Bridge point cloud data downsampling method and system based on structural feature importance

Abstract

The application discloses a bridge point cloud data downsampling method and system based on structural feature importance, and belongs to the technical field of three-dimensional laser point cloud data processing, and comprises the following steps: step S1, data preprocessing; step S2, structural feature importance index calculation; step S3, adaptive voxel downsampling; step S4, quality evaluation; and step S5, result output. According to the structural feature importance of different structural parts of the bridge, the application can automatically adjust the downsampling parameters, realizes high compression rate while guaranteeing the details of key feature regions, effectively solves the problems of serious detail loss and insufficient pertinence of the existing method, and is suitable for compression processing of various bridge point cloud data.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional laser point cloud data processing technology, specifically to a method and system for downsampling bridge point cloud data based on the importance of structural features. Background Technology

[0002] As vital transportation infrastructure, the health monitoring and maintenance of bridges are crucial for ensuring public safety. Three-dimensional laser scanning technology, with its non-contact, high-precision, and high-efficiency characteristics, has become an important tool for bridge geometric modeling and deformation monitoring. However, bridge structures are typically massive, and the point cloud data acquired through 3D laser scanning, after registration and stitching, often reaches tens of gigabytes in size, posing significant challenges to data storage, transmission, and processing.

[0003] Existing point cloud downsampling methods mainly include the following types:

[0004] (1) Voxel downsampling method: Downsampling is performed using a uniform voxel grid, which cannot distinguish the importance of different regions and is prone to losing key structural features during compression.

[0005] (2) Curvature-based downsampling method: It can preserve high curvature regions, but it does not consider the semantic information and structural importance of engineering structures, and its applicability to large engineering structures such as bridges is limited;

[0006] (3) Adaptive downsampling method based on deep learning: Although adaptive downsampling is achieved, it relies on a large amount of training data, has poor interpretability, and is not optimized for the special needs of engineering structures such as bridges;

[0007] (4) Random downsampling method: simple and efficient, but it is easy to lose key features and lacks consideration of the importance of structure.

[0008] However, existing point cloud downsampling methods have the following main drawbacks when applied to bridge point cloud data:

[0009] First, the massive amount of data makes storage and processing difficult. The complexity and large-scale characteristics of the bridge structure result in an extremely large amount of original point cloud data, and traditional downsampling methods have limited compression rates, making it difficult to effectively reduce the amount of data.

[0010] Second, the downsampling compression rate is low. Existing downsampling compression algorithms are not adequately optimized for the structural characteristics of point cloud data, especially for engineering structures like bridges with clearly defined structural partitions, where the compression effect is not ideal.

[0011] Third, significant loss of detail. Traditional downsampling methods tend to lose detailed information about critical bridge components (such as bearings and joints) during the compression process, which is crucial for bridge health monitoring.

[0012] Fourth, there is a lack of specificity. Most current downsampling methods compress the entire point cloud uniformly without considering the geometric features and importance differences of different structural locations on the bridge (such as the superstructure, substructure, and bridge deck system). This results in the loss of details in important areas rather than data redundancy in important areas.

[0013] Fifth, there is a lack of fusion of engineering semantic information. Existing methods fail to organically combine the structural knowledge of bridge engineering with the geometric features of point cloud processing, and thus cannot achieve intelligent downsampling based on structural importance.

[0014] Therefore, there is an urgent need for an adaptive point cloud downsampling method that can achieve a high compression rate while preserving the details of key feature areas, tailored to the structural characteristics of bridges. Summary of the Invention

[0015] The technical problem to be solved by this invention is: how to solve the problems of insufficient targeting, serious loss of details, and inability to achieve intelligent downsampling based on structural importance in existing bridge point cloud data downsampling methods, and provides a bridge point cloud data downsampling method based on the importance of structural features.

[0016] The present invention solves the above-mentioned technical problems through the following technical solution, and the present invention includes the following steps:

[0017] Step S1: Data Preprocessing

[0018] The original 3D point cloud data of the bridge is acquired and preprocessed, including registration, denoising and coordinate normalization.

[0019] Step S2: Bridge structure classification and importance labeling

[0020] Based on the bridge's structural and spatial geometric features, the preprocessed point cloud data is divided into regions and labeled with importance. Structural importance analysis is performed on each region, and the structural feature importance index of each region is calculated based on a multi-factor coupling algorithm.

[0021] Step S3: Calculation of the structural feature importance index

[0022] Based on the structural feature importance index of each region, an adaptive voxel downsampling algorithm is used to downsample the point cloud data. The adaptive voxel downsampling algorithm determines the corresponding voxel size according to the structural feature importance index of different regions. Regions with higher structural feature importance indices use smaller voxel sizes.

[0023] Step S4: Adaptive Voxel Downsampling

[0024] The quality of the downsampled point cloud data is evaluated, and the feature retention rate and compression rate before and after downsampling are calculated.

[0025] Step S5: Quality Assessment and Output

[0026] When the quality assessment results meet the preset requirements, the final simplified bridge point cloud data is output; if not, the voxel downsampling parameters are adjusted, and the downsampling step S3 and the quality assessment step S4 are repeated until the iteration termination condition is triggered.

[0027] Furthermore, in step S2, the preprocessed point cloud data is structured and divided into an upper structure region, a lower structure region, and a bridge deck system region. Each region is further subdivided according to component type into load-bearing components, general components, supports, piers, abutments, pavement layers, guardrails, lighting facilities, and signs. Among them, load-bearing components, general components, and supports are located in the upper structure region, piers and abutments are located in the lower structure region, and pavement layers, guardrails, lighting facilities, and signs are located in the bridge deck system region.

[0028] Furthermore, the method for calculating the structural feature importance index in step S2 is as follows:

[0029] First calculate each point within the region Point-level feature importance value Then, the arithmetic mean of the point-level feature importance values ​​of all points within the region is taken as the structural feature importance index of the region. This region is a subdivided component region, that is, a component region subdivided according to component type; among which, the point-level feature importance value The calculation formula is:

[0030]

[0031] in: For point The structural importance factor of the location; For point The curvature characteristic factor; For point Local density factor; Let be the weight coefficient, and satisfy... α ranges from 0.5 to 0.7, γ ranges from 0.2 to 0.4, and δ ranges from 0.05 to 0.15. To subdivide the index number of each point in the point cloud of the component region, , This represents the total number of points in the point cloud within the corresponding area. For the index number of the subdivided component area, , The total number of subdivided component areas.

[0032] Furthermore, structural importance factors The importance factors for each component are determined based on its type and location (in a stepped distribution), with a range of values ​​from 0 to 1.0. These factors include the following:

[0033] Load-bearing components in the superstructure: ;

[0034] General components in the superstructure: ;

[0035] Supports in the superstructure: ;

[0036] Bridge piers in the substructure: ;

[0037] Bridge abutments in the substructure: ;

[0038] Pavement layers in bridge deck systems: ;

[0039] Guardrails in the bridge deck system: ;

[0040] Lighting fixtures in the bridge deck system: ;

[0041] Bridge deck signage: .

[0042] Furthermore, curvature characteristic factor The calculation formula is:

[0043] ;

[0044] in, For point The average curvature at that point, through the principal curvature and The calculation yielded: , This represents the total number of points in the point cloud within the corresponding area.

[0045] In point cloud data, the principal curvature needs to be approximated by local neighborhood estimation, which can be obtained using various methods. This invention employs the covariance matrix eigenvalue decomposition method, obtaining three eigenvalues ​​through neighborhood search, covariance matrix construction, and eigenvalue decomposition. ;

[0046] Calculate the principal curvature using the ratio of eigenvalues:

[0047] .

[0048] Furthermore, local density factor The calculation formula is:

[0049]

[0050] in, For point The local point cloud density at a certain location is calculated using... Centered on, with radius The number of points within the spherical neighborhood is obtained; This represents the maximum local point cloud density within the corresponding region.

[0051] Furthermore, the voxel size of the adaptive voxel downsampling algorithm for each region in step S3... The calculation formula is:

[0052]

[0053] in: The minimum voxel size is set to 2-3 times the average point spacing of the point cloud. The maximum voxel size is set to 10-15 times the average point spacing of the point cloud. This is the structural feature importance index corresponding to this region; This represents the maximum importance index corresponding to the selected region; This is a voxel adjustment parameter, with a value range of 1.5-2.5;

[0054] Because the laser scanning data from the bridge site inevitably contains outliers such as atmospheric dust and strong metallic reflections, directly taking the global absolute maximum value would significantly increase the denominator due to these extreme noises. This would cause the importance index of the entire bridge's normal area to be compressed to near zero, rendering the algorithm ineffective. Therefore, to ensure the numerical stability of the normalized mathematical model in practical engineering applications, noise is removed, and the maximum effective data value falling within the confidence interval is selected as the maximum importance index. .

[0055] Furthermore, the feature retention rate in step S4 The calculation formula is:

[0056]

[0057] in, This represents the number of feature points retained after downsampling. The number of feature points before downsampling is given, and the feature points are identified using the Feature Point Detection Algorithm (ISS algorithm).

[0058] Furthermore, the compression ratio in step S4 The calculation formula is:

[0059]

[0060] in, This represents the total number of points in the point cloud after downsampling. This represents the total number of points in the point cloud before downsampling.

[0061] Furthermore, the preset requirement for quality assessment in step S5 is: feature retention rate. And compression ratio If the preset requirements are not met, adjust the voxel adjustment parameters. Then repeat steps S3-S4 until the preset requirements are met or the maximum number of iterations is reached.

[0062] This invention also provides a bridge point cloud data downsampling system based on the importance of structural features, applied to the above-mentioned method, including:

[0063] The data preprocessing module is used to acquire the original 3D point cloud data of the bridge and preprocess the original 3D point cloud data.

[0064] The importance index calculation module is used to divide the preprocessed point cloud data into regions and label their importance based on the bridge's structural and spatial geometric features, perform structural importance analysis on each region, and calculate the structural feature importance index for each region.

[0065] The voxel downsampling module is used to downsample point cloud data using an adaptive voxel downsampling algorithm based on the structural feature importance index of each region.

[0066] The quality assessment module is used to assess the quality of downsampled point cloud data and calculate the feature retention rate and compression rate before and after downsampling.

[0067] The result output module is used to output the final simplified bridge point cloud data when the quality assessment result meets the preset requirements; if it does not meet the requirements, the voxel downsampling parameters are adjusted and steps S3 to S4 are repeated until the iteration termination condition is triggered.

[0068] The present invention has the following advantages over the prior art:

[0069] 1. High adaptability: The downsampling parameters are automatically adjusted according to the importance of the characteristics of different structural parts of the bridge, avoiding the loss of details or data redundancy caused by uniform downsampling in traditional methods.

[0070] 2. High compression ratio: While ensuring the details of key feature areas, the overall compression ratio can reach more than 60%, effectively reducing the pressure of data storage and transmission.

[0071] 3. Complete feature retention: Small voxel downsampling is used for key areas such as bridge load-bearing components and bearings, with a feature retention rate of over 85%, ensuring the accuracy of subsequent analysis.

[0072] 4. Good versatility: The methodology framework has good scalability and the importance assessment strategy can be adjusted according to different bridge types and inspection requirements.

[0073] 5. High degree of automation: The entire downsampling process is completed automatically without manual intervention, which improves work efficiency. Attached Figure Description

[0074] Figure 1 This is an overall flowchart of the bridge point cloud data downsampling method based on the importance of structural features in this embodiment of the invention;

[0075] Figure 2 This is a schematic diagram of the adaptive voxel downsampling algorithm in an embodiment of the present invention;

[0076] Figure 3 This is an example diagram of a simplified bridge point cloud after adaptive voxel downsampling in an embodiment of the present invention. Detailed Implementation

[0077] The embodiments of the present invention are described in detail below. These embodiments are implemented based on the technical solution of the present invention, and provide detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the following embodiments.

[0078] Example 1

[0079] like Figure 1 , 2 As shown, this embodiment uses the three-dimensional laser scanning point cloud data of a highway bridge as the processing object, and the specific implementation steps are as follows:

[0080] Step S1: Data Preprocessing

[0081] The total amount of the original point cloud data after registration and stitching is , containing approximately There are billions of points. First, noise reduction is performed, using a statistical filter to remove outliers and setting the number of neighboring points. Standard deviation multiple Then, coordinate normalization is performed, and the result is scaled to the unit cube.

[0082] Basic information of preprocessed point cloud data: Total number of points: Average point spacing: Spatial range: direction , direction , direction .

[0083] Step S2: Bridge structure classification and importance labeling

[0084] Based on the structural characteristics of the bridge, the point cloud data is divided into the following regions:

[0085] Upper structural region (accounting for a certain percentage of the total point cloud) ): Load-bearing components (main beams): General components (sidewalk slabs, etc.): Support: .

[0086] The lower structural region (accounting for a certain percentage of the total point cloud) Bridge piers: Bridge abutment: .

[0087] Bridge surface area (accounting for a certain percentage of the total point cloud) ): Pavement layer (asphalt concrete): Guardrail (concrete guardrail): Lighting facilities (streetlights): Signage: .

[0088] Step S3: Calculate the structural feature importance index

[0089] First calculate each point within the region Point-level feature importance value Then, the arithmetic mean of the point-level feature importance values ​​of all points in the region is taken as the structural feature importance index of the region. .

[0090] Point-level feature importance value The calculation formula is:

[0091]

[0092] in, For point The structural importance factor of the location; For point The curvature characteristic factor; For point Local density factor; Let be the weight coefficient, and satisfy... .

[0093] The weighting coefficients are set as follows: , , .

[0094] Calculation example: For a point on the main beam Structural importance factor: (Load-bearing components), curvature characteristic factor: (Calculated via curvature estimation), Local density factor: (Local feature point density).

[0095] therefore:

[0096]

[0097] The structural feature importance index of this region (the area of ​​load-bearing components of the superstructure). The arithmetic mean of the point-level feature importance values ​​of all points in the region is calculated as follows: .

[0098] For a certain point on the substructure pier (Circular cross-section): Structural importance factor: (Bridge pier, load-bearing component of the substructure), curvature characteristic factor: (The pier surface is a relatively smooth cylindrical surface), local density factor: (Local feature point density)

[0099] therefore:

[0100]

[0101] The structural feature importance index of this area (substructure pier area) The arithmetic mean of the point-level feature importance values ​​of all points in the region is calculated as follows: .

[0102] Step S4: Adaptive Voxel Downsampling

[0103] The formula for calculating voxel size is:

[0104]

[0105] Parameter settings: (about (times the average point spacing) (about (times the average point spacing) , .

[0106] For the load-bearing components of the superstructure, the voxel size is:

[0107]

[0108] For the substructure pier area, its voxel size is:

[0109]

[0110] After performing adaptive voxel downsampling: Number of points after downsampling: Compression ratio: .

[0111] Step S5: Quality Assessment

[0112] Feature point detection: using The (IntrinsicShapeSignature) algorithm extracts feature points. Number of feature points before downsampling: Number of feature points after downsampling: Feature retention rate: .

[0113] It should be noted that the ISS (Intrinsic Shape Signatures) algorithm is a classic point cloud feature point detection method. Its core idea is: for each point in the point cloud, a local neighborhood is constructed and its covariance matrix is ​​calculated to analyze the variance distribution of that neighborhood along the principal directions. When the variance distribution of a point is significant across multiple principal directions (i.e., strong anisotropy), that point is identified as a feature point.

[0114] Quality assessment results: Feature retention rate (Meets requirements) Compression ratio (Requirements met)

[0115] Therefore, the final simplified bridge point cloud data is output as the final result, such as... Figure 3 As shown.

[0116] Example 2

[0117] To verify the effectiveness of the method of this invention, a comparison was made with three traditional downsampling methods:

[0118] Comparison Method 1: Random downsampling: with the same compression rate of 64.2% and feature retention rate of 72.8%, the details of the main load-bearing components were severely lost.

[0119] Comparison Method 2: Uniform voxel downsampling, with voxel size uniformly set to 10mm, compression rate: 61.5%, feature retention rate: 78.3%, details in key areas such as supports are not clear.

[0120] Comparison Method 3: Curvature-weighted downsampling, which adjusts the sampling density based on local curvature, with a compression rate of 63.8% and a feature retention rate of 84.1%, does not consider structural semantic information and oversamples flat areas such as bridge piers.

[0121] The method of this invention achieves a compression rate of 64.2%, a feature retention rate of 88.2%, preserves the details of key structural regions, and significantly reduces the amount of data.

[0122] The comparative results show that the method of the present invention is superior to the traditional method in terms of compression rate and feature retention rate, and has significant advantages in protecting the key structural features of bridges.

[0123] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A method for downsampling bridge point cloud data based on the importance of structural features, characterized in that, Includes the following steps: Step S1: Data Preprocessing Obtain the original 3D point cloud data of the bridge and preprocess the original 3D point cloud data; Step S2: Calculation of the structural feature importance index Based on the bridge's structural and spatial geometric features, the preprocessed point cloud data is divided into regions and its importance is labeled. Structural importance analysis is performed on each region, and the structural feature importance index of each region is calculated. Step S3: Adaptive Voxel Downsampling Based on the structural feature importance index of each region, an adaptive voxel downsampling algorithm is used to downsample the point cloud data; Step S4: Quality Assessment The quality of the downsampled point cloud data is evaluated, and the feature retention rate and compression rate before and after downsampling are calculated. Step S5: Output Results When the quality assessment results meet the preset requirements, the final simplified bridge point cloud data is output; if not, the voxel downsampling parameters are adjusted, and steps S3 to S4 are repeated until the iteration termination condition is triggered.

2. The bridge point cloud data downsampling method based on the importance of structural features according to claim 1, characterized in that, In step S2, the preprocessed point cloud data is structured and divided into an upper structure region, a lower structure region, and a bridge deck system region. Each region is further subdivided according to component type into load-bearing components, general components, supports, piers, abutments, pavement layers, guardrails, lighting facilities, and signs. Among them, load-bearing components, general components, and supports are located in the upper structure region, piers and abutments are located in the lower structure region, and pavement layers, guardrails, lighting facilities, and signs are located in the bridge deck system region.

3. The bridge point cloud data downsampling method based on the importance of structural features according to claim 2, characterized in that, In step S2, the structural feature importance index is calculated as follows: S21: Each point within the calculation region Point-level feature importance value : ； in, For point The structural importance factor of the location; For point The curvature characteristic factor; For point Local density factor; This represents the weighting coefficient; this area is a subdivided component area, that is, a component area subdivided according to component type. To subdivide the index number of each point in the point cloud of the component region, , This represents the total number of point clouds within the corresponding region. S22: Take the arithmetic mean of the point-level feature importance values ​​of all points in the region as the structural feature importance index of the current region. ,in, For the index number of the subdivided component area, , This represents the total number of subdivided component regions.

4. The bridge point cloud data downsampling method based on the importance of structural features according to claim 3, characterized in that, Structural importance factor The structural importance factor of each component is determined based on its type and location, and the range of values ​​is (0, 1.0).

5. The bridge point cloud data downsampling method based on the importance of structural features according to claim 3, characterized in that, Curvature characteristic factor The calculation formula is: ； in, For point The average curvature at that point, This represents the total number of point clouds within the corresponding region.

6. The bridge point cloud data downsampling method based on the importance of structural features according to claim 3, characterized in that, Local density factor The calculation formula is: ； in, For point The local point cloud density at a certain location is calculated using... Centered on, with radius The number of points within the spherical neighborhood is obtained; This represents the maximum local point cloud density within the corresponding region.

7. The bridge point cloud data downsampling method based on the importance of structural features according to claim 3, characterized in that, In step S3, the adaptive voxel downsampling algorithm determines the corresponding voxel size based on the structural feature importance index of different regions. The calculation formula is: ； in, Minimum voxel size; Maximum voxel size; This is an index representing the importance of structural features corresponding to a region. The maximum importance index corresponding to the selected region; This refers to the voxel adjustment parameter, i.e., the voxel downsampling parameter.

8. The bridge point cloud data downsampling method based on the importance of structural features according to claim 7, characterized in that, In step S4, the feature retention rate The calculation formula is: ； in, The number of feature points retained after downsampling is obtained by detecting the downsampling point cloud using a point cloud feature point detection algorithm. The number of feature points before downsampling is obtained by detecting the point cloud before downsampling using a point cloud feature point detection algorithm, which is the ISS algorithm. Compression ratio The calculation formula is: ； in, This represents the total number of points in the point cloud after downsampling. This represents the total number of points in the point cloud before downsampling.

9. The bridge point cloud data downsampling method based on the importance of structural features according to claim 8, characterized in that, In step S5, the preset requirement for quality assessment is: feature retention rate. And compression ratio ; If the preset requirements are not met, adjust the voxel adjustment parameters. Then repeat steps S3 to S4 until the preset requirements are met or the maximum number of iterations is reached.

10. A bridge point cloud data downsampling system based on the importance of structural features, characterized in that, The method applied to any one of claims 1 to 9 includes: The data preprocessing module is used to acquire the original 3D point cloud data of the bridge and preprocess the original 3D point cloud data. The importance index calculation module is used to divide the preprocessed point cloud data into regions and label their importance based on the bridge's structural and spatial geometric features, perform structural importance analysis on each region, and calculate the structural feature importance index for each region. The voxel downsampling module is used to downsample point cloud data using an adaptive voxel downsampling algorithm based on the structural feature importance index of each region. The quality assessment module is used to assess the quality of downsampled point cloud data and calculate the feature retention rate and compression rate before and after downsampling. The result output module is used to output the final simplified bridge point cloud data when the quality assessment result meets the preset requirements; if it does not meet the requirements, the voxel downsampling parameters are adjusted and steps S3 to S4 are repeated until the iteration termination condition is triggered.

Images