Power engineering building deformation monitoring method and system based on ground point cloud data

Abstract

The application belongs to the technical field of power engineering building deformation monitoring, and provides a power engineering building deformation monitoring method and system based on ground point cloud data. Firstly, when performing semantic segmentation, the geometric structure information of the building is modeled in the local feature aggregation stage. Then, based on multi-scale neighborhood search and normal vector curvature weighting mechanism, feature points are extracted from the building point cloud after semantic segmentation. Based on the extracted feature points, a straight line fitting algorithm is used to extract the feature lines of the building. Finally, the feature points and feature lines of the point clouds in different periods are matched to construct the feature correspondence relationship. The deformation monitoring is realized through the spatial position difference of the point cloud features in different periods. The power engineering building point cloud can be accurately extracted from the original mixed point cloud, and the ground, vegetation and other interference points in the background can be removed, thereby improving the detection accuracy at the corners, edges and structure transitions of the building.

Description

Technical Field

[0001] This invention belongs to the field of power engineering building deformation monitoring technology, and particularly relates to a method and system for power engineering building deformation monitoring based on ground point cloud data. Background Technology

[0002] During long-term operation, power engineering structures are subject to settlement, tilting, or structural deformation due to factors such as load, environment, and geological conditions. High-precision deformation monitoring of power engineering structures is of great significance to ensure the safe and stable operation of the power system.

[0003] The complex structure of power engineering buildings makes it difficult to accurately extract point cloud data from power engineering buildings when implementing point cloud-based deformation monitoring technology. This is due to the large amount of point cloud data and the presence of noise. Traditional filtering and segmentation methods are insufficient to accurately extract point clouds from power engineering buildings, resulting in insufficient accuracy in subsequent analysis. The extraction methods for feature points and feature lines are not very sensitive to complex building structures, which can easily lead to missing features or false detections. In multi-period point cloud comparison analysis, the registration method is easily affected by noise and external interference, resulting in errors in deformation detection results, especially at building corners, edges, and structural transitions. Summary of the Invention

[0004] To address the aforementioned problems, this invention proposes a method and system for monitoring the deformation of power engineering buildings based on ground point cloud data. This invention can accurately extract the point cloud of power engineering buildings from the original mixed point cloud, remove ground, vegetation and other interference points in the background. On this basis, it improves the feature sensitivity of building components at different scales by using a multi-scale neighborhood search and normal vector curvature weighting mechanism, thereby improving the detection accuracy at building corners, edges and structural transitions.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solution: In a first aspect, the present invention provides a method for monitoring deformation of power engineering buildings based on ground point cloud data, comprising: Acquire point cloud data of power engineering construction at different times; Semantic segmentation is performed on the acquired point cloud data; during semantic segmentation, the geometric structure information of the buildings is modeled in a focused manner during the local feature aggregation stage. Based on multi-scale neighborhood search and normal vector curvature weighting mechanism, feature points are extracted from the semantically segmented building point cloud. Based on the extracted feature points, a straight line fitting algorithm is used to extract the feature lines of the building; The feature points and feature lines of point clouds from different periods are matched to construct feature correspondences; deformation monitoring is achieved by using the spatial location differences of point cloud features from different periods.

[0006] Furthermore, the acquired point cloud data is subjected to denoising, downsampling, and coordinate system unification processing.

[0007] Furthermore, semantic segmentation of the acquired point cloud data includes: using the RAA-RandLA-Net model for semantic segmentation, extracting point clouds of power structures, and performing boundary optimization.

[0008] Furthermore, the RAA-RandLA-Net model introduces the RAA-LFA module: in the local feature aggregation stage, a residual attention mechanism is added, combining channel attention and spatial attention to focus on modeling the geometric structure information of buildings; by setting different neighborhood scales, multi-level extraction of local geometric features of point clouds is performed; and in the decoding stage, a boundary-aware loss function is added to enhance the model's segmentation accuracy at building edges.

[0009] Furthermore, feature extraction of the semantically segmented building point cloud includes: feature point extraction using the Harris3D operator.

[0010] Furthermore, during the feature point detection process, the covariance matrix is ​​calculated using multiple neighborhood scales, and points with significant responses at multiple scales are selected through cross-scale persistence index to ensure that the extracted feature points include both local details and overall structure. When constructing the covariance matrix, normal consistency weights and curvature weights are introduced. If the difference between the normal of a neighboring point and the center point is greater than a preset value, its contribution is weakened. Regions with higher curvature are given greater weight.

[0011] Furthermore, extracting the feature lines of the building includes: based on the extracted feature points, using the RANSAC straight line fitting algorithm to extract the feature lines of the building.

[0012] Secondly, the present invention also provides a power engineering building deformation monitoring system based on ground point cloud data, comprising: The data acquisition module is configured to acquire point cloud data of power engineering construction at different times; The semantic segmentation module is configured to: perform semantic segmentation on the acquired point cloud data; and during semantic segmentation, focus on modeling the geometric structure information of buildings in the local feature aggregation stage. The feature point extraction module is configured to extract feature points from the semantically segmented building point cloud based on multi-scale neighborhood search and normal vector curvature weighting mechanism. The feature line extraction module is configured to extract the feature lines of the building based on the extracted feature points using a straight line fitting algorithm. The monitoring module is configured to: match feature points and feature lines of point clouds at different times to construct feature correspondences; and monitor deformation by using the spatial position differences of point cloud features at different times.

[0013] Thirdly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the power engineering building deformation monitoring method based on ground point cloud data described in the first aspect.

[0014] Fourthly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the program to implement the steps of the power engineering building deformation monitoring method based on ground point cloud data described in the first aspect.

[0015] Fifthly, the present invention also provides a computer program product, the computer program product comprising a computer program, which, when executed by a processor, implements the steps of the power engineering building deformation monitoring method based on ground point cloud data described in the first aspect.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention first focuses on modeling the geometric structure information of buildings during the local feature aggregation stage when performing semantic segmentation. Then, based on multi-scale neighborhood search and normal vector curvature weighting mechanism, feature points are extracted from the semantically segmented building point cloud. Based on the extracted feature points, a straight line fitting algorithm is used to extract the feature lines of the buildings. Finally, feature points and feature lines of point clouds from different periods are matched to construct feature correspondences. Deformation monitoring is achieved through the spatial positional differences of point cloud features from different periods. It can accurately extract the point cloud of power engineering buildings from the original mixed point cloud, remove ground, vegetation and other interference points in the background. On this basis, by improving the feature sensitivity of building components at different scales through multi-scale neighborhood search and normal vector curvature weighting mechanism, it can improve the detection accuracy at building corners, edges and structural transitions.

[0017] 2. This invention uses an improved RAA-RandLA-Net model to perform semantic segmentation on point cloud data for each period, accurately extracting point clouds of power buildings. This enhances the ability to extract local neighborhood geometric features while effectively suppressing redundant features, improving the segmentation accuracy of building edges and details. An improved Harris3D operator is used to detect feature points in the building point clouds. Based on traditional Harris3D corner detection, a multi-scale neighborhood search mechanism is introduced to improve the sensitivity to features of building components at different scales, increasing the detection accuracy of feature points at building corners, edges, and structural transitions.

[0018] 3.b This invention extracts feature lines through RANSAC straight line fitting, which further enriches the geometric constraint information of the building and provides a stable basis for registration and deformation analysis. Attached Figure Description

[0019] The accompanying drawings, which form part of this embodiment, are used to provide a further understanding of this embodiment. The illustrative embodiments and their descriptions are used to explain this embodiment and do not constitute an improper limitation of this embodiment.

[0020] Figure 1 This is a flowchart of the feature line extraction process in Embodiment 1 of the present invention; Figure 2 This is a diagram of the overall network architecture of RAA-Randla-net according to Embodiment 1 of the present invention; Figure 3 This is a diagram of the improved RAA-LFA module network architecture of Embodiment 1 of the present invention; Figure 4 This is a rendering of the power engineering building extraction in Embodiment 1 of the present invention; Figure 5 This is a diagram showing the feature point extraction effect of Embodiment 1 of the present invention; Figure 6 This is a diagram showing the feature line extraction effect of Embodiment 1 of the present invention. Detailed Implementation

[0021] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0022] It should be noted that the following detailed descriptions are exemplary and intended to provide further explanation of this application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains.

[0023] Example 1: During long-term operation, power engineering structures may experience settlement, tilting, or structural deformation due to factors such as load, environment, and geological conditions. Therefore, high-precision deformation monitoring of power engineering structures is of great significance to ensure the safe and stable operation of the power system.

[0024] Existing methods for monitoring building deformation mainly include traditional measurement methods and monitoring methods based on 3D point clouds. Traditional methods, such as leveling, total station surveying, and GNSS monitoring, while possessing a certain level of accuracy, typically require the deployment of numerous measurement points, involve a high degree of manual intervention, have low monitoring efficiency, and struggle to achieve comprehensive monitoring of the entire building structure. With the development of LiDAR technology, deformation monitoring methods based on point cloud data have gradually become a research hotspot. Ground-based laser scanners can rapidly acquire high-density point cloud data of buildings, providing a data foundation for the overall deformation analysis of buildings.

[0025] As described in the background section, existing point cloud-based deformation monitoring technologies still have the following shortcomings: First, point cloud data is large and noisy, making it difficult for traditional filtering and segmentation methods to accurately extract point clouds of power engineering buildings, resulting in insufficient accuracy in subsequent analysis; Second, the extraction methods for feature points and feature lines are not very sensitive to complex building structures, easily leading to feature loss or false detection; Third, in multi-period point cloud comparative analysis, the registration method is easily affected by noise and external interference, resulting in errors in the deformation detection results.

[0026] To address at least one of the aforementioned problems, this embodiment provides a method for monitoring deformation of power engineering structures based on ground point cloud data, such as... Figure 1 As shown, it includes the following steps: S1. Point cloud data acquisition and preprocessing: Point cloud data of power engineering buildings at different times were acquired using a terrestrial laser scanner, and then preprocessed by denoising, downsampling, and coordinate system adjustments.

[0027] Specifically, ground-based laser scanners were used to perform laser scans on the target power engineering building at different times to obtain multi-period point cloud data of the building. The collected data underwent noise removal, invalid point removal, coordinate system unification, and point cloud registration to obtain a high-quality multi-period point cloud dataset.

[0028] S2, semantic segmentation of building point clouds: The preprocessed point cloud data is input into the improved RAA-RandLA-Net (RandLA-Net with ResidualAttention Aggregation) model for semantic segmentation, accurately extracting the point cloud of power buildings and performing boundary optimization.

[0029] like Figure 2 and Figure 3 As shown, the RAA-RandLA-Net model, based on the original RandLA-Net encoder-decoder framework, mainly makes the following improvements: Introducing the RAA-LFA module (Residual Attention Aggregation-Local Feature Aggregation): In the local feature aggregation stage, a residual attention mechanism is added, combining channel attention and spatial attention to focus on modeling the geometric structural information of buildings, thereby improving the model's ability to distinguish complex building boundaries.

[0030] Multi-scale feature enhancement: By setting different neighborhood scales, the local geometric features of the point cloud are extracted at multiple levels, which not only ensures the perception of the overall building outline, but also takes into account the capture of detailed components (such as beams, columns, and corners).

[0031] Boundary optimization strategy: Add a boundary-aware loss function (such as cross-entropy loss based on point-to-point distance or edge region weighting) during the decoding stage to enhance the segmentation accuracy of the model at the building edges, thereby avoiding "spiking" or "fractures" in the building point cloud.

[0032] Through the above improvements, the RAA-RandLA-Net model can automatically and accurately extract the point cloud of power engineering buildings from the original mixed point cloud, and remove the ground, vegetation and other interference points in the background, providing a clean data foundation for subsequent geometric feature extraction and deformation analysis.

[0033] The improved RAA-RandLA-Net model is used to perform semantic segmentation on the point cloud data of each period, accurately extracting the point cloud of power buildings. This enhances the ability to extract local neighborhood geometric features while effectively suppressing redundant features and improving the segmentation accuracy of building edges and details.

[0034] S3, Feature Point Extraction: like Figure 5 As shown, the improved Harris3D operator is used to detect feature points in the segmented building point cloud. Multi-scale neighborhood search and normal vector curvature weighting mechanism are introduced to improve the stability and repeatability of feature points.

[0035] After obtaining a clean building point cloud, an improved Harris3D operator is used for feature point extraction. Compared with the traditional Harris3D operator, improvements are made in the following aspects: Multi-scale neighborhood search: During feature point detection, the covariance matrix is ​​calculated using small, medium and large neighborhood scales respectively, and points with significant responses at multiple scales are selected by cross-scale persistence index to ensure that the extracted feature points include both local details (such as edges and corners) and overall structure (such as the turning point of a column).

[0036] Normal and curvature weighting: When constructing the covariance matrix, normal consistency weight and curvature weight are introduced. If the normal of a neighboring point differs significantly from that of the center point (greater than a preset value), its contribution is weakened; while regions with higher curvature (such as edges and corners) are given greater weight, thereby improving the detection capability of feature points at structural abrupt changes.

[0037] The improved characteristic response function can be expressed as: ; in, The weighted covariance matrix is ​​calculated by considering the point distance, normal consistency, and curvature information. This is an empirical constant.

[0038] Non-maximum suppression and spatial uniformity constraint: On the feature point response map, the non-maximum suppression (NMS) method is used to ensure that the extracted feature points are local optima; at the same time, the minimum spacing constraint is introduced to avoid excessive concentration of feature points in local regions, thereby ensuring that the feature points are evenly distributed.

[0039] An improved Harris3D operator is used to detect feature points in building point clouds. Based on traditional Harris3D corner detection, a multi-scale neighborhood search mechanism is introduced to improve the feature sensitivity of building components at different scales and improve the detection accuracy of feature points at building corners, edges and structural transitions.

[0040] S4. Feature line extraction: like Figure 6 As shown, based on the extracted feature points, the RANSAC straight line fitting method is used to extract the building feature lines, and their integrity and continuity are optimized.

[0041] Specifically, based on the extracted feature points, the RANSAC line fitting algorithm is used to extract the main feature lines of the building. This method can robustly identify the edge lines, column lines, and other geometric feature lines of the building even in the presence of noise points and outliers.

[0042] Feature lines were extracted by RANSAC straight line fitting, which further enriched the geometric constraint information of the building and provided a stable basis for registration and deformation analysis.

[0043] S5. Multi-period feature matching and deformation analysis: By matching feature points and feature lines from multiple point clouds, buildings can be visualized.

[0044] Specifically, feature points and feature lines extracted from building point clouds of different periods are matched to construct feature correspondences. By comparing and analyzing the spatial differences in the features of point clouds from multiple periods, the displacement and deformation of power engineering buildings at different times can be quantitatively calculated, enabling deformation monitoring and assessment.

[0045] Example 2: This embodiment provides a power engineering building deformation monitoring system based on ground point cloud data, including: The data acquisition module is configured to acquire point cloud data of power engineering construction at different times; The semantic segmentation module is configured to: perform semantic segmentation on the acquired point cloud data; and during semantic segmentation, focus on modeling the geometric structure information of buildings in the local feature aggregation stage. The feature point extraction module is configured to extract feature points from the semantically segmented building point cloud based on a multi-scale neighborhood search and normal vector curvature weighting mechanism. The feature line extraction module is configured to extract the feature lines of the building based on the extracted feature points using a straight line fitting algorithm. The monitoring module is configured to: match feature points and feature lines of point clouds at different times to construct feature correspondences; and monitor deformation by using the spatial position differences of point cloud features at different times.

[0046] The working method of the system is the same as that of the power engineering building deformation monitoring method based on ground point cloud data in Example 1, and will not be repeated here.

[0047] Example 3: This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the power engineering building deformation monitoring method based on ground point cloud data described in Embodiment 1.

[0048] Example 4: This embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and capable of running on the processor. When the processor executes the program, it implements the steps of the power engineering building deformation monitoring method based on ground point cloud data described in Embodiment 1.

[0049] Example 5: This embodiment provides a computer program product, which includes a computer program. When the computer program is executed by a processor, it implements the steps of the power engineering building deformation monitoring method based on ground point cloud data described in Embodiment 1.

[0050] The above description is merely a preferred embodiment of this practice and is not intended to limit the scope of this practice. Various modifications and variations can be made to this practice by those skilled in the art. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of this practice should be included within the protection scope of this practice.

Claims

1. A method for monitoring deformation of power engineering buildings based on ground point cloud data, characterized in that, include: Acquire point cloud data of power engineering construction at different times; Semantic segmentation of the acquired point cloud data is performed, specifically including: using the RAA-RandLA-Net model for semantic segmentation, extracting point clouds of power buildings, and performing boundary optimization; the RAA-RandLA-Net model is improved based on the original RandLA-Net encoder-decoder framework as follows: A RAA-LFA module is introduced: in the local feature aggregation stage, a residual attention mechanism is added, combining the channel attention and spatial attention of the GAM module to focus on modeling the geometric structural information of the buildings; different neighborhood scales are set by setting the ECA branch with the SPC module connected after the GAM module to perform multi-level extraction of local geometric features of the point cloud; a boundary-aware loss function is added in the decoding stage to enhance the segmentation accuracy of the model at the building edges; Feature point extraction was performed on the segmented building point cloud using an improved Harris3D operator. Compared with the traditional Harris3D operator, the improved Harris3D operator introduces a multi-scale neighborhood search and normal vector curvature weighting mechanism. Specifically, during feature point detection, the covariance matrix is ​​calculated using multiple neighborhood scales, and points with significant responses at multiple scales are selected through cross-scale persistence index to ensure that the extracted feature points include both local details and overall structure. When constructing the covariance matrix, normal consistency weight and curvature weight are introduced. If the difference between the normal vector of a neighboring point and the center point is greater than a preset value, its contribution is weakened. Regions with higher curvature are given greater weight. Based on the extracted feature points, a straight line fitting algorithm is used to extract the feature lines of the building; The feature points and feature lines of point clouds from different periods are matched to construct feature correspondences; deformation monitoring is achieved by using the spatial location differences of point cloud features from different periods.

2. The method for monitoring power engineering building deformation based on ground point cloud data as described in claim 1, characterized in that, The acquired point cloud data is subjected to denoising, downsampling, and coordinate system unification processing.

3. The method for monitoring deformation of power engineering buildings based on ground point cloud data as described in claim 1, characterized in that, Extracting the feature lines of a building involves: using the extracted feature points and employing the RANSAC line fitting algorithm to extract the feature lines of the building.

4. A power engineering building deformation monitoring system based on ground point cloud data, characterized in that, include: The data acquisition module is configured to acquire point cloud data of power engineering construction at different times; The semantic segmentation module is configured to perform semantic segmentation on the acquired point cloud data. Specifically, this includes: using the RAA-RandLA-Net model for semantic segmentation, extracting point clouds of power buildings, and performing boundary optimization. The RAA-RandLA-Net model is improved upon the original RandLA-Net encoder-decoder framework by: introducing a RAA-LFA module; adding a residual attention mechanism in the local feature aggregation stage, combining the channel attention and spatial attention of the GAM module to focus on modeling the geometric structure information of the buildings; setting different neighborhood scales through the ECA branch with the SPC module connected to the GAM module to perform multi-level extraction of local geometric features of the point cloud; and adding a boundary-aware loss function in the decoding stage to enhance the model's segmentation accuracy at building edges. The feature point extraction module is configured to extract feature points from the segmented building point cloud using an improved Harris3D operator. Compared to the traditional Harris3D operator, the improved Harris3D operator introduces a multi-scale neighborhood search and normal vector curvature weighting mechanism. Specifically, during feature point detection, the covariance matrix is ​​calculated using multiple neighborhood scales, and points with significant responses at multiple scales are selected through cross-scale persistence indices to ensure that the extracted feature points include both local details and overall structure. When constructing the covariance matrix, normal consistency weights and curvature weights are introduced. If the difference between the normal vector of a neighboring point and the center point is greater than a preset value, its contribution is weakened. Regions with higher curvature are given greater weight. The feature line extraction module is configured to extract the feature lines of the building based on the extracted feature points using a straight line fitting algorithm. The monitoring module is configured to: match feature points and feature lines of point clouds at different times to construct feature correspondences; and monitor deformation by using the spatial position differences of point cloud features at different times.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that, When the processor executes the program, it implements the steps of the power engineering building deformation monitoring method based on ground point cloud data as described in any one of claims 1-3.

6. A computer program product, characterized in that, The computer program product includes a computer program that, when executed by a processor, implements the steps of the power engineering building deformation monitoring method based on ground point cloud data as described in any one of claims 1-3.

Images