Ground point filtering method and system based on geometric consistency two-stage parameter optimization

Abstract

The present application relates to the technical field of point cloud filtering, and proposes a ground point filtering method and system based on geometric consistency two-stage parameter optimization, which can automatically complete the evaluation, screening and determination of ground point filtering parameters based on the geometric structure characteristics of the point cloud itself by constructing a geometric consistency scoring function and combining a two-stage parameter optimization strategy, without human intervention to obtain filtering parameters suitable for the current local area. By narrowing the parameter search range in stages and further determining the optimal parameter combination within a reasonable value range, the blindness of the parameter search process is reduced, the stability and reliability of the parameter optimization are improved, the problems of insufficient parameter evaluation basis and the search process being prone to local optimization in the prior art are solved, and the determined parameters can better adapt to changes in different terrain structures in complex terrain areas, thereby improving the stability and adaptability of the ground point filtering result under complex terrain conditions.

Description

Technical Field

[0001] This invention relates to the field of point cloud filtering technology, specifically to a ground point filtering method and system based on geometric consistency two-stage parameter optimization. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] With the continuous growth in demand for applications such as land resource surveys, topographic mapping, infrastructure planning, and geological disaster monitoring, the construction of Digital Elevation Models (DEMs) based on airborne LiDAR point cloud data has become an important technical means for acquiring surface spatial information. Among these, ground point extraction is a crucial prerequisite for DEM generation, and its accuracy directly affects the realism of the terrain representation and the reliability of subsequent analysis and engineering applications. Especially in mountainous and complex terrain areas, point cloud data often exhibits characteristics such as large terrain undulations, frequent slope changes, fragmented surface structures, and uneven point density distribution, leading to severe mixing of ground and non-ground points.

[0004] Existing ground point filtering methods, when processing point cloud data in complex mountainous areas, often require manual adjustments to key parameters such as cloth simulation filtering (CSF) and other algorithms, particularly regarding cloth resolution, stiffness, iteration count, and slope threshold. Due to the significant differences in terrain features across complex mountainous regions—including gentle slopes, steep slopes, slope breaks, and valleys—the sensitivity and adaptability of filtering parameters vary considerably across different areas. A single set of globally uniform parameters is difficult to apply to all terrain types simultaneously. Using uniform parameters can easily lead to missed ground points in steep slopes and slope breaks, introduce non-ground points in gentle areas or low-vegetation zones, and potentially cause discontinuities in ground structure at block boundaries. Furthermore, existing parameter selection processes rely heavily on manual experience or observation of filtering results, resulting in significant differences in parameter configurations provided by different operators, leading to insufficient repeatability and stability of ground point extraction results. On the other hand, existing parameter optimization methods often employ single-index or single-stage search approaches. When the number of parameter combinations is large, it is difficult to fully cover the parameter range, and these methods are easily influenced by initial parameters or random sampling results, resulting in unstable parameter configurations. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a ground point filtering method and system based on geometric consistency two-stage parameter optimization. This method fully utilizes the geometric structural features of the point cloud itself to automatically determine the ground point filtering parameters, enabling the parameter configuration to adapt to complex terrain changes. While ensuring the rationality of the ground structure, it also improves the stability and reliability of the ground point extraction results.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: The first aspect of this invention provides a ground point filtering method based on geometric consistency two-stage parameter optimization, comprising the following steps: Acquire point cloud data of the area to be processed, and divide the point cloud data into multiple local areas; Based on the geometric structure features of point clouds in each local region, a candidate space for ground point filtering parameters is constructed for each local region; Based on the geometric consistency of the ground shape, a geometric consistency scoring function is constructed; Based on the geometric consistency scoring function, a two-stage parameter optimization is performed on the parameter candidate space of each local region: first, filter parameter samples are selected in the parameter candidate space, ground point filtering is performed, and the scoring result is obtained through the geometric consistency scoring function. The reasonable value range of each filter parameter to be optimized is selected as the parameter search subspace. Then, the optimal parameter combination is determined in the parameter search subspace to obtain the final ground point filtering result.

[0007] A further technical solution, based on the geometric structure features of the point cloud in each local region, constructs a candidate space for ground point filtering parameters for each local region, including the following steps: Extract geometric indicators that characterize the topography and point cloud distribution features of the region; Based on the extracted geometric indices, the terrain complexity evaluation value of the local area is calculated, and the global feasible range of each filter parameter to be optimized is adaptively adjusted according to the terrain complexity evaluation value to obtain the parameter candidate space corresponding to each local area.

[0008] A further technical solution involves constructing a geometric consistency scoring function based on the geometric consistency of the ground shape. This geometric consistency scoring function includes multiple geometric consistency evaluation indicators and geometric anomaly penalty items.

[0009] Further technical solutions include geometric consistency evaluation indicators such as ground point connectivity evaluation indicators, slope continuity evaluation indicators, ground surface smoothness evaluation indicators, and low-level consistency evaluation indicators. In the ground point set obtained after ground point filtering, the proportion of points contained in the largest connected subset to the total number of ground points is used as the ground point connectivity evaluation index. The degree of deviation of the local slope angle of each ground point from the regional average slope angle is used as an evaluation index for slope continuity in the ground point set obtained after ground point filtering. The dispersion of the local elevation gradient amplitude of each ground point in the ground point set obtained after ground point filtering is used as the evaluation index of ground surface smoothness. The degree of deviation of the elevation of each ground point from the local elevation range in the ground point set obtained after ground point filtering is used as a low-level consistency evaluation index.

[0010] A feasible technical solution is to, for the set of ground points G(x) corresponding to the parameter combination x, for each ground point in the set of ground points... In its neighborhood Calculate the ratio of elevation difference to distance between the point and all neighboring surface points, and take the average of these ratios as the ground point value. The magnitude of the local elevation gradient.

[0011] Another feasible technical solution is to statistically analyze the elevation values ​​of various ground points in the set G(x) corresponding to the parameter combination x. The local elevation range is represented by quantiles, which means that in the ground point elevation sequence, the elevation value corresponding to the lower quantile is used as the local low reference value, and the elevation value corresponding to the higher quantile is used as the local upper limit reference value. The local elevation range is the range between the local low reference value and the local upper limit reference value.

[0012] A further technical solution involves performing two-stage parameter optimization on the parameter candidate space based on a geometric consistency scoring function for each local region, including: Phase 1: Within the parameter candidate space, generate multiple Phase 1 parameter samples according to the first sampling step size. Perform ground point filtering on each Phase 1 parameter sample to obtain the corresponding ground point set, and calculate the corresponding score based on the geometric consistency scoring function. Based on the scoring results of each Phase 1 parameter sample, select reasonable value ranges for each parameter to be optimized and construct a parameter search subspace. The second stage involves generating multiple second-stage parameter samples within the parameter search subspace determined in the first stage, using a second sampling step size smaller than the first sampling step size. Ground point filtering is then performed on each second-stage parameter sample to obtain a corresponding set of ground points, and the corresponding score is calculated based on the geometric consistency scoring function. Based on the scoring results of each second-stage parameter sample, the optimal parameter combination is determined as the ground point filtering parameters for the current local area.

[0013] A further technical solution, the process of constructing the parameter search subspace in the first stage, includes the following steps: For each local region, sampling is performed within the parameter candidate space, covering the value range of each parameter, using the set first sampling step size, to obtain multiple sets of first-stage parameter samples. For each set of first-stage parameter samples generated in the first stage, the corresponding ground filtering algorithm is used to perform ground point filtering processing on the local area to obtain a set of ground points, and the geometric consistency score is calculated based on the geometric consistency scoring function to form the first-stage scoring result set; The parameters are sorted from highest to lowest according to the comprehensive geometric consistency score corresponding to each first-stage parameter sample; according to the preset screening rules, several groups of parameter samples with scores higher than the set requirements are selected from the sorted scores to construct a high-scoring parameter set. For the set of high-scoring parameters, the parameter values ​​corresponding to each filter parameter are extracted, and the distribution range of each filter parameter in the set of high-scoring parameters is statistically analyzed. Based on the statistical results, the initial upper and lower bounds of each parameter dimension are determined. A preset buffer is added to the initial upper and lower bounds to obtain the reasonable value range of each parameter dimension. The reasonable value ranges of each parameter are combined to construct a parameter search subspace.

[0014] A second aspect of the present invention provides a ground point filtering system based on geometric consistency two-stage parameter optimization, comprising: The segmentation module is configured to acquire point cloud data of the area to be processed and divide the point cloud data into multiple local areas. The initialization module is configured to construct a candidate space for ground point filtering parameters for each local region based on the geometric structure features of the point cloud within each local region. The building module is configured to construct a geometric consistency scoring function based on the ground shape; The two-stage parameter optimization module is configured to perform two-stage parameter optimization based on the geometric consistency scoring function for the parameter candidate space of each local region: first, filter parameter samples are selected in the parameter candidate space, ground point filtering is performed, and the scoring result is obtained through the geometric consistency scoring function. The reasonable value range of each filter parameter to be optimized is selected as the parameter search subspace. Then, the optimal parameter combination is determined in the parameter search subspace to obtain the final ground point filtering result.

[0015] The third aspect of the present invention provides a ground point filtering system based on geometric consistency two-stage parameter optimization, comprising: a point cloud data acquisition terminal and a processor; Point cloud data acquisition terminal, used to acquire point cloud data of the area to be processed; The processor is configured to perform the steps of the ground point filtering method described above, which is based on geometric consistency two-stage parameter optimization.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention, by constructing a geometric consistency scoring function and combining it with a two-stage parameter optimization strategy, can automatically evaluate, screen, and determine ground point filtering parameters based on the geometric structural features of the point cloud itself. It obtains filtering parameters suitable for the current local area without human intervention, thus solving the problems of parameter selection relying on human experience and low automation in existing technologies. Furthermore, by narrowing the parameter search range in stages and further determining the optimal parameter combination within a reasonable value range, the invention reduces the blindness of the parameter search process, improves the stability and reliability of parameter optimization, and solves the problems of insufficient parameter evaluation basis and the search process easily getting trapped in local optima in existing technologies. The determined parameters can better adapt to the changes in different terrain structures within complex terrain areas, thereby improving the stability and adaptability of ground point filtering results under complex terrain conditions.

[0017] The advantages of the present invention, as well as its additional advantages, will be described in detail in the following specific embodiments. Attached Figure Description

[0018] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute a limitation thereof.

[0019] Figure 1 This is a schematic diagram of the overall process of the ground point filtering method based on geometric consistency two-stage parameter optimization according to Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of the geometric consistency score calculation process in Embodiment 1 of the present invention; Figure 3 This is a schematic diagram of the parameter candidate space construction process in Embodiment 1 of the present invention; Figure 4 This is a schematic diagram of the two-stage parameter optimization process in Embodiment 1 of the present invention. Detailed Implementation

[0020] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0021] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. 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 invention pertains.

[0022] It should be noted that the terminology used herein is for describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof. It should be noted that, without conflict, the various embodiments and features within those embodiments can be combined with each other. The embodiments will now be described in detail with reference to the accompanying drawings.

[0023] Example 1 In one or more of the technical solutions disclosed in the embodiments, such as Figures 1 to 4 As shown, a ground point filtering method based on geometric consistency two-stage parameter optimization includes the following steps: Step 1: Obtain the point cloud data of the area to be processed, and divide the point cloud data into multiple local areas; Step 2: Based on the geometric structure features of the point cloud in each local region, construct a candidate space for ground point filtering parameters for each local region; Step 3: Construct a geometric consistency scoring function based on the geometric consistency of the ground shape; Step 4: Based on the geometric consistency scoring function, perform two-stage parameter optimization for the parameter candidate space of each local region: First, select filter parameter samples in the parameter candidate space, perform ground point filtering, and obtain the scoring results through the geometric consistency scoring function. Then, select the reasonable value range of each filter parameter to be optimized as the parameter search subspace. Finally, determine the parameter combination with the best score in the parameter search subspace to obtain the final ground point filtering result.

[0024] This embodiment achieves automatic determination of ground point filtering parameters by constructing a parameter candidate space for each local region and performing a two-stage optimization of the filtering parameters based on a geometric consistency scoring function. Since both parameter evaluation and parameter search are based on the geometric structural features of the point cloud itself, there is no need to rely on repeated adjustments of parameters based on human experience, thus solving the problems of manual parameter configuration and low automation in existing technologies. Simultaneously, by first using the first-stage search to determine the parameter search subspace for each parameter to be optimized, and then performing the second-stage optimization within the parameter search subspace, the instability caused by blind searching in a large parameter space can be reduced. This improves the problems in existing technologies where parameter search is easily affected by initial conditions, prone to getting trapped in local optima, and difficult to obtain stable and reliable parameter results. Furthermore, since the determined parameters are optimized based on the geometric structural features of local regions, they can better adapt to the terrain differences between different regions under complex terrain conditions, improving the stability and reliability of ground point filtering results.

[0025] In step 1, point cloud data can be obtained through airborne lidar measurement, vehicle-mounted lidar measurement, ground laser scanning, UAV oblique photogrammetry reconstruction, or other methods that can acquire three-dimensional spatial information of the ground surface. Preferably, it can be obtained through airborne lidar measurement.

[0026] It is feasible to preprocess the acquired point cloud data of the area to be processed, including outlier removal, data format standardization, and data size control: Optionally, a combination of statistical filtering and radius filtering can be used to remove outliers. Specifically, for each point A, the average distance between the current point A and its neighboring points is calculated, and outliers that deviate significantly from the overall distribution are removed based on the statistical distribution of the average distance. Then, based on the preset search radius, the number of points in the neighborhood is counted, and isolated points with fewer points in the neighborhood than the threshold are further removed.

[0027] The above processing can reduce the interference of noise points and extreme anomalies on the subsequent evaluation and optimization of filter parameters.

[0028] Optionally, after removing outliers, the point cloud data is uniformly converted into standard XYZ coordinate format data; the coordinate units, coordinate precision, and attribute fields are uniformly processed to ensure the consistency of point cloud data from different sources in subsequent calculations.

[0029] For point cloud data with a large volume, while preserving key terrain features such as topographic relief, slope breaks, and valleys as much as possible, sampling or hierarchical retention methods can be used to control the data size, so as to balance computational efficiency and the integrity of terrain representation.

[0030] In step 1, the point cloud data is divided into multiple local regions. This can be done using a regular grid division method based on spatial neighborhood relationships. Specifically, the preprocessed point cloud data is projected onto the XY plane and divided into regular square grids at preset equal intervals. The point cloud in each grid constitutes an independent local region, and a 5% to 10% overlap buffer is set between adjacent grids to avoid breaking the terrain features at the region boundary, so as to preserve the continuous terrain information near the region boundary. Furthermore, the grid size can be set according to the actual point cloud density and terrain complexity. In areas with high point density or drastic terrain changes, the grid size can be appropriately reduced, while in areas with low point density or relatively flat terrain, the grid size can be appropriately increased. This approach allows the subsequent evaluation and optimization of filtering parameters to be performed within a local range, thereby improving the parameters' adaptability to different terrain conditions.

[0031] In step 2, for each local region, filtering parameters are selected as optimization objects for the adopted ground point filtering algorithm, and a reasonable value range is set for each filtering parameter. The candidate space of filtering parameters can be initialized based on the geometric structure features of the point cloud in the current local region to obtain the parameter candidate space for each local region. : ; in, Let i represent the i-th filter parameter to be optimized. For parameter dimensions; They represent the first Within the local area, the first Local lower and local upper bounds for each parameter.

[0032] Specifically, for the cloth algorithm, the filtering parameters to be optimized may include cloth resolution, stiffness, number of iterations, and slope threshold; In step 2, based on the geometric structure features of the point cloud in each local region, a candidate space for ground point filtering parameters is constructed for each local region. It includes the following steps: Step 21: Extract geometric indicators that characterize the topography and point cloud distribution features of the region; Specifically, geometric indices include one or more of the following: average point spacing, mean slope, standard deviation of slope, standard deviation of elevation, local elevation difference, surface roughness, and point density.

[0033] Step 22: Calculate the terrain complexity evaluation value of the local area based on the geometric indicators extracted in Step 21. Based on the terrain complexity evaluation value, the global feasible range of each filter parameter to be optimized is adaptively adjusted to obtain the parameter candidate space corresponding to each local region. Specifically, it includes: Step 221: After normalizing the geometric indices of the local area, weight the results to obtain the evaluation value of the terrain complexity of the local area. ; Optionally, one or more of the average point spacing, slope standard deviation, elevation standard deviation, surface roughness, and point density extracted in step 21 can be normalized and then weighted according to preset weights to obtain the terrain complexity evaluation value. ,in The value range of is 0 to 1, and The larger the value, the more complex the terrain in the local area.

[0034] Step 222: Based on the terrain complexity evaluation value The global feasible range of each filter parameter to be optimized is adaptively adjusted to determine the local value range of each filter parameter in the current local region; the method for adaptively adjusting the global feasible range of the filter parameters to be optimized is as follows: Step 2221: For continuous parameters, use the interval contraction method to determine the first... Within the local area, the first Local lower bounds of the parameters and local upper bound ; For continuous parameters, the first... Local lower bounds of the parameters and local upper bound They can be represented as: ; ; in, and They represent the first The global lower bound and global upper bound of the filtering parameters to be optimized. and The preset shrinkage coefficient satisfies , ,and .

[0035] Step 2222: For discrete parameters, a candidate value screening method is used to determine the local candidate subset within the current local region.

[0036] Step 223: Combine the local value intervals or local candidate subsets obtained after adaptive adjustment of each filter parameter to be optimized to form the parameter candidate space corresponding to the current local region. .

[0037] In this embodiment, before the parameter search begins, the global feasible range of each filter parameter to be optimized is locally adaptively adjusted based on the geometric structure features of the point cloud in the current local area to form a parameter candidate space corresponding to the current local area. This limits the parameter search to a reasonable parameter range and avoids the inefficiency and unstable results caused by unconstrained search in the global parameter space.

[0038] To achieve automatic evaluation and selection of ground point filtering parameters without manual intervention or actual ground point annotation, this embodiment constructs a geometric consistency parameter evaluation mechanism based on point cloud geometric structure features. This mechanism is used to quantitatively evaluate the ground point filtering results obtained under different parameter combinations, providing an objective basis for subsequent two-stage parameter selection and determination.

[0039] The geometric consistency parameter evaluation mechanism in this embodiment does not evaluate the classification accuracy of a single point, but rather comprehensively characterizes the spatial continuity, slope consistency, surface smoothness, and relative elevation distribution characteristics of the ground point set in a local area from the perspective of the overall geometric structure rationality.

[0040] In any local area Internally, parameter combinations are used. Perform ground point filtering to obtain a set of ground points. , can be represented as: ; in, Indicates the first One ground point, The number of ground points. .

[0041] In step 3, based on the geometric consistency of the ground shape, a geometric consistency scoring function is constructed. This function includes multiple geometric consistency evaluation indicators and a geometric anomaly penalty term, used to evaluate the geometric rationality of the ground point filtering results. The geometric consistency scoring function is defined as follows:

[0042] in, For the first Geometric consistency evaluation index For the corresponding weights; This is a penalty term for geometric anomalies. For penalty weighting; This refers to the number of geometric consistency evaluation indicators. To ensure comparability between different evaluation indicators, preferably, each geometric consistency evaluation indicator... Normalization to Interval, geometric anomaly penalty term Similarly normalized to .

[0043] Further technical solutions include geometric consistency evaluation indicators such as ground point connectivity evaluation indicators, slope continuity evaluation indicators, ground surface smoothness evaluation indicators, and low-level consistency evaluation indicators. In some embodiments, the proportion of points in the largest connected subset of the ground point set obtained after ground point filtering is used as the ground point connectivity evaluation index. ; Within a local area, the actual ground surface typically exhibits a unified structure with strong spatial continuity. If the parameters are not set appropriately, the ground point results can easily become fragmented or disjointed.

[0044] Specifically, for parameter combinations For the corresponding set of ground points, an adjacency graph is constructed based on the spatial adjacency relationships of the ground points. The connected subset with the most points is then counted, and the ratio of the number of points in the connected subset to the total number of points in the set of ground points is defined as the ground point connectivity evaluation index. The formula is:

[0045] in, This represents the connected subset with the most vertices in the adjacency graph. This indicates the number of points. When the filtering result shows good spatial continuity and low fragmentation of ground points,... The value of approaches 1.

[0046] Specifically, an adjacency graph is constructed based on the spatial adjacency relationship of ground points. A distance threshold can be set. If the distance between two ground points is less than the distance threshold, the two points are considered to be adjacent. The adjacency graph is formed by connecting each adjacent point in turn. The ground points of the interconnected subgraphs form a connected subset. In some embodiments, the degree of deviation of the local slope angle of each ground point from the regional average slope angle in the ground point set obtained after ground point filtering is used as an evaluation index of slope continuity. ; To constrain the geometric rationality of ground points on the slope, a slope continuity evaluation index based on local slope changes is introduced.

[0047] Specifically, for each ground point in the set of ground points In its neighborhood The local plane is fitted using principal component analysis (PCA) or least squares method to obtain the corresponding unit normal vector. The point is defined according to the unit normal vector. The local slope angle is: ; in, To prevent extremely small positive numbers from being divided by zero; The average slope angle of ground points within the region can be expressed as the mean or median as follows: ; Based on the degree of dispersion of local slope angles, the slope continuity evaluation index can be defined as: ; The `clip(.)` option restricts the result to the range of 0 to 1. Another feasible technical solution is to construct slope continuity evaluation indicators in an exponential form. The formula is:

[0048] in, The upper limit of the maximum permissible slope angle, This is a slope scale parameter.

[0049] When the change of ground points on the slope is relatively gentle and continuous The value is relatively large. When the changes of ground points on the slope in the filtering result are relatively gentle and continuous, and the local slope angle is consistent with the overall slope characteristics of the area, the slope continuity evaluation index is relatively high. The value is relatively large; conversely, when there are obvious abrupt changes in slope, local abnormal undulations, or non-ground points mixed in the ground point results, the value is relatively small. The value of is relatively small. By introducing this index, the quality of the filtering parameters can be evaluated from the perspective of slope geometric continuity, thereby improving the rationality and reliability of the automatic parameter selection results.

[0050] In some embodiments, the dispersion of the local elevation gradient magnitude of each ground point in the ground point set obtained after ground point filtering is used as an evaluation index of ground surface smoothness. ; Within a small scale, the real ground typically has a relatively smooth surface morphology. If the parameters are not set appropriately, the filtering results may easily include trees, building edges, or other non-ground points, or omit some real ground points, leading to abrupt changes in local elevation and affecting the geometric rationality of the ground surface. To evaluate the smoothness of the ground point set in the elevation direction, a ground surface smoothness evaluation index is introduced. Specifically, for parameter combinations The corresponding set of ground points For each ground point in the set of ground points In its neighborhood Calculate the ratio of elevation difference to distance between the point and all neighboring surface points, and take the average of these ratios as the ground point value. Local elevation gradient magnitude : ; in, Represents ground point The set of neighborhood points, express Neighboring ground points in the middle, and Representing ground points and neighboring ground points elevation, Represents ground point Its neighboring points The planar or spatial distance between them To prevent extremely small positive numbers with a denominator of zero.

[0051] Based on the dispersion of local elevation gradient amplitudes at various ground points, the surface smoothness evaluation index can be defined as: ; Alternatively, in another feasible approach, a normalized difference form can be used to construct the ground surface smoothness evaluation index as follows: ; in, Represents the set of ground points The set of local elevation gradient magnitudes at various ground points in China. This represents the variance of the set of local elevation gradient magnitudes. For gradient scaling parameters, This is the upper bound parameter for the variance of the local elevation gradient magnitude. This means that the calculation results will be restricted to a specified range.

[0052] When the surface elevation change is relatively smooth and the local elevation gradient change is relatively uniform in the ground point filtering results, the dispersion of the local elevation gradient amplitude of each surface point is low, and the surface smoothness evaluation index is good. The value of is relatively large; conversely, when there are obvious elevation changes, local abnormal fluctuations, or non-ground points mixed in the filtering results, the dispersion of the local elevation gradient amplitude increases, and the ground surface smoothness evaluation index becomes less effective. The value is relatively small.

[0053] When the surface elevation change is relatively smooth and the local gradient change is relatively uniform in the ground point filtering results, the dispersion of the local elevation gradient amplitude of each surface point is low, and the ground surface smoothness evaluation index is good. The value of is relatively large; conversely, when there are obvious elevation abrupt changes, local abnormal fluctuations, or non-ground points mixed in the filtering result, the dispersion of the local elevation gradient amplitude increases. The value of is relatively small. By introducing this index, the quality of the filtering parameters can be evaluated from the perspective of the smoothness of surface elevation changes, thereby improving the rationality and reliability of the automatic parameter selection results.

[0054] In some embodiments, the degree of deviation of the elevation of each ground point in the ground point set obtained after ground point filtering from the local elevation range is used as a low-level consistency evaluation index. ; Within a local area, the actual ground surface is typically located at a relatively low geometric position. If the parameters are not set appropriately, trees, buildings, or other non-ground points higher than the ground can easily be included in the filtering results, leading to an overall higher elevation of the ground point set and reducing the geometric validity of the filtering results. To evaluate whether the ground point set is distributed throughout a locally low-lying area, a low-lying consistency evaluation index is introduced. Specifically, for the set of ground points G(x) corresponding to the parameter combination x, the elevation values ​​of each ground point in the set G(x) are statistically analyzed. To enhance robustness to local abnormal highs and lows, quantiles are used to represent the local elevation range; that is, the elevation value corresponding to the lowest quantile in the ground point elevation sequence is used as the local low reference value. The elevation value corresponding to the highest quantile is used as the reference value for the upper limit of the local elevation. The local elevation range is the area between the local low reference value and the local upper reference value, which can be expressed as: ; in, For lower quantile parameters, a value of 1% to 5% can be used. For high quantile parameters, 95% to 99% can be used.

[0055] Based on the normalized position of each surface point relative to the local elevation range, the low-level consistency evaluation index can be defined as: ; When the overall distribution of ground points is closer to local low-lying areas The larger the value, the better.

[0056] When the overall distribution of ground points in the ground point filtering results is closer to the local low-lying area, the normalized position of each ground point relative to the local elevation range is lower, and the low-lying consistency evaluation index... A larger value indicates a higher elevation; conversely, when the filtered result contains many non-ground points above the ground, the overall elevation of the ground point set is higher. The value of is relatively small. By introducing this index, the quality of the filtering parameters can be evaluated from the perspective of the rationality of the overall elevation position of the ground points, thereby improving the accuracy and stability of the automatic parameter selection results.

[0057] Furthermore, the proportion of outliers or outlier regions in the ground point set obtained after ground point filtering that meet the preset geometric anomaly conditions is used as a geometric anomaly penalty term. ; Within a local area, the actual ground surface typically exhibits continuous, low-lying geometric features with relatively harmonious slope changes. If the parameters are not set appropriately, significant misfiltering can easily occur at locations such as the toe of a slope, steep slopes, building edges, or regional boundaries. For example, ground point results may show large-area breaks, non-ground structures may be mixed into the ground point set, or the local geometry may clearly not conform to the continuous variation of the terrain. To suppress the impact of these anomalies on the parameter evaluation results, a geometric anomaly penalty term is introduced. Specifically, for the set of ground points G(x) corresponding to the parameter combination x, points that satisfy the preset geometric anomaly conditions are defined as anomaly points and form an anomaly point set A(x), wherein the geometric anomaly conditions include at least one of the following: (a) The size of the connected subset to which the ground point belongs is less than the preset minimum connected size threshold; (b) The magnitude of the local elevation gradient at a ground point is greater than the preset gradient threshold; (c) The difference between the local slope angle of a ground point and the average slope angle of its neighborhood is greater than the preset slope change threshold; (d) The ground point is located within the buffer zone of the region and there is a significant discontinuity between it and the ground structure of the adjacent region; Based on the proportion of outlier points in the total set of ground points, the geometric outlier penalty term P(x) can be defined as: ; in, Let G(x) represent the set of outlier points that satisfy the preset geometric anomaly conditions, and let G(x) represent the set of ground points corresponding to the parameter combination x. Indicates the number of points.

[0058] This embodiment introduces a geometric anomaly penalty term to suppress obvious misfiltering phenomena that may occur at locations such as slope toes, steep slopes, building edges, or area boundaries. This is used to penalize large-area breaks, the inclusion of non-ground structures, and anomalies that clearly do not conform to terrain logic in the ground point results. The geometric anomaly penalty term can be defined based on the proportion of anomaly points or anomaly regions in the ground point set, and preferably normalized to a certain value. Interval.

[0059] In step 4, for each local region, a two-stage parameter optimization is performed on the parameter candidate space based on the geometric consistency scoring function, including: Step 41, First Stage: Within the parameter candidate space determined in Step 2, generate multiple first-stage parameter samples according to the first sampling step size. Perform ground point filtering on each first-stage parameter sample to obtain the corresponding ground point set, and calculate the corresponding score based on the geometric consistency scoring function. Based on the scoring results of each first-stage parameter sample, select reasonable value ranges for each parameter to be optimized and construct a parameter search subspace. Step 42, Second Stage: Within the parameter search subspace determined in the first stage, generate multiple second-stage parameter samples with a second sampling step size smaller than the first sampling step size. Perform ground point filtering on each second-stage parameter sample to obtain the corresponding ground point set, and calculate the corresponding score based on the geometric consistency scoring function. Based on the scoring results of each second-stage parameter sample, determine the parameter combination with the optimal score as the ground point filtering parameters for the current local area. The purpose of the first stage of processing in step 41 is to use geometric consistency scores to judge the overall rationality of the ground point geometry generated by different parameter values ​​within the complete parameter candidate space, thereby eliminating obviously unreasonable parameter value ranges, identifying reasonable value intervals in which the geometric consistency scores are relatively concentrated in each parameter dimension, and providing a parameter search subspace for subsequent parameter determination.

[0060] In the first stage, due to the use of a large parameter sampling step size and a limited number of evaluations, the geometric consistency score is mainly used to reflect the geometric rationality distribution characteristics of different parameter values ​​on an overall scale, but it lacks stability in response to small parameter changes. Therefore, the first stage does not aim to obtain the final ground point filtering parameters, but rather to statistically determine the parameter value range, and its output is not directly used as the final filtering parameters.

[0061] The process of constructing the parameter search subspace in the first stage of step 41 includes the following steps: Step 411: For each local region, with the set first sampling step size, sample within the parameter candidate space obtained in step 2, covering the value range of each parameter, to obtain multiple sets of first-stage parameter samples. Specifically, the parameter candidate space includes each filter parameter to be optimized and its corresponding value range. In this step, a larger sampling step size is used to select the parameter combination of ground filtering. Each parameter combination is used as a parameter sample to obtain multiple first-stage parameter samples. Based on uniform sampling, each parameter dimension is covered within the allowed value range to obtain the first-stage parameter sample set.

[0062] Step 412: For each set of first-stage parameter samples generated in the first stage, the corresponding ground filtering algorithm is used to perform ground point filtering processing on the local area to obtain a set of ground points, and the geometric consistency score is calculated based on the geometric consistency scoring function constructed in step 3 to form the first-stage scoring result set.

[0063] For the first-stage parameter sample of group t, each parameter in this group is first assigned to the control parameters corresponding to the ground point filtering algorithm. The ground point filtering algorithm can be a cloth simulation filtering algorithm. Then, the assigned ground point filtering algorithm is used to classify the original point cloud data of the current local area into ground points and non-ground points, extracting the ground point set. ; set of ground points Using these as inputs, the following are calculated: the proportion of the number of points in the largest connected subset of the ground point set to the total number of ground points, yielding the ground point connectivity evaluation index; the dispersion of the local slope angle of each ground point, yielding the slope continuity evaluation index; the dispersion of the local elevation gradient amplitude of each ground point, yielding the ground surface smoothness evaluation index; the normalized position of the elevation of each ground point relative to the local elevation range, yielding the low-level consistency evaluation index; and the proportion of abnormal points or abnormal areas that meet the preset geometric anomaly conditions, yielding the geometric anomaly penalty term.

[0064] Substituting the above evaluation indicators into the geometric consistency scoring function yields the comprehensive score for this set of parameter samples, and the parameter sample and its corresponding score are saved as a scoring record. After completing the calculation for all first-stage parameter samples, the first-stage scoring result set is obtained, which can be represented as: ; in, This represents the parameter sample of the first stage in the t-th group. This represents the overall geometric consistency score corresponding to this set of parameter samples. This represents the total number of parameter samples in the first stage.

[0065] Step 413: For the first-stage scoring result set obtained in step 412 The parameters are sorted from highest to lowest according to their comprehensive geometric consistency scores corresponding to each first-stage parameter sample. Based on preset screening rules, several groups of parameter samples with scores higher than the set requirements are selected from the sorted scores to construct a high-scoring parameter set. ; Specifically, the preset filtering rules can be any one of the following or a combination thereof: Select the top K parameter samples by score ranking; Select scores greater than the preset score threshold Parameter samples; Select parameter samples whose scores fall within the pre-set ratio range.

[0066] The set of parameters for high scores can be represented as: ; This step yields a set of parameter samples that can produce better ground point filtering results within the current local area, which can be used to characterize the distribution characteristics of the better scoring parameters across various parameter dimensions.

[0067] Step 414: For the set of high-scoring parameters obtained in step 413, extract the parameter values ​​corresponding to each filter parameter and count the distribution range of each filter parameter in the set of high-scoring parameters; based on the statistical results, determine the initial upper and lower bounds of each parameter dimension; add a preset buffer amount on the basis of the initial upper and lower bounds to obtain the reasonable value range of each parameter dimension, and combine the reasonable value ranges of each parameter to construct a parameter search subspace. Specifically, for the i-th parameter to be optimized, extract the value sequence of that parameter from the set of high-scoring parameters: ; Where L represents the number of samples of the i-th parameter in the set of high-scoring parameters.

[0068] Determine the initial lower bound of the i-th parameter based on the parameter value sequence. and the initial upper bound Specifically, the initial lower bound and initial upper bound can be determined in any of the following ways: 1) using the minimum and maximum values ​​of the parameter in the set of high-scoring parameters; 2) using the lower and higher quantiles of the parameter; 3) using the range of the mean combined with the standard deviation range to determine the upper and lower bounds.

[0069] After determining the initial upper and lower bounds, a preset buffer amount is introduced. This yields the reasonable range of values ​​for the i-th parameter: ; For the reasonable value ranges of each parameter dimension obtained in step 414, the ranges of each parameter dimension are combined to construct the parameter search subspace corresponding to the current local region, and the second-stage parameter search subspace is used as the output result of the first-stage coarse search; the parameter search subspace can be represented as: ; This step yields a parameter search subspace that is smaller than the complete parameter candidate space but more concentrated in the high-scoring region. This provides parameter range constraints for the second-stage fine search, thereby reducing invalid searches and improving parameter optimization efficiency and stability.

[0070] The first stage of processing in this embodiment involves coarse-grained sampling, filtering, and geometric consistency scoring of parameter samples within the complete parameter candidate space. Based on high-scoring parameter samples, reasonable value ranges for each parameter dimension are determined. This allows for the rapid identification of the distribution range of advantageous parameters that match the terrain features of the current local area without excessively increasing the computational load. This avoids subsequent parameter searches being conducted blindly within a large parameter space, thereby reducing the degree to which parameter searches are affected by random initialization and improving the efficiency, stability, and reliability of automatic parameter selection.

[0071] Within the reasonable parameter search subspace determined in the first stage, the second stage uses denser parameter sampling and geometric consistency scoring to finely compare the consistency and stability of ground point geometry under similar parameter conditions, thereby automatically determining the parameter combination for ground point filtering. Compared to the first stage, the second stage employs a smaller parameter sampling step size and a higher evaluation density, making the geometric consistency score more stable in response to small parameter changes, thus ensuring the reliability of the parameter determination results in engineering applications.

[0072] In step 42, a second-stage fine search is performed within the parameter search subspace determined in the first stage to determine the ground point filtering parameters for the current local region. The implementation process of the second stage includes the following steps: Step 421: For the parameter search subspace output in step 41, generate multiple second-stage parameter samples within the parameter search subspace according to a second sampling step size smaller than the first sampling step size; The parameter sampling step size in the second stage is smaller than that in the first stage, in order to improve the accuracy of characterizing parameter changes and score changes within the subspace.

[0073] Step 422: For the second-stage parameter samples obtained in step 421, input each set of parameter samples into the ground point filtering algorithm to perform ground point filtering processing on the point cloud data of the current local area to obtain the ground point set corresponding to each set of parameter samples; then, based on the geometric consistency scoring function constructed in step 3, calculate the comprehensive geometric consistency score for each ground point set to obtain the second-stage scoring result set.

[0074] Specifically, for the t-th group of second-stage parameter samples First, the control parameters of the ground point filtering algorithm are assigned values ​​using parameter samples. Then, the assigned ground point filtering algorithm is used to classify the point cloud data of the current local area into ground points and non-ground points, and the ground point set is extracted. Based on the set of ground points, the evaluation indices for ground point connectivity, slope continuity, ground surface smoothness, low-level consistency, and geometric anomaly penalty are calculated. Then, based on each evaluation index and its corresponding weight, the comprehensive geometric consistency score of the parameter sample is calculated. Repeat the above process for all second-stage parameter samples to obtain the second-stage scoring result set; the second-stage scoring result set can be represented as: ; in, This represents the parameter sample of the second stage in the t-th group. This indicates the corresponding overall geometric consistency score. This represents the total number of parameter samples in the second stage.

[0075] Step 423: Parameter Sample Evaluation. Based on the second-stage scoring result set obtained in Step 422, the comprehensive geometric consistency scores corresponding to each group of parameter samples are compared. The parameter sample with the best score is selected as the target parameter combination for the current local area, and the ground filtering result of the target parameter combination is used as the final ground point filtering result for the current local area.

[0076] A further technical solution is that when multiple sets of parameter samples in the second-stage scoring result set have the same or approximately optimal scores, selection can be made based on preset discrimination rules. The discrimination rules include one or more of the following: priority of parameter stability, priority of smaller geometric anomaly penalty term, or priority of smaller parameter change amplitude.

[0077] In this embodiment, by performing higher-density secondary sampling and scoring evaluation within the high-scoring parameter search subspace determined in the first stage, the optimal parameter combination can be more accurately located on the basis of narrowing the search range, avoiding the problem of excessive computation caused by performing high-density search directly in the complete parameter space. At the same time, since the second-stage search is carried out within the range of advantageous parameters selected in the first stage, it helps to reduce the degree to which parameter search is affected by random sampling and initial conditions, thereby improving the stability, accuracy and reliability of automatic parameter selection results.

[0078] Example 2 Based on Example 1, this example provides a ground point filtering system based on geometric consistency two-stage parameter optimization, including: The segmentation module is configured to acquire point cloud data of the area to be processed and divide the point cloud data into multiple local areas. The initialization module is configured to construct a candidate space for ground point filtering parameters for each local region based on the geometric structure features of the point cloud within each local region. The building module is configured to construct a geometric consistency scoring function based on the ground shape; The two-stage parameter optimization module is configured to perform two-stage parameter optimization based on the geometric consistency scoring function for the parameter candidate space of each local region: first, filter parameter samples are selected in the parameter candidate space, ground point filtering is performed, and the scoring result is obtained through the geometric consistency scoring function. The reasonable value range of each filter parameter to be optimized is selected as the parameter search subspace. Then, the optimal parameter combination is determined in the parameter search subspace to obtain the final ground point filtering result.

[0079] It should be noted that each module in this embodiment corresponds one-to-one with each step in embodiment 1, and their specific implementation process is the same, so it will not be repeated here.

[0080] Example 3 Based on Embodiment 1, this embodiment provides a ground point filtering system based on geometric consistency two-stage parameter optimization, including: a point cloud data acquisition terminal and a processor; Point cloud data acquisition terminal, used to acquire point cloud data of the area to be processed; The processor is configured to perform the steps of the ground point filtering method based on geometric consistency two-stage parameter optimization as described in Example 1.

[0081] Optionally, the point cloud data acquisition terminal can be any one of airborne LiDAR equipment, vehicle-mounted LiDAR equipment, ground-based laser scanning equipment, or UAV oblique photogrammetry acquisition equipment, or a combination of the above devices. Preferably, the point cloud data acquisition terminal is an airborne LiDAR equipment.

[0082] It is feasible to connect the processor and the point cloud data acquisition terminal via wired or wireless communication; the wireless communication method can be Wi-Fi, 4G, 5G or other data transmission methods.

[0083] Furthermore, the processor is connected to a memory containing a computer program. When the processor calls the computer program, it executes the steps of the ground point filtering method based on geometric consistency two-stage parameter optimization as described in Embodiment 1.

[0084] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0085] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A ground point filtering method based on geometric consistency two-stage parameter optimization, characterized in that, Includes the following steps: Acquire point cloud data of the area to be processed, and divide the point cloud data into multiple local areas; Based on the geometric structure features of point clouds in each local region, a candidate space for ground point filtering parameters is constructed for each local region; Based on the geometric consistency of the ground shape, a geometric consistency scoring function is constructed; Based on the geometric consistency scoring function, a two-stage parameter optimization is performed on the parameter candidate space of each local region: first, filter parameter samples are selected in the parameter candidate space, ground point filtering is performed, and the scoring result is obtained through the geometric consistency scoring function. The reasonable value range of each filter parameter to be optimized is selected as the parameter search subspace. Then, the optimal parameter combination is determined in the parameter search subspace to obtain the final ground point filtering result.

2. The ground point filtering method based on geometric consistency two-stage parameter optimization as described in claim 1, characterized in that, Based on the geometric structure features of the point cloud in each local region, a candidate space for ground point filtering parameters is constructed for each local region, including the following steps: Extract geometric indicators that characterize the topography and point cloud distribution features of the region; Based on the extracted geometric indices, the terrain complexity evaluation value of the local area is calculated, and the global feasible range of each filter parameter to be optimized is adaptively adjusted according to the terrain complexity evaluation value to obtain the parameter candidate space corresponding to each local area.

3. The ground point filtering method based on geometric consistency two-stage parameter optimization as described in claim 1, characterized in that, Based on the geometric consistency of the ground shape, a geometric consistency scoring function is constructed, which includes multiple geometric consistency evaluation indicators and geometric anomaly penalty terms.

4. The ground point filtering method based on geometric consistency two-stage parameter optimization as described in claim 3, characterized in that, Geometric consistency evaluation indicators include ground point connectivity evaluation indicators, slope continuity evaluation indicators, ground surface smoothness evaluation indicators, and low-level consistency evaluation indicators. In the ground point set obtained after ground point filtering, the proportion of points contained in the largest connected subset to the total number of ground points is used as the ground point connectivity evaluation index. In the set of ground points obtained after ground point filtering, the degree of deviation of the local slope angle of each ground point from the regional average slope angle is used as the evaluation index of slope continuity. The dispersion of the local elevation gradient amplitude of each ground point in the ground point set obtained after ground point filtering is used as the evaluation index of ground surface smoothness. The degree of deviation of the elevation of each ground point from the local elevation range in the ground point set obtained after ground point filtering is used as a low-level consistency evaluation index.

5. The ground point filtering method based on geometric consistency two-stage parameter optimization as described in claim 4, characterized in that, For the set of ground points G(x) corresponding to the parameter combination x, for each ground point in the set of ground points... In its neighborhood Calculate the ratio of elevation difference to distance between the point and all neighboring surface points, and take the average of these ratios as the ground point value. The magnitude of the local elevation gradient.

6. The ground point filtering method based on geometric consistency two-stage parameter optimization as described in claim 4, characterized in that, For the set of ground points G(x) corresponding to the parameter combination x, count the elevation values ​​of each ground point in the set. The local elevation range is represented by quantiles, which means that in the ground point elevation sequence, the elevation value corresponding to the lower quantile is used as the local low reference value, and the elevation value corresponding to the higher quantile is used as the local upper limit reference value. The local elevation range is the range between the local low reference value and the local upper limit reference value.

7. The ground point filtering method based on geometric consistency two-stage parameter optimization as described in claim 1, characterized in that, For each local region, a two-stage parameter optimization is performed on the parameter candidate space based on the geometric consistency scoring function, including: Phase 1: Within the parameter candidate space, generate multiple Phase 1 parameter samples according to the first sampling step size. Perform ground point filtering on each Phase 1 parameter sample to obtain the corresponding ground point set, and calculate the corresponding score based on the geometric consistency scoring function. Based on the scoring results of each Phase 1 parameter sample, select reasonable value ranges for each parameter to be optimized and construct a parameter search subspace. The second stage involves generating multiple second-stage parameter samples within the parameter search subspace determined in the first stage, using a second sampling step size smaller than the first sampling step size. Ground point filtering is then performed on each second-stage parameter sample to obtain a corresponding set of ground points, and the corresponding score is calculated based on the geometric consistency scoring function. Based on the scoring results of each second-stage parameter sample, the optimal parameter combination is determined as the ground point filtering parameters for the current local area.

8. The ground point filtering method based on geometric consistency two-stage parameter optimization as described in claim 7, characterized in that, The process of constructing the parameter search subspace in the first stage includes the following steps: For each local region, sampling is performed within the parameter candidate space, covering the value range of each parameter, using the set first sampling step size, to obtain multiple sets of first-stage parameter samples. For each set of first-stage parameter samples generated in the first stage, the corresponding ground filtering algorithm is used to perform ground point filtering processing on the local area to obtain a set of ground points, and the geometric consistency score is calculated based on the geometric consistency scoring function to form the first-stage scoring result set; The parameters are sorted from highest to lowest according to the comprehensive geometric consistency score corresponding to each first-stage parameter sample; according to the preset screening rules, several groups of parameter samples with scores higher than the set requirements are selected from the sorted scores to construct a high-scoring parameter set. For the set of high-scoring parameters, the parameter values ​​corresponding to each filter parameter are extracted, and the distribution range of each filter parameter in the set of high-scoring parameters is statistically analyzed. Based on the statistical results, the initial upper and lower bounds of each parameter dimension are determined. A preset buffer is added to the initial upper and lower bounds to obtain the reasonable value range of each parameter dimension. The reasonable value ranges of each parameter are combined to construct a parameter search subspace.

9. A ground point filtering system based on geometric consistency two-stage parameter optimization, characterized in that, include: The segmentation module is configured to acquire point cloud data of the area to be processed and divide the point cloud data into multiple local areas. The initialization module is configured to construct a candidate space for ground point filtering parameters for each local region based on the geometric structure features of the point cloud within each local region. The building module is configured to construct a geometric consistency scoring function based on the ground shape; The two-stage parameter optimization module is configured to perform two-stage parameter optimization based on the geometric consistency scoring function for the parameter candidate space of each local region: first, filter parameter samples are selected in the parameter candidate space, ground point filtering is performed, and the scoring result is obtained through the geometric consistency scoring function. The reasonable value range of each filter parameter to be optimized is selected as the parameter search subspace. Then, the optimal parameter combination is determined in the parameter search subspace to obtain the final ground point filtering result.

10. A ground point filtering system based on geometric consistency two-stage parameter optimization, characterized in that, include: Point cloud data acquisition terminal and processor; Point cloud data acquisition terminal, used to acquire point cloud data of the area to be processed; The processor is configured to perform the steps of the ground point filtering method based on geometric consistency two-stage parameter optimization as described in any one of claims 1-8.

Images