it is necessary to reconstruct the sampled point cloud.

[0092] The effect of point cloud reconstruction is visualized by reconstructing the RGB color signal. After step S104, the graph-based 3D point cloud fast sampling method further includes: using the graph Laplacian regularization expression to reconstruct the sampled signal set to obtain the target reconstructed point cloud signal.

[0093] Among them, the target reconstructed point cloud signal is used for high-quality modeling, improving the accuracy of shape-based object recognition and tracking algorithms, etc.

[0094] Specifically, using the expression of graph Laplacian regularization, the sampled signal set is reconstructed to obtain the target reconstructed point cloud signal, including:

[0095] Step E1: Calculate the sampled signal set using the expression of graph Laplace regularization to obtain the reconstructed signal set.

[0096] The expression for the reconstructed signal set is: ,in, To reconstruct the signal set, The sampled signal set corresponding to the 3D point cloud data. Let Laplace matrix be the distance between unsampled nodes in the sampled signal set. Let be the Laplace matrix between unsampled nodes and sampled nodes in the sampled signal set.

[0097] Step E2: Merge the reconstructed signal set and the sampled signal set to obtain the target reconstructed point cloud signal.

[0098] Based on the above Figure 1As can be seen from the implementation method, this embodiment of the invention utilizes the K-nearest neighbor algorithm to generate a weighted adjacency matrix of 3D point cloud data; according to a preset number, the 3D point cloud data and the weighted adjacency matrix are divided into multiple sub-blocks and corresponding multiple sub-adjacency matrices, and based on the multiple sub-blocks and multiple sub-adjacency matrices, the subgraph corresponding to each sub-block and the graph Laplacian matrix corresponding to each subgraph are determined; based on the eigenvalues ​​corresponding to the graph Laplacian matrix, a preset spectral truncation threshold, and the pulse signals of nodes in each subgraph, the Chebyshev approximation algorithm is used to determine the filter and the local energy of nodes in each subgraph; based on the filter, the local energy, and the preset regularization parameters, multiple target nodes are sampled in the nodes of each subgraph, and the sampled signal set corresponding to the 3D point cloud data is determined based on the multiple target nodes. In this way, 3D point cloud data can be divided into multiple sub-blocks, the weighted adjacency matrix can be divided into multiple sub-adjacency matrices, and multiple subgraphs corresponding to the sub-blocks can be determined based on the divided sub-blocks and sub-adjacency matrices. Each subgraph is sampled, and the sampled signal set can be obtained through the multiple target nodes sampled in each subgraph. This results in low computational complexity, strong robustness, and small error when processing large-scale point cloud data. Furthermore, by sampling multiple target nodes in the nodes of each subgraph based on filters, local energy, and preset regularization parameters, the local energy can be maximized and the correlation interference minimized, thereby achieving high sampling efficiency.

[0099] Based on the same inventive concept, as an implementation of the above-mentioned graph-based three-dimensional point cloud fast sampling method, this embodiment of the invention also provides a graph-based three-dimensional point cloud fast sampling device. Figure 2 This is a structural diagram of the graph-based 3D point cloud rapid sampling device in an embodiment of the present invention. See also... Figure 2 As shown, the graph-based 3D point cloud rapid sampling device may include:

[0100] The generation module 201 is used to generate a weighted adjacency matrix of 3D point cloud data using the K-nearest neighbor algorithm;

[0101] The segmentation module 202 is used to segment the 3D point cloud data and the weighted adjacency matrix according to a preset number, to obtain multiple sub-blocks and corresponding multiple sub-adjacency matrices, and to determine the subgraph corresponding to each sub-block and the graph Laplacian matrix corresponding to each subgraph based on the multiple sub-blocks and multiple sub-adjacency matrices.

[0102] The determination module 203 is used to determine the filter and the local energy of each node in the subgraph based on the eigenvalues ​​corresponding to the graph Laplacian matrix, the preset spectral truncation threshold and the pulse signal of each node in the subgraph, using the Chebyshev approximation algorithm.

[0103] The sampling module 204 is used to sample multiple target nodes in each subgraph according to the filter, local energy and preset regularization parameters, and determine the sampling signal set corresponding to the three-dimensional point cloud data according to the multiple target nodes.

[0104] The graph-based 3D point cloud fast sampling device may further include: a reconstruction module, which samples multiple target nodes in each subgraph according to the kernel function filter, local energy and preset regularization parameters, and after determining the sampling signal set corresponding to the 3D point cloud data according to the multiple target nodes, reconstructs the sampling signal set using the graph Laplacian regularization expression to obtain the target reconstructed point cloud signal.

[0105] The generation module 201 is specifically used to construct an adjacency graph of the 3D point cloud data using the K nearest neighbor algorithm. The adjacency graph includes each point in the 3D point cloud data and the K nearest neighbor points of each point. Based on each point and the K neighbor points, the edge weight between each point and the K neighbor points is calculated. Based on the edge weights, a weighted adjacency matrix is ​​generated.

[0106] The segmentation module 202 is specifically used to segment the 3D point cloud data into multiple sub-blocks according to a preset number. The 3D point cloud data corresponding to each sub-block is different, and each sub-block has a sub-block index. Based on the sub-block index, the weighted adjacency matrix is ​​segmented to obtain the sub-adjacency matrix corresponding to each sub-block. The point set of each sub-block is used as the vertex set, and a subgraph corresponding to each sub-block is constructed based on the vertex set and the sub-adjacency matrix. The degree matrix is ​​calculated based on the elements in the sub-adjacency matrix. Based on the degree matrix and the sub-adjacency matrix, the graph Laplacian matrix corresponding to each subgraph is determined.

[0107] The determination module 203 is specifically used to determine the kernel function based on the eigenvalues ​​corresponding to the graph Laplacian matrix and the preset spectral truncation threshold; to process the kernel function using the Chebyshev approximation algorithm to obtain the filter; and to determine the local energy of the nodes in each subgraph based on the filter and the pulse signal of each node in each subgraph.

[0108] The sampling module 204 is specifically used to: determine the first localized filter response of the selected nodes based on the filter and the pulse signals of the selected nodes; determine the second localized filter response of the unselected nodes based on the filter and the pulse signals of the unselected nodes; determine the correlation square weighted cumulative amount and energy coupling cumulative amount between the selected nodes and the unselected nodes based on the first localized filter response, the second localized filter response, the first local energy, and the preset regularization parameter; sample multiple target nodes in the nodes of each sub-graph based on the correlation square weighted cumulative amount, the energy coupling cumulative amount, the second local energy, and the preset regularization parameter to obtain all target nodes of all sub-graphs, where the multiple target nodes are nodes among the selected nodes; and merge all target nodes of all sub-graphs to obtain the sampling signal set corresponding to the 3D point cloud data, wherein the nodes of each sub-graph include selected nodes and unselected nodes, the pulse signals of the nodes in each sub-graph include the pulse signals of the selected nodes and the pulse signals of the unselected nodes, and the local energy includes the first local energy of the selected nodes and the second local energy of the unselected nodes.

[0109] In sampling module 204, the expression for the correlation squared weighted cumulant is:

[0110] ;

[0111] in, Selected nodes and unselected nodes The squared weighted cumulative correlation between them Selected nodes The first localization filter response, Selected nodes The transpose of the first localization filter response, Unselected node The second localization filter response, Selected nodes The first local energy, To preset regularization parameters, This is the set of sampled signals corresponding to the 3D point cloud data.

[0112] In sampling module 204, the expression for the energy coupling accumulation is:

[0113] ;

[0114] in, Selected nodes and unselected nodes The cumulative amount of energy coupling between them Selected nodes The first localization filter response, Selected nodes The transpose of the first localization filter response, Unselected node The second localization filter response, Selected nodes The first local energy, To preset regularization parameters, This is the set of sampled signals corresponding to the 3D point cloud data.

[0115] In sampling module 204, the expressions for multiple target nodes are as follows:

[0116] ;

[0117] in, For multiple target nodes, Unselected node The second localization filter response, Unselected node The second local energy, To preset regularization parameters, Selected nodes and unselected nodes The squared weighted cumulative correlation between them Selected nodes and unselected nodes The cumulative amount of energy coupling between them This is the set of unselected sampled signals formed by the unselected nodes.

[0118] It should be noted that the above description of the graph-based 3D point cloud rapid sampling device embodiment is similar to the description of the graph-based 3D point cloud rapid sampling method embodiment, and has similar beneficial effects. For any technical details not disclosed in the embodiments of the graph-based 3D point cloud rapid sampling device of this invention, please refer to the description of the graph-based 3D point cloud rapid sampling method embodiment of this invention for understanding.

[0119] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A fast sampling method for 3D point clouds based on graph structure, characterized in that, include: The weighted adjacency matrix of 3D point cloud data is generated using the K-nearest neighbor algorithm. According to a preset number, the 3D point cloud data and the weighted adjacency matrix are divided into multiple sub-blocks and corresponding multiple sub-adjacency matrices, and the subgraph corresponding to each sub-block and the graph Laplacian matrix corresponding to each subgraph are determined based on the multiple sub-blocks and the multiple sub-adjacency matrices. Based on the eigenvalues ​​corresponding to the graph Laplacian matrix, the preset spectral truncation threshold, and the pulse signal of each node in the subgraph, the Chebyshev approximation algorithm is used to determine the filter and the local energy of each node in the subgraph. Based on the filter, the local energy, and the preset regularization parameters, multiple target nodes are sampled in the nodes of each subgraph, and the sampling signal set corresponding to the three-dimensional point cloud data is determined based on the multiple target nodes. The step involves segmenting the 3D point cloud data and the weighted adjacency matrix according to a preset number of segments, obtaining multiple sub-blocks and corresponding sub-adjacency matrices, and determining the subgraph corresponding to each sub-block and the graph Laplacian matrix corresponding to each subgraph based on the multiple sub-blocks and the multiple sub-adjacency matrices, including: The three-dimensional point cloud data is divided into multiple sub-blocks according to the preset number. The three-dimensional point cloud data corresponding to each sub-block is different, and each sub-block has a sub-block index. Based on the sub-block index, the weighted adjacency matrix is ​​divided to obtain the sub-adjacency matrix corresponding to each sub-block; The set of points of each sub-block is used as a vertex set, and a subgraph corresponding to each sub-block is constructed based on the vertex set and the sub-adjacency matrix; Calculate the corresponding degree matrix based on the elements in the sub-adjacency matrix; The graph Laplacian matrix corresponding to each subgraph is determined based on the degree matrix and the subadjacency matrix.

2. The method for fast sampling of 3D point clouds based on graph structure according to claim 1, characterized in that, After sampling multiple target nodes in each subgraph based on the filter, the local energy, and the preset regularization parameters, and determining the sampling signal set corresponding to the 3D point cloud data based on the multiple target nodes, the method further includes: The sampled signal set is reconstructed using a graph Laplacian regularization expression to obtain the target reconstructed point cloud signal.

3. The method for fast sampling of 3D point clouds based on graph structure according to claim 1, characterized in that, The process of generating a weighted adjacency matrix for 3D point cloud data using the K-nearest neighbor algorithm includes: Using the K-nearest neighbor algorithm, an adjacency graph of the 3D point cloud data is constructed. Each node in the adjacency graph corresponds to each point in the 3D point cloud data, and each point corresponds to the K nearest neighboring points. Calculate the edge weight between each point and its K neighboring points; The weighted adjacency matrix is ​​generated based on the edge weights.

4. The method for fast sampling of 3D point clouds based on graph structure according to claim 1, characterized in that, The step of determining the filter and the local energy of each node in the subgraph using the Chebyshev approximation algorithm based on the eigenvalues ​​corresponding to the graph Laplacian matrix, a preset spectral truncation threshold, and the impulse signal of each node in the subgraph includes: The kernel function is determined based on the eigenvalues ​​corresponding to the graph Laplacian matrix and the preset spectral truncation threshold; The kernel function is processed using the Chebyshev approximation algorithm to obtain the filter; The local energy of each node in each subgraph is determined based on the filter and the pulse signal of each node in the subgraph.

5. The method for fast sampling of 3D point clouds based on graph structure according to claim 1, characterized in that, Each subgraph's nodes include selected nodes and unselected nodes, and the pulse signal of each node in each subgraph includes the pulse signal of the selected node and the pulse signal of the unselected node. The local energy includes the first local energy of the selected node and the second local energy of the unselected node. The step of sampling multiple target nodes in each sub-graph node according to the filter, the local energy, and the preset regularization parameters, and determining the sampling signal set corresponding to the 3D point cloud data according to the multiple target nodes, includes: Based on the filter and the pulse signal of the selected node, determine the first localization filter response of the selected node; Based on the filter and the pulse signal of the unselected node, determine the second localization filter response of the unselected node; Based on the first localization filter response, the second localization filter response, the first local energy, and the preset regularization parameter, determine the correlation squared weighted cumulative amount and energy coupling cumulative amount between the selected node and the unselected node; Based on the correlation squared weighted cumulative amount, the energy coupling cumulative amount, the second local energy and the preset regularization parameter, the plurality of target nodes are sampled in the nodes of each subgraph to obtain all target nodes of all subgraphs, and the plurality of target nodes are nodes among the selected nodes; By merging all target nodes of all subgraphs, the sampled signal set corresponding to the 3D point cloud data is obtained.

6. The method for fast sampling of 3D point clouds based on graph structure according to claim 5, characterized in that, The expression for the correlation squared weighted cumulant is: ； in, For the selected node and the unselected nodes The squared weighted cumulative correlation between them For the selected node The first localization filter response, For the selected node The transpose of the first localization filter response, For the unselected node The second localization filter response, For the selected node The first local energy, The preset regularization parameter is... This refers to the set of sampled signals corresponding to the three-dimensional point cloud data.

7. The method for fast sampling of 3D point clouds based on graph structure according to claim 5, characterized in that, The expression for the energy coupling accumulation is: ； in, For the selected node and the unselected nodes The cumulative amount of energy coupling between them For the selected node The first localization filter response, For the selected node The transpose of the first localization filter response, For the unselected node The second localization filter response, For the selected node The first local energy, The preset regularization parameter is... This refers to the set of sampled signals corresponding to the three-dimensional point cloud data.

8. The method for fast sampling of 3D point clouds based on graph structure according to claim 5, characterized in that, The expressions for the multiple target nodes are: ； in, For the multiple target nodes, For the unselected node The second localization filter response, For the unselected node The second local energy, The preset regularization parameter is... For the selected node and the unselected nodes The squared weighted cumulative correlation between them For the selected node and the unselected nodes The cumulative amount of energy coupling between them The unselected sampling signal set consists of the unselected nodes.

9. A rapid sampling device for three-dimensional point clouds based on graph structure, characterized in that, include: The generation module is used to generate a weighted adjacency matrix of 3D point cloud data using the K-nearest neighbor algorithm. The segmentation module is used to segment the 3D point cloud data and the weighted adjacency matrix according to a preset number, respectively, to obtain multiple sub-blocks and corresponding multiple sub-adjacency matrices, and to determine the subgraph corresponding to each sub-block and the graph Laplacian matrix corresponding to each subgraph based on the multiple sub-blocks and the multiple sub-adjacency matrices. The determination module is used to determine the filter and the local energy of the nodes in each subgraph based on the eigenvalues ​​corresponding to the graph Laplacian matrix, the preset spectral truncation threshold, and the pulse signal of the node in each subgraph, using the Chebyshev approximation algorithm. The sampling module is used to sample multiple target nodes in the nodes of each sub-graph according to the filter, the local energy and the preset regularization parameters, and to determine the sampling signal set corresponding to the three-dimensional point cloud data according to the multiple target nodes; The segmentation module is specifically used to segment the 3D point cloud data into multiple sub-blocks according to the preset number, wherein the 3D point cloud data corresponding to each sub-block is different, and each sub-block has a sub-block index; to segment the weighted adjacency matrix according to the sub-block index to obtain a sub-adjacency matrix corresponding to each sub-block; to use the point set of each sub-block as a vertex set, and to construct a subgraph corresponding to each sub-block according to the vertex set and the sub-adjacency matrix; to calculate the corresponding degree matrix according to the elements in the sub-adjacency matrix; and to determine the graph Laplacian matrix corresponding to each subgraph according to the degree matrix and the sub-adjacency matrix.

Images