Asphalt panel point cloud block reconstruction and flatness evaluation method

Abstract

The application discloses a method for point cloud block reconstruction and flatness evaluation of asphalt panels, and belongs to the technical field of engineering measurement. The application solves the problems of low efficiency and easy omission of local mutations of an existing manual detection method, low calculation efficiency of a global fitting method, poor adaptability of a single geometric model, and discontinuity of boundaries caused by conventional block processing. Technical solution points are as follows: collecting and preprocessing laser radar point cloud data of asphalt panels; adaptively block processing based on local features of point clouds and establishing an index; performing mixed fitting of multiple types of curved surfaces according to features of sub-regions; performing neighborhood interpolation repair and global plane fitting for invalid blocks; fusing adjacent curved surface parameters to realize boundary smoothing; calculating signed orthogonal distance and optimizing boundary point distance calculation; and outputting a curved surface model, color point cloud and a statistical report after quantitatively evaluating flatness.

Description

Technical Field

[0001] This invention belongs to the field of engineering measurement technology, specifically relating to a method for point cloud segmentation reconstruction and flatness assessment of asphalt panels. Background Technology

[0002] With the rapid development of laser scanning, UAV aerial surveying, and 3D sensor technologies, point cloud data, as an important means of acquiring geometric information of structural surfaces, has been gradually applied in infrastructure inspection, quality assessment, and refined modeling. In the construction of the upper and lower reservoirs of pumped storage power stations, asphalt concrete panels, as key seepage prevention structures, directly affect the safety and stability of reservoir operation. According to specifications, the shape and flatness of each layer of the panel should be strictly controlled, and abrupt changes are not allowed. Currently, the inspection of panel flatness and surface morphology still mainly relies on two-dimensional cross-sectional methods or manual point measurements, which makes it difficult to comprehensively and continuously acquire geometric information across the entire panel area, resulting in low efficiency and a tendency to miss local abrupt changes. Against this backdrop, introducing 3D laser scanning to collect point cloud data of the panel surface has become an effective solution to improve inspection accuracy and efficiency.

[0003] Relevant patent documents retrieved:

[0004] This document, published in China (CN117109481A), discloses a method for evaluating the overall smoothness and flow capacity of concrete flow surfaces. The method first uses a three-dimensional laser scanner to measure and collect three-dimensional point cloud data of the flow surface to be evaluated. Then, it performs planar / curved surface feature fitting on the processed three-dimensional point cloud data, calculates the distance and average distance between each point in the three-dimensional point cloud data and the fitted optimal planar / curved surface, and subsequently calculates the overall standard deviation, overall smoothness, and equivalent roughness. Finally, it substitutes the equivalent roughness into the roughness calculation formula according to applicable conditions to obtain the roughness ratio, and compares it with design parameter indicators to evaluate and judge the flow capacity of the concrete flow surface.

[0005] The prior art represented by the aforementioned documents has at least the following unresolved technical problems or defects: (1) Only global planar / surface feature fitting is used to process point cloud data. No block division and block fitting technical solutions are disclosed. For large-scale point cloud data, there will be a large amount of global fitting calculation, which is difficult to meet the actual needs of rapid evaluation on the engineering site. The relevant evidence is that the literature only records the technical means of overall fitting of the flow surface point cloud data, without mentioning any block division or block fitting related operations. (2) The global fitting method lacks adaptability to complex geometric shapes and cannot adapt to the structural features of the asphalt panel edge of the pumped storage power station with significant changes in curvature and local curvature. It is difficult to accurately express such complex panel geometry. The relevant evidence is that the literature only records a single overall plane / surface fitting strategy and does not involve differentiated fitting methods for local geometric features. (3) No optimization processing scheme for the fitting boundary is mentioned, which cannot solve the problem of boundary error mutation and poor continuity caused by block division in traditional point cloud processing. It also does not establish an efficient and intuitive error analysis and quality evaluation mechanism to assist in the judgment of construction quality. The relevant evidence is that the literature only completes the evaluation of flow capacity by calculating the overall flatness, equivalent roughness and other indicators, and does not record the relevant technical means of boundary optimization and refined error analysis.

[0006] In view of this, the present invention is hereby proposed. Summary of the Invention

[0007] To address the aforementioned technical problems in the existing technology, this invention provides a method for point cloud block reconstruction and flatness assessment of asphalt panels, which solves the problems of low efficiency and easy omission of local mutations in existing manual inspection methods, low computational efficiency of global fitting methods, poor adaptability of single geometric models, and discontinuous boundaries caused by conventional block processing.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows: A method for point cloud segmentation reconstruction and flatness assessment of asphalt panels includes: S1. Collect lidar point cloud data of the asphalt panel surface, extract the three-dimensional coordinate information of the point cloud and perform preprocessing to determine the spatial range of the asphalt panel detection area. S2. Based on the local features of the point cloud, dynamically adjust the block size, divide the detection area into several sub-regions, establish the mapping relationship between points and blocks, and generate a block index table containing regional feature information. S3. Select the appropriate surface type for the geometric features of different sub-regions, obtain the surface coefficients of each sub-region, perform the block fitting operation in parallel through multiple processes, and count the fitting results of each block. S4. Perform neighborhood interpolation repair on invalid blocks that fail to fit or have insufficient point cloud quantity. Use global plane fitting as a fallback method for invalid blocks that fail to interpolate repair. Mark and count the repair results of invalid blocks. S5. Generate a uniform grid for each sub-region, determine the boundary points and fusion regions of the blocks, calculate the spatial weight factor within the fusion region, and perform weighted fusion of the surface parameters of adjacent sub-regions based on the spatial weight factor. S6. Calculate the signed orthogonal distance from each point in the point cloud to the corresponding fitted surface. Use the neighborhood surface distance optimization strategy to determine the final distance for the block boundary points. Classify the signed orthogonal distance results according to the preset error threshold. Complete the quantitative evaluation of asphalt panel flatness based on distance distribution statistics. S7. Generate a curved mesh model of the asphalt panel, a color precision point cloud, and a flatness statistical report.

[0009] Furthermore, the specific operations of step S1 include: S11. Use fixed, mobile or handheld lidar sensors to scan and collect three-dimensional point cloud data of the asphalt panel surface, and combine the attitude sensor IMU and the positioning system GNSS / RTK to synchronously obtain spatial pose information to achieve point cloud spatial registration and extract the three-dimensional coordinate information of each point. S12. Remove noise points from the point cloud using at least one of statistical filtering and radius filtering methods. S13. Use the voxel grid method to downsample the point cloud data; S14. Crop the point cloud according to the boundary range set by the detection task, extract the point cloud of the target area, and determine the spatial range of the detection area. S15. Convert the point cloud data to the engineering coordinate system or the local coordinate system to complete the coordinate system.

[0010] Furthermore, the specific operations of step S2 include: S21. Based on the spatial boundary of the point cloud , Divide the point cloud into regular grid blocks; S22. According to the preset block size Calculate the initial number of blocks; S23. Dynamically adjust the block size according to the local features of the point cloud in each initial grid to generate sub-regions that adapt to the local features; S24. Assign each point to the corresponding sub-region and establish a mapping relationship from point to block; S25. Generate a unique identifier for each sub-region and establish a block index table containing regional feature information.

[0011] Furthermore, step S3 specifically includes: S31. Perform point count determination and sampling for each block: For each block, determine whether the number of point clouds it contains is not less than the preset threshold MIN_TILE_POINTS. If it meets the requirement, randomly sample from the point cloud set of the block, and the number of samples does not exceed the preset threshold PLANE_SAMPLE. Use the sampled point set for subsequent surface fitting. If it does not meet the requirement, mark the block as an invalid block and include it in the repair process of step S4.

[0012] Furthermore, step S3 also includes: S32. Perform feature analysis on the sampling points and make a preliminary judgment on the surface type tendency: Calculate the centroid of the sampling point set and perform SVD / PCA decomposition to obtain singular values / eigenvalues ​​and corresponding eigenvectors; take the eigenvector corresponding to the smallest eigenvalue as the initial normal and unify the normal, and calculate the plane constant term; The surface type tendency is initially determined based on the ratio of singular values ​​to eigenvalues. The surface type tendency includes: planar tendency, cylindrical tendency, and quadratic tendency.

[0013] Furthermore, the specific operations of step S3 also include: S33. Surface type adaptation and robust optimization: If it is determined to be a planar tendency, the planar parameters obtained from feature analysis are directly used, and the orthogonal distance residual from the point to the plane is calculated to complete the fitting. If it is determined to be a cylindrical inclination, the axis direction and point cloud centroid obtained from feature analysis are used as the initial solution. The axis position and radius are optimized, and the least squares method is used to minimize the absolute value of the signed orthogonal distance residual to complete the fitting. If it is determined to be a quadratic surface tendency, the planar solution obtained from feature analysis is converted into the initial equation of the quadratic surface to obtain the initial parameters. The absolute value of the signed orthogonal distance residual is minimized by the least squares method to complete the fitting. If the coefficients of the quadratic terms after optimization are all lower than the preset threshold SURFACE_EPS, it is determined to be a fitting degradation, and the quadratic surface fitting result is reverted to the planar solution obtained from feature analysis. S34. Parallel Execution and Statistics: The operations of S31-S33 are executed in parallel on all blocks through a multi-process pool. The final fitted surface type and fitting result of each block are recorded, and the number of effective surfaces, the number of fitted degenerate surfaces, and the number of cylindrical fitted surfaces are counted.

[0014] Furthermore, the specific operations of step S4 include: S41. For invalid blocks, use their centers... For reference, by radius The search proceeds step-by-step from 1 to a preset threshold MAX_NEIGHBOR_RADIUS, gradually searching for the center point of the surrounding fitted blocks. Obtain the surface of each fitted block in place Values ​​are used to construct a set of neighborhood support points; S42. Construct a weighted least squares model based on inverse distance weights, solve the model to obtain the surface parameters of invalid blocks and complete the interpolation repair. If the number of points in the neighborhood support point set is less than the preset threshold NEIGHBOR_MIN_USED or the weighted least squares model fails to solve, then stop this interpolation repair. S43. For invalid blocks where interpolation repair fails, randomly sample from the global point cloud and the number of samples does not exceed the preset threshold GLOBAL_SAMPLE. Force the use of planar ODR fitting to obtain global planar coefficients and complete global fallback fitting. S44. Mark the number of blocks that have been repaired by interpolation or fallback, and calculate the percentage of effective blocks after repair.

[0015] Furthermore, the specific operations of step S5 include: S51. Generate a uniform grid for each sub-region whose size can be set as needed; S52, according to the preset boundary threshold Determine the merging region around the four edges of the block when the normalized coordinates of the points are... , or When this point is reached, it is determined to be a boundary point; S53. Calculate the shortest distance from a point within the fusion region to the block boundary. , distance Normalization yields ,according to Calculate the spatial weighting factor, wherein the spatial weighting factor is calculated by linear weighting or Gaussian smoothing weighting; S54. Within the fusion region, the surface parameters of adjacent blocks are weighted and averaged according to the spatial weighting factor. S55, merge the region in parameter space The internal discretization is performed into a fine mesh, and the fusion parameters are calculated only for the mesh vertices. During global surface reconstruction, the surface coefficient values ​​of any point in the fusion region are derived from the fusion parameters of the mesh vertices through bilinear interpolation.

[0016] Furthermore, the specific operations of step S6 include: S61. Calculate the signed orthogonal distance from each point in the original point cloud to its corresponding fitted surface. For different types of fitted surfaces, such as planar, quadratic, and cylindrical surfaces, appropriate analytical or numerical calculation methods are adopted respectively. The sign of the signed orthogonal distance is determined according to the relative position of the point and the local normal vector of the corresponding fitted block. S62. For quadratic surfaces and cylindrical non-planar fitted surfaces, a numerical optimization algorithm is used to solve for the projection points from the points to the corresponding surfaces, and the signed orthogonal distance is accurately calculated based on the projection points. S63. For the block boundary point, calculate the signed orthogonal distance to the fitted surface of the current block and the neighboring blocks within the neighborhood range, and select the signed orthogonal distance with the smallest absolute value and retaining the original sign as the final distance of the point. S64. Divide the point cloud into multiple error regions according to the preset error threshold, mark each region with color visualization, and realize the above distance calculation and error analysis through multi-process parallel method. Realize the quantitative evaluation of flatness based on the distribution statistics of signed orthogonal distance.

[0017] Furthermore, the color precision point cloud generated in step S7 is color-coded based on the distance classification results from step S64, with signed orthogonal distances. The dots are marked in blue. The points are marked in green. The dots are marked in red; The output flatness statistics report includes at least one of the following information: point cloud distance distribution statistics, effective block ratio, fitted surface type statistics, and out-of-limit point ratio. The color precision point cloud supports thinning and compression processing of point clouds in well-fitted areas.

[0018] The beneficial effects of this invention are as follows: (1) By using an adaptive block-segmentation strategy based on local features of point cloud, the block size is dynamically adjusted, avoiding the overfitting or underfitting problems caused by traditional fixed-size block-segmentation. At the same time, the number of point clouds in each sub-region is controlled within a reasonable range, effectively reducing computational complexity and significantly improving the efficiency and scalability of large-scale point cloud data processing. (2) Based on the geometric features of different sub-regions, a hybrid fitting of plane, cylindrical or quadric surface is adaptively selected. Compared with single model fitting, it fits the complex local geometry of the asphalt panel better, reduces fitting residuals, and improves the accuracy and robustness of surface fitting. At the same time, the stability of quadric surface fitting is ensured by fitting degradation back-off mechanism. (3) Handling invalid blocks that fail to fit or have insufficient points. Neighborhood inverse distance weighted interpolation is used as the first choice for repair. If it fails, global plane fitting is used as a fallback to ensure that all blocks can generate effective surfaces. This completely avoids the problems of empty blocks and surface breaks that occur in traditional methods and ensures the integrity of the overall surface model. (4) By using the adjacent surface parameter fusion strategy, the spatial weight factor in the fusion area is calculated, and the surface parameters of adjacent blocks are weighted and averaged to achieve a smooth transition at the block boundary. The neighborhood minimum distance strategy is adopted for the boundary points, and the signed orthogonal distance with the smallest absolute value is selected as the final result, which effectively eliminates the error mutation at the block joint and greatly improves the smoothness and continuity of the overall surface model. (5) For different types of fitted surfaces, adopt appropriate analytical or numerical calculation methods to accurately calculate the signed orthogonal distance from the point cloud to the fitted surface; and classify the distance results based on the preset error threshold, and intuitively distinguish the regions of different error levels through color visualization marks. At the same time, combine the distance distribution statistics to generate a flatness quantitative evaluation report, providing a high-precision and quantifiable reliable basis for judging the engineering quality. (6) The multi-process parallel processing method is adopted in key steps such as surface fitting and distance calculation, which greatly improves the data processing speed. At the same time, the output surface mesh model, color precision point cloud and flatness statistical report can be directly used for construction quality control and feedback adjustment, which perfectly meets the actual needs of rapid evaluation and decision-making on the engineering site. Attached Figure Description

[0019] Figure 1 A flowchart of the method for point cloud segmentation reconstruction and flatness evaluation of asphalt panels provided in an embodiment of the present invention; Figure 2 A schematic diagram illustrating the principle of multi-type surface hybrid fitting provided in an embodiment of the present invention; Figure 3 A schematic diagram of the original point cloud of the asphalt panel during the construction stage of the pumped storage power station asphalt panel according to an embodiment of the present invention. Figure 4 This is a schematic diagram of the original point cloud of panel two during the construction stage of the asphalt panel of a pumped storage power station, provided in an embodiment of the present invention. Figure 5 A color visualization of point cloud reconstruction and flatness assessment of asphalt panel provided in an embodiment of the present invention; Figure 6 This is a color visualization of the point cloud reconstruction and flatness assessment of an asphalt panel provided in an embodiment of the present invention. Detailed Implementation

[0020] The technical solution of the present invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are not all embodiments of the present invention. All other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0021] It should be noted that, unless otherwise specifically stated, the relative arrangement and numerical expressions of the components and steps described in these embodiments should not be construed as limiting the scope of the invention.

[0022] The following description of exemplary embodiments is merely illustrative and is not intended to limit the invention or its application or use in any way. Techniques, methods, and apparatus known to those skilled in the art may not be discussed in detail herein, but where applicable, such techniques, methods, and apparatus should be considered part of this specification.

[0023] Example 1 See Figure 1 , Figure 1 This is a flowchart of a point cloud segmentation reconstruction and flatness assessment method for asphalt panels proposed in this invention. It is applicable to high-precision geometric inspection and quality acceptance of structures such as asphalt concrete panels in the upper and lower reservoirs of pumped storage power stations. It can also be extended to fields with high surface flatness requirements, such as water conservancy and hydropower projects, transportation engineering, and industrial flooring. Specific steps may include: S1. Point Cloud Data Acquisition and Preprocessing: Acquire point cloud data from the asphalt panel surface using lidar, extract 3D coordinate information, and perform preprocessing to determine the spatial range of the detection area. LiDAR can be a fixed, mobile, or handheld scanning device. Select an appropriate resolution and sampling interval based on the detection task requirements to ensure the ability to capture minute surface changes. Specifically, this includes: S11. Use fixed, mobile or handheld lidar sensors to scan and collect three-dimensional point cloud data of the asphalt panel surface, and combine the attitude sensor IMU and the positioning system GNSS / RTK to synchronously obtain spatial pose information to achieve point cloud spatial registration and extract the three-dimensional coordinate information of each point. S12. Use at least one of statistical filtering and radius filtering to remove noise points; remove isolated points and abnormal height differences to reduce errors caused by environmental factors, equipment vibration, etc. S13. The point cloud data is downsampled using the voxel grid method to compress redundant data, thereby reducing the amount of data while preserving the surface geometric features. S14. Trim the point cloud according to the boundary range set by the detection task to determine the spatial range of the detection area, remove irrelevant area data, and ensure processing efficiency and accuracy; S15. Convert the point cloud data to the engineering coordinate system or the local coordinate system to complete the coordinate system, providing a consistent spatial reference for subsequent surface fitting and error analysis.

[0024] After processing in step S1, high-precision, low-noise, and limited-range three-dimensional point cloud data of the asphalt panel surface are obtained, laying the data foundation for subsequent surface fitting and deformation analysis.

[0025] S2. Adaptive Block Partitioning and Indexing: Based on local point cloud features, the block size is dynamically adjusted to divide the detection region into several sub-regions, establishing a mapping relationship between points and blocks and generating a block index table; specific operations include: S21. Based on the spatial boundary of the point cloud , Divide the point cloud into regular grid blocks; S22. Calculate the initial number of blocks based on the preset block size T. The specific formula is as follows: , .

[0026] S23. Dynamically adjust the block size based on at least one local feature of the point cloud in each initial grid, such as point density and curvature change, to generate a sub-region that adapts to the local features. S24. Assign each point to the corresponding sub-region and establish a mapping relationship from point to block to facilitate subsequent block processing, surface fitting or statistical analysis. S25. Generate a unique identifier for each sub-region and establish a block index table containing regional feature information to enable fast access and retrieval.

[0027] S3, Multi-type Surface Hybrid Fitting: See also Figure 2 For different sub-regions, a suitable surface type is selected for fitting based on their geometric features. The surface coefficients of each sub-region are obtained, and the fitting operation is performed in parallel across multiple processes, with the fitting results statistically analyzed. Specifically, this includes: S31. Perform point count determination and sampling for each block: For each block, determine whether the number of point clouds it contains is not less than the preset threshold MIN_TILE_POINTS. If it meets the requirement, randomly sample from the point cloud set of that block. The number of samples shall not exceed the preset threshold PLANE_SAMPLE. Use the sampled point set for subsequent surface fitting. If it does not meet the requirement, mark the block as an invalid block and include it in the repair process of step S4. S32. Perform feature analysis on the sampling points and make a preliminary judgment on the surface type tendency: Calculate the centroid of the sampling point set and perform SVD / PCA decomposition to obtain singular values / eigenvalues. ) and the corresponding feature vectors ( ); Take the smallest eigenvalue Corresponding feature vector As the initial normal and unify legal direction Calculate the plane constant term satisfy ; The surface type tendency is initially determined based on the ratio of singular values ​​to eigenvalues. The surface type tendency includes planar tendency, cylindrical tendency, and quadratic surface tendency.

[0028] like (Approaching the plane): Prepare a plane solution.

[0029] like (Approaching the cylinder): Preparing the cylindrical solution (axis direction) ).

[0030] Otherwise (no obvious dominant type): prepare a quadratic surface solution.

[0031] S33. Surface Type Adaptation and Robust Optimization: If the surface is determined to be planar, the planar parameters obtained from feature analysis are directly used. The fitting is completed by calculating the orthogonal distance residuals from the point to the plane; If the slope is determined to be cylindrical, the axis direction (axis direction ≈ V3) and point cloud centroid (centroid ≈ point cloud centroid) obtained from feature analysis are used as the initial solutions. The axis position and radius are optimized, and the absolute value of the signed orthogonal distance residual is minimized. Complete the fitting; the objective function is the equation of a cylinder (or a cylinder in a specific direction). If the surface is determined to be a quadric surface with a tendency to dip, the planar solution obtained from the characteristic analysis is converted into the initial equation of the quadric surface. To obtain the initial parameters Minimize the absolute value of the signed orthogonal distance residual. Fitting complete; target surface is: ; The degradation / failure handling specifically includes: if the selected surface type is a quadratic surface, and the coefficients of the quadratic term after optimization are simultaneously lower than the preset threshold SURFACE_EPS, then it is determined to be a fitting degradation, and the quadratic surface fitting result is reverted to the planar solution obtained by feature analysis.

[0032] S34. Parallel Execution and Statistics: The operations of S31-S33 are executed in parallel on all blocks through a multi-process pool. The final fitted surface type (plane / cylindrical / quadratic surface) and fitting results of each block are recorded. The number of effective surfaces, the number of fitted degradations (the number of times it reverts to the plane), and the number of cylindrical fittings are counted.

[0033] S4. Block Fitting Repair and Cusp Handling: Invalid blocks that fail to fit or have insufficient points are repaired using neighborhood interpolation. Invalid blocks that fail to interpolate are handled using global planar fitting as a fallback. The repair results are then marked and statistically analyzed. Specifically, this includes: S41. For blocks that fail to fit or have insufficient points, first perform neighborhood interpolation repair: for each invalid block, use its center... For reference, by radius The search proceeds step-by-step from 1 to a preset threshold MAX_NEIGHBOR_RADIUS, gradually searching for the center point of the surrounding fitted blocks. Obtain the surface of each fitted block in place Values ​​are used to construct a set of neighborhood support points; S42. Construct a weighted least squares model based on inverse distance weights, solve the model to obtain the surface parameters of invalid blocks to complete the interpolation repair, the specific expression is:

[0034] If the number of points in the neighborhood support point set is less than the preset threshold NEIGHBOR_MIN_USED or the weighted least squares model fails to solve, the current interpolation repair will be terminated and the process will be handled by the S43 global fallback process. S43, Global fallback: For invalid blocks where interpolation repair fails, randomly sample from the global point cloud with the number of samples not exceeding the preset threshold GLOBAL_SAMPLE, force the use of planar ODR fitting to obtain global planar coefficients, and complete the global fallback fitting. S44. Repair Result Marking: Mark the number of blocks repaired by interpolation or fallback, and calculate the proportion of effective blocks after repair.

[0035] S5. Adjacent Surface Parameter Fusion: Generate a uniform mesh for each sub-region, determine the boundary points of the blocks and the fusion region, calculate the spatial weighting factor, and perform weighted fusion of the surface parameters of adjacent sub-regions to achieve a smooth transition of the overall surface model; specifically including: S51, Mesh Generation: Generate a uniform mesh for each sub-region (mesh size: ); S52. Boundary Determination: Based on the preset boundary threshold. Determine the merging region around the four edges of the block when the normalized coordinates of the points are... , or When this point is reached, it is determined to be a boundary point; For a parameter is Blocks The fusion region on its right is all normalized coordinates. The points within this region, their final coefficients will be affected by the block division. parameters And its right neighbor parameters The combined effects of these factors.

[0036] S53, Calculate the spatial weighting factor : Define distance For a point within the merge region, calculate its shortest distance to the block boundary. For example, in the right merge region, a point... The range of d is .

[0037] Normalized distance : Distance Normalizing to the [0, 1] interval yields ; according to Calculate spatial weighting factor The spatial weighting factor is calculated using either linear weighting or Gaussian smoothing weighting, and is used to control the fusion ratio of parameters between adjacent blocks; the specific scheme is as follows: Option A: Linear weights

[0038] When the point is located at the boundary ( ), with a weight of 0; when the point is located at the inner edge of the fusion region ( With a weight of 1, a linear transition is achieved.

[0039] Option B: Gaussian smoothed weights

[0040] At the boundary ( The weight is 0; it gradually and smoothly increases to close to 1 within the fusion region. Parameter The decay rate can be flexibly set according to the data characteristics.

[0041] S54. Within the fusion region, the surface parameters of adjacent blocks are weighted and averaged according to the spatial weighting factor, using the following formula:

[0042] in, Corresponding to the current block Part of; Corresponding Adjacency Blocks Part; as points gradually move away from the boundary and enter the block. Internally, its parameters gradually shift towards Transition is necessary to ensure overall continuity.

[0043] S55. Theoretically, a set of fusion parameters can be calculated for each point within the fusion region, but this method is computationally too costly. To improve efficiency, the following strategy is adopted: The fusion region in parameter space Discretize the inner mesh into a finer mesh, and calculate the fusion parameters only for the mesh vertices. During global surface reconstruction, the surface coefficient values ​​at any point within the fusion region are derived from the fusion parameters of the mesh vertices using bilinear interpolation. This method ensures both computational efficiency and maintains smooth continuity at the boundaries.

[0044] S6. Distance Calculation and Flatness Evaluation: Calculate the signed orthogonal distance from the point cloud to the corresponding fitted surface, apply the neighborhood surface distance optimization strategy to the block boundary points, and classify the distance results according to the preset error threshold. S61. Calculate the signed orthogonal distance from each point in the original point cloud to its corresponding fitted surface. For different types of fitted surfaces—planar, quadratic, and cylindrical—adaptive analytical or numerical calculation methods are used. The sign of the signed orthogonal distance is determined based on the relative position of the point and the local normal vector of the corresponding fitted block. Specifically, this includes: Planar blocks: Calculated directly using analytical formulas; Quadratic surface block: The projection point is solved by minimizing the square of the Euclidean distance from the point to the surface; Cylindrical block: The projection point is solved by minimizing the square of the Euclidean distance from the point to the cylindrical model.

[0045] S62. Numerical Optimization of Surface / Cylinder Distance: For quadratic surfaces and cylindrical non-planar fitted surfaces, numerical optimization is employed. The optimization algorithm solves for the projection points from a point to the corresponding surface, and accurately calculates the signed orthogonal distance based on these projection points; specifically, it includes: Quadratic surface: Optimization variables are Substitute into the surface equation to find Construct the squared distance objective function and iterate until convergence; Cylindrical surface: The optimization variable is the axial position. with the angle around the axis The coordinates of the projection points are generated based on the cylindrical parametric equations, the objective function of the squared distance is constructed and iterated until convergence.

[0046] S63. Boundary point distance calculation strategy: For block boundary points, calculate the signed orthogonal distance to the fitted surface of the current block and neighboring blocks within the neighborhood range, and select the signed orthogonal distance with the smallest absolute value and retaining the original sign as the final distance of the point. If the point is located at the block boundary, then in S61, its distance to the current point is calculated additionally. and 8 neighbors The signed orthogonal distance of the (3×3 neighborhood) model is used, and the one with the smallest absolute value (with the sign retained) is selected as the final result to reduce the error of the block joint.

[0047] S64. Divide the point cloud into multiple error regions according to a preset error threshold, mark each region with color visualization, and implement the above distance calculation and error analysis in a multi-process parallel manner. Quantitatively evaluate the flatness based on the distribution statistics of signed orthogonal distances; divide the points into three categories according to the error threshold t: Blue (Distance to the lower part of the plane exceeds the limit); Green (Good); Red (distance to the top of the plane exceeds the limit); Supports error statistical analysis, thinning and compression functions (the green dot indicates the retention ratio r).

[0048] Multi-process parallel acceleration: By executing block fitting and error calculation tasks in parallel through a process pool, the overall processing speed is significantly accelerated.

[0049] S7. Output of test results: Generate a curved mesh model of the asphalt panel, a color precision point cloud, and a flatness statistical report; the output flatness statistical report includes at least one of the following information: point cloud distance distribution statistics, effective block ratio, fitted surface type statistics, and proportion of out-of-limit points.

[0050] Example 2 This embodiment uses the point cloud segmentation reconstruction and flatness evaluation method for asphalt panels proposed in this invention to verify the effectiveness of the method. The specific steps include: B1. Point cloud reading and preprocessing: B11, See also Figure 3 As shown, raw point cloud data is imported from the measuring device, where... Figure 3 , Figure 4 These are the point cloud maps for Panel 1 and Panel 2, respectively. The point cloud data contains approximately 1,634,567 points, covering an area with a span of approximately 126.75m in the X direction and approximately 177.10m in the Y direction. B12. Perform format conversion, outlier removal, and normalization on the point cloud to eliminate data noise and format differences, ensuring the stability of subsequent fitting.

[0051] B2. Blocking and Adaptive Blocking: B21. Set the initial tile size to 2m×2m, resulting in 64×89=5696 initial tiles; B22. Dynamically adjust the block size based on the point density and local height fluctuation characteristics within each tile: If the local point density is high or the surface undulation is complex, it is subdivided into 2×2 sub-blocks (the size of the sub-block is halved, for example, 1m×1m). If the local points are sparse and the terrain is flat, then they are merged into a larger one. (e.g., 4m×4m) to enhance fitting stability and reduce computational cost.

[0052] B3. Segmented Surface Hybrid Fitting: B31. Sampling and Point Count Determination: For each tile, determine whether the number of point clouds it contains is greater than or equal to the preset threshold MIN_TILE_POINTS=10; If the point requirement is met, from this Random sampling is performed on the point cloud, with the number of samples not exceeding PLANE_SAMPLE=5000; if the number of sampling points is less than the minimum value, it is marked as an invalid block and included in the subsequent interpolation repair process.

[0053] B32. Eigenvalue decomposition and preliminary determination of surface type: B321. Calculate the centroid of the sampled point set and perform SVD / PCA decomposition to obtain singular values. and corresponding feature vectors ; B322, Find the minimum singular value Corresponding feature vector As the initial normal And unify the normal direction; B323. Determining the type tendency of a surface by the ratio of singular values: like If it is close to a plane, then proceed to plane fitting; like (Approaching the cylindrical surface) then proceeds to cylindrical fitting; Otherwise (if it is clearly curved), it is determined to be a quadratic surface tendency.

[0054] B33. Surface Fitting and Robust Optimization: Based on the initial classification, the least squares method is used for parameter optimization: Plane fitting: Calculate and minimize the orthogonal distance residuals from a point to a plane.

[0055] Cylindrical fitting: using the axial direction ( The objective function is the sum of squared orthogonal distances between cylindrical surfaces, which optimizes the position and radius of the point cloud distribution axis.

[0056] Quadratic surface fitting: Initial equation To obtain the initial parameters We employ weighted least squares optimization. If the coefficients of the quadratic terms after optimization are simultaneously lower than SURFACE_EPS = 1e-3, then the solution is considered degenerate and regresses to a plane solution.

[0057] B34. Degradation and Statistics: If the fit fails or degenerates, revert to a lower-order model (e.g., quadratic surface degenerates into a plane). Record the final surface type (plane / cylindrical / quadratic surface) and residual statistics for each tile.

[0058] B4. Interpolation and fallback repair: B41. For degradation or ineffectiveness Neighborhood interpolation is used: B42. Search for the maximum neighborhood radius Use at least 5 neighboring blocks (NEIGHBOR_MIN_USED=5); B43. Calculate the weighted average of the neighborhood surfaces, with the weights based on the inverse distance ratio (minimum weight NEIGHBOR_WEIGHT_EPS = 1e-6). B44. After repair, the effective block ratio increased from 53.22% to 94.94%; B45. Use global random sampling (GLOBAL_SAMPLE=1,000,000) to perform surface blending across the entire domain.

[0059] B5. Boundary Blending and Reconstructed Mesh Generation: After the segmented surface fitting is completed and interpolation or fallback repair is performed, the surface parameters of adjacent segments are further blended to ensure the smoothness and continuity of the overall surface model at the segment boundaries, and a global mesh is generated. This includes the following steps: B51 Mesh Generation: Generates a uniform mesh for each sub-region. The mesh size can be set according to application requirements, for example, 0.1m × 0.1m. This mesh provides the basis for subsequent parameter discretization and interpolation.

[0060] B52 Boundary Determination: Determine the boundary regions of the blocks based on a preset boundary threshold M (e.g., M=0.2); When the normalized coordinates of a point satisfy ,or When the boundary point is determined, it is considered a boundary point.

[0061] B53. Calculate the spatial weighting factor : Define distance For a point within the merge region, calculate its shortest distance to the block boundary. For example, in the right merge region, a point... The range of d is ; Normalized distance : Distance Normalization to interval Based on this, the weighting factor is defined. This is used to control the fusion ratio of parameters between adjacent blocks; Using linear weights:

[0062] When the point is located at the boundary ( ), with a weight of 0; when the point is located at the inner edge of the fusion region ( 1) With a weight of 1, a linear transition is achieved.

[0063] B54. Parameter Fusion Calculation: Within the fusion region, based on the weighting factor Weighted average of parameters from adjacent blocks:

[0064] in: Corresponding to the current block Part of; Corresponding Adjacency Blocks Part; as points gradually move away from the boundary and enter the block. Internally, its parameters gradually shift towards Transition is necessary to ensure overall continuity.

[0065] B55. Actual Implementation Method: The fusion region in parameter space The internal discretization is performed into a finer mesh; Calculate fusion parameters only for these mesh vertices ; During global surface reconstruction, coefficient values ​​for any point within the region are derived from these vertex parameters using bilinear interpolation.

[0066] B6. Calculation of the distance from a point to a surface: B61. For all points, calculate their distance to the current point. and 8 neighbors The signed orthogonal distance of the (3×3 neighborhood) model was calculated, and the result with the smallest absolute value (preserving the sign) was selected as the final result, yielding the statistical distribution: Points within a distance of [-0.01, 0.01] m accounted for 97.61%; Points larger than 0.1m account for only 0.61%, indicating that the method has high fitting accuracy.

[0067] B62. The final output is a point cloud file with color markings, where: Blue dots represent negative distance points (below the fitted surface); Green dots represent points near zero distance (good fit regions); The red dots represent positive distance points (above the fitted surface).

[0068] After completing point cloud reconstruction and flatness assessment, the output color visualization effect in this embodiment is as follows: Figure 5 and Figure 6 The images shown are the evaluation results for panel one and panel two, respectively. Blue dots represent points below the fitted surface that exceed the limit, green dots represent areas with good fit, and red dots represent points above the fitted surface that exceed the limit, which visually reflect the flatness of each area of ​​the asphalt panel.

[0069] The above specific embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to examples, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for asphalt panel point cloud block reconstruction and roughness evaluation, characterized in that, include: S1. Collect lidar point cloud data of the asphalt panel surface, extract the three-dimensional coordinate information of the point cloud and perform preprocessing to determine the spatial range of the asphalt panel detection area. S2. Based on the local features of the point cloud, dynamically adjust the block size, divide the detection area into several sub-regions, establish the mapping relationship between points and blocks, and generate a block index table containing regional feature information. S3. Select the appropriate surface type for the geometric features of different sub-regions, obtain the surface coefficients of each sub-region, perform the block fitting operation in parallel through multiple processes, and count the fitting results of each block. S4. Perform neighborhood interpolation repair on invalid blocks that fail to fit or have insufficient point cloud quantity. Use global plane fitting as a fallback method for invalid blocks that fail to interpolate repair. Mark and count the repair results of invalid blocks. S5. Generate a uniform grid for each sub-region, determine the boundary points and fusion regions of the blocks, calculate the spatial weight factor within the fusion region, and perform weighted fusion of the surface parameters of adjacent sub-regions based on the spatial weight factor. S6. Calculate the signed orthogonal distance from each point in the point cloud to the corresponding fitted surface. Use the neighborhood surface distance optimization strategy to determine the final distance for the block boundary points. Classify the signed orthogonal distance results according to the preset error threshold. Complete the quantitative evaluation of asphalt panel flatness based on distance distribution statistics. S7. Generate a curved mesh model of the asphalt panel, a color precision point cloud, and a flatness statistical report.

2. The asphalt panel point cloud block reconstruction and flatness evaluation method of claim 1, wherein, The specific operations of step S1 include: S11. Use fixed, mobile or handheld lidar sensors to scan and collect point cloud data on the surface of the asphalt panel. Combine the attitude sensor IMU and the positioning system GNSS / RTK to synchronously acquire spatial pose information to achieve point cloud spatial registration and extract the three-dimensional coordinate information of each point. S12. Remove noise points from the point cloud using at least one of statistical filtering and radius filtering methods. S13. Use the voxel grid method to downsample the point cloud data; S14. Crop the point cloud according to the boundary range set by the detection task, extract the point cloud of the target area, and determine the spatial range of the detection area. S15. Convert the point cloud data to the engineering coordinate system or the local coordinate system to complete the coordinate system.

3. The method of claim 1, wherein, The specific operations of step S2 include: S21. Based on the spatial boundary of the point cloud , Divide the point cloud into regular grid blocks; S22、according to the preset block size calculating the preliminary block quantity; S23. Dynamically adjust the block size according to the local features of the point cloud in each initial grid to generate sub-regions that adapt to the local features; S24. Assign each point to the corresponding sub-region and establish a mapping relationship from point to block; S25. Generate a unique identifier for each sub-region and establish a block index table containing regional feature information.

4. The asphalt panel point cloud block reconstruction and roughness evaluation method of claim 1, wherein, Step S3 specifically includes: S31. Perform point count determination and sampling for each block: For each block, determine whether the number of point clouds it contains is not less than the preset threshold MIN_TILE_POINTS. If it meets the requirement, randomly sample from the point cloud set of the block, and the number of samples does not exceed the preset threshold PLANE_SAMPLE. Use the sampled point set for subsequent surface fitting. If it does not meet the requirement, mark the block as an invalid block and include it in the repair process of step S4.

5. The method of claim 4, wherein, Step S3 also includes: S32. Perform feature analysis on the sampling points and make a preliminary judgment on the surface type tendency: Calculate the centroid of the sampling point set and perform SVD / PCA decomposition to obtain singular values / eigenvalues ​​and corresponding eigenvectors; take the eigenvector corresponding to the smallest eigenvalue as the initial normal and unify the normal, and calculate the plane constant term; The surface type tendency is initially determined based on the ratio of singular values ​​to eigenvalues. The surface type tendency includes: planar tendency, cylindrical tendency, and quadratic tendency.

6. The asphalt panel point cloud block reconstruction and flatness evaluation method of claim 5, wherein, The specific operations of step S3 also include: S33. Surface type adaptation and robust optimization: If it is determined to be a planar tendency, the planar parameters obtained from feature analysis are directly used, and the orthogonal distance residual from the point to the plane is calculated to complete the fitting. If it is determined to be a cylindrical inclination, the axis direction and point cloud centroid obtained from feature analysis are used as the initial solution. The axis position and radius are optimized, and the least squares method is used to minimize the absolute value of the signed orthogonal distance residual to complete the fitting. If it is determined to be a quadratic surface tendency, the planar solution obtained from feature analysis is converted into the initial equation of the quadratic surface to obtain the initial parameters. The absolute value of the signed orthogonal distance residual is minimized by the least squares method to complete the fitting. If the coefficients of the quadratic terms after optimization are all lower than the preset threshold SURFACE_EPS, it is determined to be a fitting degradation, and the quadratic surface fitting result is reverted to the planar solution obtained from feature analysis. S34. Parallel Execution and Statistics: The operations of S31-S33 are executed in parallel on all blocks through a multi-process pool. The final fitted surface type and fitting result of each block are recorded, and the number of effective surfaces, the number of fitted degenerate surfaces, and the number of cylindrical fitted surfaces are counted.

7. The asphalt panel point cloud block reconstruction and roughness evaluation method of claim 1, wherein, The specific operations of step S4 include: S41. For invalid blocks, use their centers... For reference, by radius The search proceeds step-by-step from 1 to a preset threshold MAX_NEIGHBOR_RADIUS, gradually searching for the center point of the surrounding fitted blocks. Obtain the surface of each fitted block in place Values ​​are used to construct a set of neighborhood support points; S42. Construct a weighted least squares model based on inverse distance weights, solve the model to obtain the surface parameters of invalid blocks and complete the interpolation repair. If the number of points in the neighborhood support point set is less than the preset threshold NEIGHBOR_MIN_USED or the weighted least squares model fails to solve, then stop this interpolation repair. S43. For invalid blocks where interpolation repair fails, randomly sample from the global point cloud and the number of samples does not exceed the preset threshold GLOBAL_SAMPLE. Force the use of planar ODR fitting to obtain global planar coefficients and complete global fallback fitting. S44. Mark the number of blocks that have been repaired by interpolation or fallback, and calculate the percentage of effective blocks after repair.

8. The method of claim 1, wherein, The specific operations of step S5 include: S51. Generate a uniform grid for each sub-region whose size can be set as needed; S52, according to the preset boundary threshold Determine the merging region around the four edges of the block when the normalized coordinates of the points are... , or When this point is reached, it is determined to be a boundary point; S53. Calculate the shortest distance from a point within the fusion region to the block boundary. , distance Normalization yields ,according to Calculate the spatial weighting factor, wherein the spatial weighting factor is calculated by linear weighting or Gaussian smoothing weighting; S54. Within the fusion region, the surface parameters of adjacent blocks are weighted and averaged according to the spatial weighting factor. S55, merge the region in parameter space The internal discretization is performed into a fine mesh, and the fusion parameters are calculated only for the mesh vertices. During global surface reconstruction, the surface coefficient values ​​of any point in the fusion region are derived from the fusion parameters of the mesh vertices through bilinear interpolation.

9. The method for point cloud segmentation reconstruction and flatness evaluation of asphalt panels according to claim 1, characterized in that, The specific operations of step S6 include: S61. Calculate the signed orthogonal distance from each point in the original point cloud to its corresponding fitted surface. For different types of fitted surfaces, such as planar, quadratic, and cylindrical surfaces, appropriate analytical or numerical calculation methods are adopted respectively. The sign of the signed orthogonal distance is determined according to the relative position of the point and the local normal vector of the corresponding fitted block. S62. For quadric surfaces and cylindrical surfaces that are not planar fitted surfaces, numerical optimization algorithms are used to solve for the projection points from the points to the corresponding surfaces, and the signed orthogonal distance is accurately calculated based on the projection points. S63. For the block boundary point, calculate the signed orthogonal distance to the fitted surface of the current block and the neighboring blocks within the neighborhood range, and select the signed orthogonal distance with the smallest absolute value and retaining the original sign as the final distance of the point. S64. Divide the point cloud into multiple error regions according to the preset error threshold, mark each region with color visualization, and realize the above distance calculation and error analysis through multi-process parallel method. Realize the quantitative evaluation of flatness based on the distribution statistics of signed orthogonal distance.

10. The method for point cloud segmentation reconstruction and flatness evaluation of asphalt panels according to claim 9, characterized in that, The color precision point cloud generated in step S7 is color-coded according to the distance classification results in step S64, with signed orthogonal distances. The dots are marked in blue. The points are marked in green. The dots are marked in red; The output flatness statistics report includes at least one of the following information: point cloud distance distribution statistics, effective block ratio, fitted surface type statistics, and out-of-limit point ratio. The color precision point cloud supports thinning and compression processing of point clouds in well-fitted areas.

Images