Modeling information and model relationship extraction method and system based on point cloud data

Abstract

This application provides a method and system for extracting modeling information and model relationships based on point cloud data. It uses an improved k-means clustering algorithm based on feature fusion to segment point cloud data, obtaining multiple model point cloud subsets. This method can resist interference from complex scenes such as occlusion and strong light. Modeling information is extracted from each model point cloud subset. During the extraction process, key modeling information is identified and enhanced. Feature fusion and feature enhancement effectively solve the extraction bias problems caused by noise and occlusion, ensuring the accuracy of the modeling information. An association matrix between models is established based on the model point cloud subsets and modeling information. Machine learning algorithms are used to analyze this matrix, accurately identifying the position, connection, and dependency relationships between models, ensuring the completeness and rationality of scene modeling. Furthermore, it eliminates the need for point-by-point traversal and single feature matching, enabling efficient processing of large amounts of point cloud data, thereby improving extraction efficiency, accuracy, and reliability.

Description

Technical Field

[0001] This invention relates to the field of point cloud data processing technology, and in particular to a method and system for extracting modeling information and model relationships based on point cloud data. Background Technology

[0002] With the rapid development of technologies such as laser scanning and Simultaneous Localization and Mapping (SLAM), point cloud data has become a core data source for 3D modeling, widely used in fields such as power operation and maintenance (e.g., point clouds of transmission lines and substation equipment), smart cities (e.g., point clouds of urban buildings and terrain), and industrial manufacturing (e.g., point clouds of production equipment). Point cloud data is characterized by its large volume (the number of point clouds in a single scene can reach tens of millions or even hundreds of millions), complex structure, high noise interference, and disordered data. Existing technologies struggle to achieve ideal results when extracting modeling information from point clouds and identifying the relationships between models.

[0003] Traditional point cloud modeling information extraction methods often employ point-by-point traversal and single-feature matching. For large-scale point clouds with tens of millions of points, the processing time is long (usually exceeding several hours), which cannot meet the efficiency requirements of scenarios such as real-time modeling and rapid operation and maintenance. Furthermore, point cloud data contains problems such as noise, occlusion, and missing data. Existing methods have limited ability to identify features, which can easily lead to deviations in the extraction of modeling information (such as model size, geometric features, and texture information), resulting in low modeling accuracy and an inability to accurately reproduce the real scene. In addition, existing technologies mostly focus on extracting modeling information from a single model and lack the ability to identify the relationships between multiple models. They cannot accurately determine the correlation logic between different models in the point cloud, affecting the completeness and rationality of the overall scene modeling. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for extracting modeling information and model relationships based on point cloud data, which can improve extraction efficiency, accuracy and reliability.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A method for extracting modeling information and model relationships based on point cloud data includes: Receive point cloud data; The point cloud data is segmented using an improved k-means clustering algorithm based on feature fusion to obtain multiple model point cloud subsets; Extract modeling information from each subset of the model point cloud. During the extraction process, identify key modeling information and enhance the key modeling information. Establish an association matrix between the models based on the model point cloud subset and the modeling information; The correlation matrix is ​​analyzed using machine learning algorithms to obtain the relationships between the models, including positional relationships, connectivity relationships, and dependency relationships.

[0006] To solve the above-mentioned technical problems, another technical solution adopted by the present invention is as follows: A system for extracting modeling information and model relationships based on point cloud data includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the aforementioned method for extracting modeling information and model relationships based on point cloud data.

[0007] The beneficial effects of this invention are as follows: It uses an improved k-means clustering algorithm based on feature fusion to segment point cloud data, obtaining multiple model point cloud subsets. This can resist interference from complex scenes such as occlusion and strong light. It extracts modeling information from each model point cloud subset. During the extraction process, key modeling information is identified and enhanced. Feature fusion and feature enhancement effectively solve the extraction bias problems caused by noise and occlusion, ensuring the accuracy of the modeling information. Based on the model point cloud subsets and modeling information, an association matrix is ​​established between the models. Machine learning algorithms are used to analyze the association matrix to obtain the relationships between the models, accurately identifying the positions, connections, and dependencies between models, ensuring the completeness and rationality of scene modeling. Furthermore, it eliminates the need for point-by-point traversal and single feature matching, enabling efficient processing of large amounts of point cloud data, thereby improving extraction efficiency, accuracy, and reliability. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating a method for extracting modeling information and model relationships based on point cloud data, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of a system for extracting modeling information and model relationships based on point cloud data, according to an embodiment of the present invention. Detailed Implementation

[0009] Definitions:

[0010] To explain in detail the technical content, objectives, and effects of the present invention, the following description is provided in conjunction with the embodiments and accompanying drawings.

[0011] In existing technologies, when faced with point cloud data of tens of millions or even hundreds of millions of points acquired by technologies such as laser scanning and SLAM, the traditional point-by-point traversal and single feature matching methods are computationally inefficient, with processing times typically reaching several hours. At the same time, the inherent problems of noise, occlusion, and data loss in point clouds lead to insufficient robustness of feature recognition, and the extraction of modeling information is prone to deviation. In addition, existing methods lack the ability to identify the correlation logic between multiple models, making it difficult to ensure the integrity and rationality of the overall scene modeling.

[0012] To address at least some of the aforementioned issues, an improved k-means clustering algorithm based on feature fusion is used to segment point cloud data. Modeling information is extracted from each subset of point clouds obtained from the segmentation. During the extraction process, key modeling information is identified and enhanced. An association matrix is ​​established between the models based on the model point cloud subsets and the modeling information. Machine learning algorithms are used to analyze the association matrix to obtain the positional, connectivity, and dependency relationships between the models. This approach allows for point cloud data segmentation using the improved k-means clustering algorithm based on feature fusion, resisting interference from complex scenes such as occlusion and strong light. Feature fusion and feature enhancement effectively solve extraction bias problems caused by noise and occlusion, ensuring the accuracy of modeling information. Machine learning algorithms analyze the association matrix to obtain the relationships between models, accurately identifying their positional, connectivity, and dependency relationships, ensuring the completeness and rationality of scene modeling. Furthermore, it eliminates the need for point-by-point traversal and single-feature matching, enabling efficient processing of large amounts of point cloud data, thereby improving extraction efficiency, accuracy, and reliability.

[0013] The following describes in detail a method for extracting modeling information and model relationships based on point cloud data, wherein the point cloud data is point cloud data of power equipment, applicable to scenarios such as power inspection, risk warning, and three-dimensional digital ledger management. Please refer to [the relevant documentation]. Figure 1 The method 100 includes steps 101 to 105: Step 101: Receive point cloud data.

[0014] Specifically, it receives point cloud data from power equipment from multiple data sources, such as laser scanning equipment and SLAM equipment.

[0015] Step 102: The point cloud data is segmented using an improved k-means clustering algorithm based on feature fusion to obtain multiple model point cloud subsets.

[0016] Specifically, each subset of point clouds corresponds to an independent model.

[0017] Step 103: Extract modeling information for each subset of the model point cloud. During the extraction process, identify key modeling information and enhance the key modeling information.

[0018] Step 104: Establish the correlation matrix between the models based on the model point cloud subset and the modeling information.

[0019] Specifically, based on the geometric modeling information, the spatial coordinates, normal vector angle, and distance parameters of each subset of the model point cloud are calculated. Combined with the attribute modeling information, the feature similarity between each model is represented by matrix elements, and an association matrix between each model is established.

[0020] Step 105: Analyze the correlation matrix using machine learning algorithms to obtain the relationships between the models, including positional relationships, connection relationships, and dependency relationships.

[0021] Specifically, the feature similarity data in the association matrix is ​​input into a machine learning classification algorithm (such as the random forest algorithm) for analysis, and the relationship categories between the models are output.

[0022] In one embodiment of the present invention, before step 102, the method further includes: The point cloud data is subjected to format standardization processing to obtain standardized point cloud data; An adaptive denoising algorithm is applied to the standardized point cloud data, combining point cloud density and local neighborhood features, to identify and remove noise points in the standardized point cloud data, thus obtaining denoised point cloud data.

[0023] Specifically, the point cloud data is converted into a preset format to obtain standardized point cloud data. An adaptive denoising algorithm is then applied to the standardized point cloud data to calculate the number of neighboring points and the distance variance of each point within a preset neighborhood radius. A noise threshold is set to remove noise points (such as isolated points and outliers) and retain key feature points (such as transformer winding edge points and circuit breaker connection points).

[0024] For example, point cloud data in PLY, PCD, and / or LAS formats can be uniformly converted to PCD format.

[0025] The preset neighborhood radius is 5mm.

[0026] As described above, data format differences can be eliminated, noise points can be effectively filtered out, anti-interference ability can be improved, and processing efficiency can be increased through format standardization and adaptive denoising algorithms.

[0027] In one embodiment of the present invention, step 102 includes steps 1021 to 1024: Step 1021: Determine the initial cluster centers.

[0028] Specifically, the local density of each point cloud data in the point cloud data is calculated, where the local density is the number of point clouds within a preset radius centered on the point cloud data. The local densities are sorted in descending order, and the point cloud data corresponding to the top k local densities are selected as candidate centers. Calculate the Euclidean distance between the candidate centers; Candidate centers whose Euclidean distance is less than the second preset threshold are removed, and one point cloud data is sequentially selected from the remaining point cloud data corresponding to the local density that is not in the first k in the sorting to fill the candidate center, thus obtaining the final candidate center; The final candidate centers are fine-tuned based on semantic priors (such as device size and typical geometry) of the point cloud data to obtain initial cluster centers. For example, for devices with known dimensions such as transformers and insulators, the cluster centers are constrained to be near their geometric centers to improve the specificity of the initial clustering.

[0029] The second preset threshold is dynamically adjusted according to the point cloud density, and is usually 2-3 times the average neighborhood distance of the point cloud.

[0030] As described above, it is important to ensure that the initial cluster centers are evenly distributed and far away from areas with dense noise.

[0031] Step 1022: Calculate the clustering distance between each point cloud data point and the initial cluster center based on the geometric, texture, and semantic features of the point cloud data. Specifically: ; In the formula, Point i With point j Cluster distance, , , All represent weighting coefficients. Point i With point j The geometric distance (i.e., the Euclidean distance in three-dimensional coordinates) is calculated based on geometric features. Point i With point j The texture distance (i.e., the absolute difference between reflection intensity and grayscale value) is calculated based on texture features. Point i With point j The semantic distance (i.e., similarity based on device type labels, with 0 for the same type and 1 for different types) is calculated based on semantic features.

[0032] The weighting coefficients can be dynamically adjusted according to the point cloud type (e.g., reducing semantic weight for terrain point clouds and increasing semantic weight for industrial equipment point clouds) to ensure the synergistic effect of features from different dimensions and improve segmentation accuracy. In one optional implementation, =0.5、 =0.3、 =0.2.

[0033] Step 1023: Assign each point of cloud data to the cluster containing the initial cluster center that is closest to the clustering distance.

[0034] Step 1024: If the variance of the geometric features of all point cloud data within a cluster is less than or equal to the first preset threshold and the number of iterations reaches the preset number, then stop the iteration and treat each cluster as a subset of the model point cloud. Otherwise, recalculate the cluster center of each cluster and return to step 1022.

[0035] For example, if the variance of the angle between the normal vectors of all point cloud data within a cluster is less than or equal to 5 and the number of iterations reaches 10, then stop the iteration and treat each cluster as a subset of the model point cloud; otherwise, recalculate the cluster center of each cluster and return to step 1022.

[0036] As described above, the traditional k-means clustering algorithm randomly selects initial cluster centers, which easily leads to local optima and unstable segmentation results. The above-mentioned density peak combined with distance screening strategy determines the initial cluster centers, improving the reliability of the segmentation results. The traditional k-means clustering algorithm only calculates the cluster distance based on the three-dimensional coordinates (geometric features) of the point cloud, which cannot distinguish between geometrically similar models but different semantics or textures (such as a metal circuit breaker and a concrete base). This invention integrates geometric features, texture features, and semantic features to construct a weighted distance, ensuring the synergistic effect of features of different dimensions. In addition, the traditional k-means clustering algorithm uses "cluster centers no longer changing" as the termination condition, which is prone to premature convergence due to noise. This invention adds a judgment on the consistency threshold of point cloud features within a cluster to ensure that the point clouds within a cluster belong to a unified model.

[0037] In one embodiment of the present invention, before step 103, the method further includes: The model point cloud subset is optimized using a region growing mechanism to obtain an optimized model point cloud subset, including steps (1) to (3): (1) Pretreatment: Establishing seed points and growth criteria Seed point selection: Each cluster center after segmentation by the improved k-means clustering algorithm based on feature fusion is used as the initial seed point. At the same time, the point with the highest feature consistency within the cluster (such as the point with the smallest geometric feature variance) is selected as the auxiliary seed point to avoid growth deviation caused by noise due to a single seed point. Growth Criterion Definition (Multiple Constraints): Define geometric consistency: The Euclidean distance between the growth point and the seed point is less than or equal to the growth threshold (dynamically adjusted according to the model size, such as 5mm for the device model and 10mm for the background model); the angle between the normal vectors of the growth point and the seed point is less than or equal to 8° (to ensure surface continuity). Define texture consistency: The difference in reflection intensity between the growth point and the seed point is less than or equal to 10% (applicable to laser scanning point clouds); Define semantic consistency: The semantic label of the point to be grown (such as device type) is consistent with the seed point (based on the prior knowledge of the power scenario); Define connectivity constraints: The point to be grown must be directly connected to the already grown region (to avoid cross-model growth).

[0038] (2) Growth process: Adaptive region expansion and conflict resolution Layer-by-layer growth: Starting from the seed point, traverse its neighboring points (based on fast retrieval using a KD tree). Points that meet the growth criteria are added to the current region, and the seed point is used as a new seed point to continue expanding until no points that meet the criteria can be grown. Undersegmentation repair: To address the undersegmentation problem where the same model is split into multiple clusters after segmentation by the improved k-means clustering algorithm based on feature fusion, the feature similarity of adjacent clusters is calculated (weighted similarity of geometric, texture, and semantic features greater than or equal to 90%). If the similarity threshold is met, adjacent clusters are merged into one region, for example, merging multiple scattered clusters of transformer windings into a complete winding point cloud. Oversegmentation Removal: To address the oversegmentation problem caused by excessive subdivision of individual models into small clusters after segmentation by the improved k-means clustering algorithm based on feature fusion, a minimum cluster size threshold is set (e.g., the number of cluster points is less than or equal to 0.5% of the total number of point clouds). The excessively small clusters are then matched with the adjacent largest clusters for feature matching. If the matching degree is greater than or equal to 85%, they are merged; otherwise, they are judged as noise points and removed. Boundary optimization: Smooth the boundary points of the grown region and remove isolated points at the boundary (with fewer than or equal to 3 neighboring points) to ensure that the boundaries of the model point cloud subset are continuous and complete.

[0039] (3) Post-processing: Regional verification and adjustment Integrity verification: Calculate the matching degree between the geometry of each grown region and the preset model library (such as the typical geometry of transformers and circuit breakers). If the matching degree is greater than or equal to 80%, it is determined to be a valid model subset; if the matching degree is insufficient, the growth criteria are readjusted (such as increasing the growth threshold), and the growth is repeated. Redundant region removal: If the feature similarity of multiple growth regions is greater than or equal to 95% and the spatial overlap rate is greater than or equal to 30%, then the region with the most points and the highest feature consistency is retained, and the remaining redundant regions are removed. Final output: After the above optimization, each model point cloud subset corresponds to a complete independent model, with no undersegmentation or oversegmentation errors, and the segmentation accuracy is greater than or equal to 98%.

[0040] As described above, the region growing mechanism can eliminate the undersegmentation and oversegmentation errors after clustering by the improved k-means clustering algorithm, ensuring that each model point cloud subset strictly corresponds to a complete and independent model.

[0041] In one embodiment of the present invention, step 103 includes steps 1031 to 1034: Step 1031: Use a point cloud fitting algorithm to extract geometric modeling information from each optimized model point cloud subset. The geometric modeling information includes the model's size parameters, shape parameters, and key feature points.

[0042] The dimensional parameters include length, width, height, and radius, the shape parameters include geometric shapes such as planes, cylinders, and spheres, and the key feature points include device connection points and vertices.

[0043] Specifically, after normalizing the coordinates of each optimized model point cloud subset, the PCA algorithm is used to extract the three-dimensional extreme points to calculate the size parameters for each optimized model point cloud subset, the RANSAC algorithm is used to fit the geometric equation to obtain the shape parameters, and the SIFT-3D algorithm is used to identify curvature change points as key feature points, thereby achieving accurate extraction of geometric modeling information.

[0044] Step 1032: Use a semantic recognition algorithm to extract attribute modeling information from each optimized model point cloud subset. The attribute modeling information includes the model type, material, and purpose.

[0045] Specifically, a pre-defined power equipment attribute prior library (including typical features and corresponding attributes) is obtained. The extracted geometric modeling information is then matched with the pre-defined power equipment attribute prior library for similarity. If the similarity is greater than or equal to 85%, the attribute is directly assigned. If the matching is insufficient, manual annotation is used to supplement the information. The model type, material, and purpose are output through a semantic recognition algorithm to obtain attribute modeling information.

[0046] Step 1033: Use a texture mapping algorithm to extract texture modeling information from each optimized model point cloud subset. The texture modeling information includes texture features and color information of the model surface.

[0047] Specifically, if each optimized model point cloud subset is a laser point cloud, the native reflection intensity is directly extracted and its distribution is statistically analyzed as texture modeling information; if each optimized model point cloud subset is an image fusion point cloud, the correspondence between three-dimensional coordinates and image pixels is established through camera calibration, RGB (red, green, blue) color values ​​and texture grayscale distribution are mapped and extracted, a texture map is generated, and texture modeling information is obtained.

[0048] Step 1034: During the extraction process, key modeling information is determined and the key modeling information is enhanced.

[0049] Specifically, the extracted modeling information is first compared with the corresponding measured values, and features whose deviation exceeds a threshold are marked as key modeling information. Then, the key modeling information is given high weight. If the key modeling information is geometric modeling information, it is optimized by fitting neighborhood points. If the key modeling information is texture modeling information, it is smoothed by Gaussian filtering. If the key modeling information is attribute modeling information, no enhancement operation is performed, and the subsequent steps are executed directly. Finally, the enhanced key modeling information is iteratively substituted into the extraction process of steps 1031 to 1033, and the parameters are adjusted until the deviation is less than or equal to the preset accuracy (such as geometric dimensions less than or equal to 1%).

[0050] As described above, extracting geometric modeling information, attribute modeling information, and texture modeling information separately makes the modeling information more comprehensive and reliable. Further strengthening the key modeling information can reduce extraction errors and ensure more accurate identification of model relationships.

[0051] In one embodiment of the present invention, after step 105, the method further includes: Cross-validation is performed on the relationships between the models to eliminate contradictory or unreasonable relationships, resulting in the validated relationships between the models.

[0052] As described above, by cross-validating the relationships between the models and eliminating contradictory or unreasonable relationships, the accuracy of model relationship identification can be further improved.

[0053] Please refer to Figure 2 The present invention also provides a system 200 for extracting modeling information and model relationships based on point cloud data, including a memory 201, a processor 202, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the various steps of the method for extracting modeling information and model relationships based on point cloud data as described above.

[0054] In summary, the method and system for extracting modeling information and model relationships based on point cloud data described above shorten the processing time for large-scale point cloud data through adaptive denoising, feature fusion segmentation, and parallel processing technologies, meeting the efficiency requirements of scenarios such as real-time modeling and rapid operation and maintenance. The use of feature fusion combined with feature enhancement effectively improves the accuracy of modeling information extraction, attribute recognition accuracy, and segmentation accuracy, effectively solving the extraction deviation problem caused by noise and occlusion, ensuring the accuracy of modeling information. Furthermore, the combination of association matrix and machine learning accurately identifies the position, connection, and dependency relationships between models, improving recognition accuracy and ensuring the integrity and rationality of scene modeling. The adaptive denoising algorithm effectively filters noise points, and the multi-feature fusion strategy can resist interference from complex scenes such as occlusion and strong light. It also supports point cloud data of different formats, types, and densities, adapting to power operation and maintenance scenarios. Simultaneously, it can quickly provide accurate modeling information and model relationship data for 3D modeling, scene simulation, and equipment operation and maintenance, significantly reducing manual processing costs and improving work efficiency, showing broad application prospects in fields such as digital twins and power operation and maintenance.

[0055] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent modifications made based on the content of the present invention specification and drawings, or direct or indirect applications in related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A modeling information and model relationship extraction method based on point cloud data, characterized in that, include: Receive point cloud data; The point cloud data is segmented using an improved k-means clustering algorithm based on feature fusion to obtain multiple model point cloud subsets; Extract modeling information from each subset of the model point cloud. During the extraction process, identify key modeling information and enhance the key modeling information. Establish an association matrix between the models based on the model point cloud subset and the modeling information; The correlation matrix is ​​analyzed using machine learning algorithms to obtain the relationships between the models, including positional relationships, connectivity relationships, and dependency relationships.

2. The method of claim 1, wherein, The point cloud data was segmented using an improved k-means clustering algorithm based on feature fusion to obtain multiple model point cloud subsets, including: Determine the initial cluster centers; Based on the geometric, texture, and semantic features of the point cloud data, calculate the clustering distance between each point cloud data and the initial cluster center; Each point of cloud data is assigned to the cluster containing the initial cluster center that is closest to the clustering distance; If the variance of the geometric features of all point cloud data within a cluster is less than or equal to a first preset threshold, and the number of iterations reaches a preset number, then the iteration stops, and each cluster is treated as a subset of the model point cloud. Otherwise, the cluster center of each cluster is recalculated, and the process returns to the step of calculating the cluster distance between each point cloud data and the initial cluster center based on the geometric, texture, and semantic features of the point cloud data.

3. The method of claim 2, wherein, Determining the initial cluster centers includes: Calculate the local density of each point cloud data in the point cloud data, where the local density is the number of point clouds within a preset radius centered on the point cloud data. The local densities are sorted in descending order, and the point cloud data corresponding to the top k local densities are selected as candidate centers. Calculate the Euclidean distance between the candidate centers; Candidate centers whose Euclidean distance is less than the second preset threshold are removed, and one point cloud data is sequentially selected from the remaining point cloud data corresponding to the local density that is not in the first k in the sorting to fill the candidate center, thus obtaining the final candidate center; The final candidate centers are fine-tuned based on the semantic priors of the point cloud data to obtain the initial cluster centers.

4. The method of claim 2, wherein, Based on the geometric, texture, and semantic features of the point cloud data, the clustering distance between each point cloud data point and the initial cluster center is calculated, specifically as follows: ； wherein representing a point i and a point j , , , , representing a point i and a point j , representing a point i and a point j , representing a point i and a point j , 5. The method of claim 1, wherein, Before extracting the modeling information for each subset of the model point cloud, the process also includes: The model point cloud subset is optimized using a region growing mechanism to obtain an optimized model point cloud subset.

6. The method of claim 5, wherein, Extracting modeling information for each subset of the model point cloud includes: Geometric modeling information is extracted from each optimized model point cloud subset using a point cloud fitting algorithm. The geometric modeling information includes the model's size parameters, shape parameters, and key feature points. The semantic recognition algorithm is used to extract attribute modeling information from each optimized model point cloud subset, the attribute modeling information including the model type, material and purpose; Texture modeling information, including texture features and color information of the model surface, is extracted from each optimized model point cloud subset using a texture mapping algorithm.

7. The method according to claim 6, characterized in that, Establishing the correlation matrix between models based on the model point cloud subset and the modeling information includes: Based on the geometric modeling information, calculate the spatial coordinates, normal vector angle, and distance parameters of each subset of the model point cloud. Combined with the attribute modeling information, use matrix elements to represent the feature similarity between models to establish the association matrix between models.

8. The method according to claim 1, characterized in that, After analyzing the correlation matrix using machine learning algorithms to obtain the relationships between the models, the process also includes: Cross-validation is performed on the relationships between the models to eliminate contradictory or unreasonable relationships, resulting in the validated relationships between the models.

9. The method according to claim 1, characterized in that, Before segmenting the point cloud data using an improved k-means clustering algorithm based on feature fusion to obtain multiple model point cloud subsets, the following steps are also included: The point cloud data is subjected to format standardization processing to obtain standardized point cloud data; An adaptive denoising algorithm is applied to the standardized point cloud data, combining point cloud density and local neighborhood features, to identify and remove noise points in the standardized point cloud data, thus obtaining denoised point cloud data.

10. A system for extracting modeling information and model relationships based on point cloud data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements each step of the method for extracting modeling information and model relationships based on point cloud data according to any one of claims 1 to 9.

Images