Ground filtering method and system based on cloth simulation parameter dynamic optimization

Abstract

The invention relates to the technical field of point cloud data processing, and provides a ground filtering method and system based on cloth simulation parameter dynamic optimization, and the method comprises the following steps: obtaining point cloud data of a to-be-processed region, carrying out the preprocessing, constructing a complexity index, and calculating a complexity index value based on the preprocessed data; clustering the obtained complexity indexes by adopting a clustering method fusing an elbow rule and a spatial neighborhood constraint to obtain a terrain complexity category; and selecting distribution parameters of a distribution algorithm based on the terrain complexity category and the dynamic mapping of the constructed complexity category-optimal parameter group, and further performing ground point filtering and ground extraction by adopting the distribution algorithm. According to the invention, distribution simulation parameter configuration is guided through terrain complexity analysis, accurate adaptation of different terrain areas is realized, and the accuracy and global consistency of ground point extraction are improved.

Description

Technical Field

[0001] This invention relates to the field of point cloud data processing technology, specifically to a ground filtering method and system based on dynamic optimization of fabric simulation parameters. 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 rapid development of LiDAR and UAV remote sensing technologies, acquiring large-scale, high-density, and high-precision 3D point cloud data has become a reality. This type of data demonstrates significant application value in fields such as digital mapping, terrain modeling, geological disaster monitoring, and forestry resource surveys, promoting the integrated development of spatial information technology and geographic information systems. Among these, ground point extraction, as a fundamental step in building digital terrain models (DTMs), directly impacts the overall performance of subsequent terrain analysis, modeling, and geographic decision support systems due to its processing accuracy and efficiency. Especially in mountainous scenarios with large coverage areas and complex terrain structures, traditional point cloud processing methods face serious challenges in balancing accuracy and efficiency due to dramatic elevation changes, significant slope variations, and large differences in surface roughness. Therefore, how to achieve accurate identification and efficient filtering of complex terrain environments has become a key technical challenge in the current field of point cloud processing.

[0004] Currently, commonly used ground point extraction techniques mainly include raster filtering, morphological filtering, and Cloth Simulation Filter (CSF). Among them, the CSF method has been widely adopted in mountainous point cloud processing tasks in recent years due to its good preservation of terrain continuity. However, these algorithms generally rely on fixed or manually set parameters (such as cloth particle size, number of simulation iterations, elevation threshold, etc.), and have the following significant drawbacks when dealing with natural terrain that is uneven in height and complex in shape: (1) Fixed parameters in the cloth algorithm lead to insufficient adaptability. The terrain features of complex mountainous areas exhibit significant spatial heterogeneity, ranging from gentle plateaus to steep valleys, with huge differences in elevation, slope, and surface roughness. Uniformly set parameters cannot dynamically adapt to different landform types: in areas with gentle terrain, it may cause computational redundancy and reduce efficiency, while in areas with rugged terrain, it may result in insufficient extraction accuracy, leading to confusion between ground points and non-ground points and reducing the quality of DTM.

[0005] (2) Limited dimensions for characterizing terrain complexity, leading to unstable classification. Existing methods often rely solely on a single slope index for terrain classification, and the classification process frequently employs non-deterministic clustering strategies such as fixed cluster numbers and random initial centroids. This results in sensitivity to initial conditions, exhibiting instability and weak generalization ability. More importantly, these methods lack the ability to comprehensively express terrain complexity. For example, they struggle to distinguish between "slight hills with similar average elevations but significant undulations" and "plateaus with concentrated elevations," and they also have difficulty identifying areas that are "generally flat but with local micro-topographical undulations." This lack of expression for terrain complexity severely limits the effectiveness of subsequent parameter configuration and regional adaptive processing strategies, thereby affecting the overall point cloud processing quality and efficiency. Summary of the Invention

[0006] To address the aforementioned problems, this invention proposes a ground filtering method and system based on dynamic optimization of fabric simulation parameters. By guiding the configuration of fabric simulation parameters through terrain complexity analysis, it achieves accurate adaptation to different terrain regions and improves the accuracy and global consistency of ground point extraction.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: One or more embodiments provide a ground filtering method based on dynamic optimization of fabric simulation parameters, including the following steps: The point cloud data of the area to be processed is acquired and preprocessed to construct a complexity index. The complexity index value is calculated based on the preprocessed data. A clustering method that integrates the elbow rule and spatial neighborhood constraints is used to cluster the obtained complexity indices to obtain terrain complexity categories; Based on the terrain complexity category and the dynamic mapping of the constructed complexity category-optimal parameter group, the cloth parameters of the cloth placement algorithm are selected, and then the cloth placement algorithm is used for ground point filtering and ground extraction.

[0008] One or more embodiments provide a ground filtering system based on dynamic optimization of fabric simulation parameters, including: The complexity calculation module is configured to acquire point cloud data of the area to be processed, preprocess it, construct a complexity index, and calculate the complexity index value based on the preprocessed data. The clustering module is configured to use a clustering method that combines the elbow rule and spatial neighborhood constraints to cluster the obtained complexity indices and obtain the terrain complexity categories. The mapping extraction module is configured to select the cloth parameters of the cloth algorithm based on the dynamic mapping of the constructed complexity category-optimal parameter group based on the terrain complexity category, and then use the cloth algorithm to perform ground point filtering and ground extraction.

[0009] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the steps in the above-described ground filtering method based on dynamic optimization of fabric simulation parameters.

[0010] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the steps in the above-described ground filtering method based on dynamic optimization of fabric simulation parameters.

[0011] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention significantly improves the recognition accuracy of complex terrain areas through multi-dimensional terrain feature quantization and clustering, making it particularly suitable for mountainous or forested scenes with significant elevation undulations and local micro-topographic features. The dynamic mapping mechanism supports a precise "one type of terrain, one parameter" application algorithm, effectively solving the adaptability problem caused by fixed parameters in traditional algorithms, and improving the overall accuracy and processing efficiency of ground point extraction. The clustering strategy incorporating spatial constraints enhances the stability and generalization ability of the classification, reduces sensitivity to random initialization, and ensures the consistency and reliability of terrain complexity recognition results. It balances filtering quality and computational efficiency, improving its engineering practicality in large-scale complex terrain point cloud data processing tasks.

[0012] 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

[0013] 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.

[0014] Figure 1 This is a schematic flowchart of the ground filtering method according to Embodiment 1 of the present invention; Figure 2 This is a flowchart of the ground filtering method according to Embodiment 1 of the present invention. Detailed Implementation

[0015] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0016] 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.

[0017] 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.

[0018] Example 1 In one or more of the technical solutions disclosed in the embodiments, such as Figures 1 to 2 As shown, a ground filtering method based on dynamic optimization of fabric simulation parameters includes the following steps: Step 1: Obtain point cloud data of the area to be processed, perform preprocessing, construct a complexity index, and calculate the complexity index value based on the preprocessed data; Step 2: Using a clustering method that combines the elbow rule and spatial neighborhood constraints, the obtained complexity indexes are clustered to obtain the terrain complexity categories; Step 3: Based on the terrain complexity category and the dynamic mapping of the constructed complexity category-optimal parameter group, select the cloth parameters of the cloth placement algorithm, and then use the cloth placement algorithm to perform ground point filtering and ground extraction.

[0019] In this embodiment, an adaptive clustering algorithm that integrates the elbow rule and spatial neighborhood constraints is first used to cluster these complexity features, thereby forming multiple terrain complexity categories. This clustering method automatically determines the optimal number of categories through the elbow rule and combines spatial neighborhood information to constrain the cluster boundaries, improving clustering stability and practical adaptability of classification. Then, the system calls a pre-established complexity category-optimal parameter group mapping model and automatically matches the parameter configuration of the cloth simulation algorithm according to the identified categories. Finally, the cloth simulation filtering algorithm extracts ground points from the point cloud data and filters out non-ground points, realizing a complexity-aware adaptive ground filtering process.

[0020] This embodiment significantly improves the recognition accuracy of complex terrain areas through multi-dimensional terrain feature quantization and clustering, making it particularly suitable for mountainous or forested scenes with significant elevation undulations and local micro-topographic features. The dynamic mapping mechanism supports a precise "one type of terrain, one parameter" application algorithm, effectively solving the adaptability problem caused by fixed parameters in traditional algorithms, and improving the overall accuracy and processing efficiency of ground point extraction. The clustering strategy incorporating spatial constraints enhances the stability and generalization ability of the classification, reduces sensitivity to random initialization, and ensures the consistency and reliability of terrain complexity recognition results. It balances filtering quality and computational efficiency, improving its engineering practicality in large-scale complex terrain point cloud data processing tasks.

[0021] Before extracting ground points, the original 3D laser point cloud data needs to be preprocessed to improve data quality, remove abnormal noise, and reduce data size, laying the foundation for subsequent complexity analysis and cloth simulation. Step 1 involves preprocessing the point cloud data of the area to be processed, including format conversion, statistical filtering for noise reduction, and voxel filtering for downsampling. The specific preprocessing process is as follows: Step 11: Convert and standardize the acquired point cloud data. The raw point cloud data comes from various sources (such as LAS, PCD, TXT, etc.). To ensure processing consistency and compatibility, the raw data is uniformly converted to LAS 1.4 format, retaining key fields. This step ensures that all subsequent algorithm modules can be compatible with and uniformly process point cloud data. Key fields may include 3D coordinates, intensity, and other data. Step 12: Perform statistical filtering and noise reduction on the acquired point cloud data; During laser scanning, especially in complex mountainous environments, point cloud data often contains isolated and outlier points. These outliers can be caused not only by reflections from tree canopy gaps, water reflections, and sensor misreadings, but are also affected by multiple factors such as complex terrain undulations, dense vegetation obstruction, atmospheric interference, and the stability of the scanning platform. Since these noisy points typically lack spatial continuity, directly involving them in subsequent analysis will severely interfere with the accuracy and robustness of ground point extraction. In this embodiment, various filtering methods, including statistical filtering, are employed to effectively remove outliers, improving point cloud quality and processing efficiency.

[0022] Specifically, the specific processing flow for statistical filtering denoising is as follows: Step 121: For each point Pi in the point cloud, construct a set of its k nearest neighbors; Step 122: Calculate the mean Euclidean distance di from point Pi to its neighboring points; Step 123: Calculate the mean μ and standard deviation σ of di in the entire point cloud, and set the threshold T = μ + a•σ (usually a = 1-2). Step 124: If the mean Euclidean distance di from point Pi to its neighboring points is greater than T, then point Pi is considered an outlier and is removed.

[0023] Step 13, Voxel filtering downsampling: Perform voxel mesh downsampling on the denoised point cloud; To reduce data volume and improve computational efficiency, the process of downsampling the point cloud using voxel meshes after denoising is as follows: Step 131: Define the side length v of the cube voxel based on the terrain density of the collection area; Terrain density refers to the number of point clouds contained in a unit area or unit volume (point density), which is used to measure the spatial density of point cloud data in a certain area. Specifically, the side length v of the cube voxel can be defined, for example, set to v=0.5m, which can be adjusted to 0.3m-1.0m according to the specific terrain density; Step 132: Based on the side length v of the cube voxel, divide the entire point cloud space into several voxel grids; for each voxel, retain the centroid of the three-dimensional coordinates of all points inside it as the representative point, discard other points, and obtain the downsampled data, thus achieving data compression.

[0024] The downsampling process in this embodiment can preserve the main geometric features of the terrain while significantly reducing the number of points, providing data support and computational foundation for subsequent complexity index calculation, K-means clustering, and cloth simulation parameter matching.

[0025] To achieve refined analysis of terrain complexity and targeted parameter configuration, the preprocessed point cloud is first divided into several independent sub-blocks, and a multi-dimensional complexity index is calculated for each sub-block. This constructs a quantitative system that comprehensively reflects terrain features, laying the foundation for subsequent clustering, classification, and parameter mapping. Step 1 involves constructing a complexity index, calculating the index value based on the preprocessed data, and includes the following steps: Step 101, Spatial partitioning strategy: Based on the point cloud density, set the spatial voxel side length and the overlapping boundary width of the partitioned region for the preprocessed point cloud data, and divide the point cloud data to obtain multiple sub-block point cloud data. Specifically, using 100m×100m as the basic unit, to enhance the adaptability of the algorithm, it supports adaptive adjustment based on point cloud density within the range of 80-120m. When the average point spacing within a sub-block is <0.5m, the sub-block is automatically expanded to 120m×120m to reduce computational redundancy; when the average point spacing within a sub-block is >1.0m, the sub-block is automatically shrunk to 80m×80m to ensure that each sub-block contains enough points for complexity calculation.

[0026] Specifically, each terrain sub-block introduces a 10m overlapping area of ​​a set width in the X / Y direction as the overlapping boundary width of the sub-block, enhancing the contextual expressiveness of the boundary transition area; In mountainous point clouds, terrain changes frequently, especially in transition zones such as gullies, ridges, and steep slopes. If there are no redundant areas between sub-blocks, the complexity calculation and the cloth simulation results will show significant inconsistencies between blocks. In this embodiment, the overlapping area setting is based on measured data, with the average width of the terrain transition zone estimated to be 8-12m. Finally, 10m is set as the redundant width, which satisfies the boundary consistency requirement and avoids the redundancy burden caused by data duplication.

[0027] Step 102: Calculate the complexity index value for each sub-block of data, and standardize the complexity index value. Furthermore, the complexity metrics constructed include: 1) Elevation variation coefficient (CV) h The elevation variation coefficient is constructed by dividing the standard deviation by the mean. It reflects the vertical undulation of the terrain, is dimensionless, and adapts well to variations at different elevation scales. The formula for calculating the elevation variation coefficient is: ; ; ; in, Let be the elevation value of the i-th point, and n be the number of points in the sub-block. This represents the standard deviation of the point cloud elevation values ​​within a sub-block. The mean elevation value of point cloud within a sub-block.

[0028] 2) Roughness: The average of the squared Euclidean distances from all points in each sub-block to the center point (mean point) of that sub-block is used as the roughness. Roughness is used to measure the dispersion of surface points in a sub-block, and can characterize the micro-topographic undulations. The formula is: ; in, The center point of the sub-block (mean of the coordinates of all points); (x i y i , z i Let be the three-dimensional coordinates of the i-th point.

[0029] The larger the value R, the more irregular and rough the surface.

[0030] 3) Mean curvature of terrain: The mean Gaussian curvature of all points in each sub-block is used as the mean curvature of terrain; Topographic curvature is used to describe the overall curvature of the Earth's surface. The greater the curvature, the more pronounced the surface undulations. Gaussian curvature K is used to calculate the curvature at each point, and then the mean value within each sub-block is taken.

[0031] First, perform quadratic surface fitting on each point within its neighborhood ( ), and the fitted surface is given by the formula; ; Where a, b, c, d, e, f are fitting coefficients; Specifically, the neighborhood window size can be set to 5×5; Based on the fitting coefficients of the fitted surface, the Gaussian curvature is calculated using the following formula: ; Then, calculate the mean of the curvature values ​​of all points within the sub-block: ; in, This represents the average curvature of the terrain within a sub-block, reflecting the overall degree of surface curvature of the sub-block. 'n' represents the total number of points within the sub-block. This represents the Gaussian curvature value of the i-th point within the sub-block.

[0032] 4) Slope Standard Deviation: The single-point slope of each point is defined as the normal vector. With vertical vector =The included angle of (0,0,1); Within the sub-block, the standard deviation of the single-point slope of all points is the slope standard deviation; The standard deviation of slope reflects the degree of drastic change in slope within a sub-block. First, the slope at each point is calculated, then the range of fluctuation across the entire sub-block is statistically analyzed; the slope at a single point is defined as the normal vector. With vertical vector The angle between points (0,0,1), expressed in radians or degrees, is calculated using the following formula: ; The standard deviation of the slope at all points within this sub-block is calculated using the following formula: ; in, The average slope at all points. This represents the slope of the i-th point within the sub-block. It represents the standard deviation of the slope within a sub-block, reflecting the degree of drastic change in slope within the sub-block.

[0033] Optionally, the standardization process can employ the Z-score standardization method to standardize the aforementioned four-dimensional complexity metrics, as shown in the formula: ; Where: I is a certain primitive complexity index; These are the mean and standard deviation of the complexity index I across all sub-blocks, respectively; I norm The standardized values ​​satisfy zero mean and unit variance; In this implementation, Z-score standardization eliminates dimensional differences (e.g., roughness is in the square dimension of length, and curvature is in the inverse dimension of length) and numerical distribution differences among the four indicators: elevation variation coefficient, roughness, mean curvature, and slope standard deviation. This ensures a balanced weight across the four dimensions of the complexity feature vector. The standardized feature vector accurately reflects the overall complexity differences among different sub-blocks, providing reliable input for the subsequent K-means clustering algorithm to achieve a reasonable division of high, medium, and low complexity based on feature similarity. Furthermore, through the dynamic mapping mechanism of complexity category-optimal parameter group, a data foundation is laid for the accurate adaptation of fabric simulation parameters.

[0034] The complexity index described in this embodiment classifies terrain complexity from both macroscopic and microscopic levels. At the macroscopic level, the overall undulation is characterized by the elevation variation coefficient, and the slope standard deviation represents the magnitude of slope fluctuation, solving the problem that a single slope cannot distinguish between gentle and steep mixed areas. At the microscopic level, surface roughness is used to identify surface unevenness, and the mean curvature of the terrain is used to determine the degree of surface curvature, solving the classification distortion problem of different surface textures or micro-topography for the same slope, such as the difficulty in identifying rock piles and bare land using existing complexity indices.

[0035] To enable the algorithm system to automatically distinguish different types of terrain regions in complex mountainous areas and to provide a basis for the adaptive application of simulation parameters, this embodiment designs a terrain complexity clustering method that integrates spatial information. Simply put, it allows the computer to automatically group similar terrain and geographically close areas into one category based on the characteristics and spatial location of each terrain sub-block, thereby achieving automatic zoning that more closely reflects the real-world landform patterns. In step 2, a clustering method that combines the elbow rule and spatial neighborhood constraints is used to cluster the obtained complexity indices to obtain terrain complexity categories, including the following steps: Step 21: Construct a one-dimensional feature vector from the calculated complexity index value; Each sub-block has several numerical features representing terrain complexity, including elevation variation coefficient, roughness, mean terrain curvature, and standard deviation of slope.

[0036] These features, after standardization, form a four-dimensional feature vector: ; The computer uses the feature vector of each terrain sub-block as input data for the clustering algorithm.

[0037] Step 22: Automatically determine the optimal number of clusters (k value) using the elbow rule, and obtain the initial classification results using K-means clustering; Traditional clustering typically requires manual specification of the number of clusters, but manual setting is often unreasonable. This embodiment uses the elbow rule to automatically determine the optimal number of clusters. Furthermore, the method for automatically determining the optimal number of clusters (k value) using the elbow rule includes the following steps: Step 221: Traverse the candidate cluster numbers (k values) and perform clustering; In this embodiment, the number of clusters can be set from 1 to 15. Step 222: For the clustering results under each k value, calculate the sum of squared Euclidean distances from each data point to the center of its class, i.e., the sum of squared errors within the group, with the formula as follows; ; Step 223: Plot the curve of the sum of squared errors within groups (WCSS) as a function of k, identify the elbow inflection point, and the k value corresponding to the slowdown of the decrease in SSE is the optimal number of clusters. In this way, the system can automatically determine how many categories the terrain should be divided into without human intervention.

[0038] Step 23: Integrate spatial constraints for cluster optimization. Based on the spatial neighborhood constraint mechanism, optimize the initial classification results to obtain the adjusted clustering results, including the following steps; Step 231: Obtain the center coordinates (x_center, y_center) of each terrain sub-block; Step 232: Establish a spatial index structure, which can be constructed as a KD-Tree; Step 233: For each sub-block, search for neighboring sub-blocks within its neighborhood. Specifically, the neighborhood range can be set to 4% of the coordinate range; Step 234: Execute the neighborhood majority correction rule: If the proportion of a certain category in the neighborhood exceeds the set proportion, and the current sub-block belongs to a different category, then automatically change its category to the dominant category of the neighborhood; execute the neighborhood majority correction rule for all sub-blocks to suppress edge jumps and isolation misjudgments.

[0039] Optionally, if the proportion of a certain category in the neighborhood exceeds a set percentage, where the set percentage can be 60%; Ordinary K-means only considers numerical features and ignores the spatial distribution of terrain, which can easily lead to the phenomenon of "the same area being classified into different categories". To address this, this embodiment introduces a spatial neighborhood constraint mechanism to ensure that the clustering results conform to geographic spatial continuity. This spatial constraint clustering mechanism is the first to introduce geographic neighborhood relationships into the complexity grading process, ensuring that the terrain classification results maintain spatial consistency with the geomorphological structure and providing a more stable partitioning basis for subsequent dynamic parameter mapping.

[0040] Existing clustering methods suffer from two major drawbacks. First, traditional K-means methods use random initial centroids and empirically determined cluster number k, which can easily lead to local optima in multimodal mountainous terrain. Second, they ignore the spatial continuity of the terrain, resulting in the misclassification of spatially adjacent terrain elements of the same type. These problems together cause a disconnect between the complexity classification and the actual terrain distribution, directly affecting the effectiveness of parameter mapping.

[0041] To address the problem of distorted complexity classification caused by coarse clustering methods in existing technologies, this embodiment first adaptively determines the optimal number of clusters K by incorporating the elbow rule, avoiding empirical bias. Then, after obtaining the initial classification through K-means clustering, the results are optimized by combining spatial neighborhood constraints (based on the neighborhood majority rule of the block center point) to ensure spatial continuity of the classification. Ultimately, this achieves a high degree of matching between complexity classification and actual terrain, providing a reliable classification basis for dynamic parameter mapping.

[0042] In step 3, the dynamic mapping of complexity category-optimal parameter set is constructed as a supervised learning regression model. Its input is the clustering category of the complexity of each terrain sub-block, and its output is the corresponding matched optimal cloth simulation parameter set. The fabric simulation parameters include: stiffness coefficient k1, friction coefficient μ, damping coefficient d, and fabric density ρ; optionally, the supervised learning regression model can use gradient boosting tree (LightGBM). Furthermore, the training process of the supervised learning regression model includes the following steps: Step 31: Obtain the training samples. After performing the complexity calculation, sub-block partitioning and clustering in Step 1 and Step 2, obtain representative sample sub-blocks under multiple complexity categories. Step 32: For each parameter combination, perform cloth simulation on a representative sample sub-block to obtain accuracy and stability indices. Step 33: For the cloth simulation results of samples in each complexity category, calculate the comprehensive score based on the accuracy index and stability index, and select the best parameter set for the cloth simulation of each complexity category. The formula for calculating the overall score is: ; Where Var is the score variance of the same group of fabric simulation parameters on different samples; λ represents the coefficient, which is automatically adjusted according to the complexity level to strengthen the stability constraints of complex regions. Indicates ground point extraction Fraction; Indicates recall rate; Indicates the error rate; If multiple sets of results have the same overall score, the physical constraint verification module is called to compare the stiffness coefficient with the complexity category (e.g., complex terrain tends to have high stiffness parameters), and finally the unique optimal combination is determined. Step 34: Using the complexity category as input and the optimal parameter set for cloth simulation as output, input these parameters into a supervised learning regression model for training. The trained supervised learning regression model is then used to map the complexity category to the parameters of the cloth simulation. The training objective is to minimize the following loss function: ; in, This represents the set of parameters predicted by the model. Indicates the target parameter set. This represents the adaptively adjusted weighting coefficient. Penalty for unreasonable parameter combinations; Step 31, the method for obtaining representative sample sub-blocks under multiple complexity categories, includes the following steps: Step 311: Obtain all sub-blocks under each complexity category from the clustering results; Step 312: Divide the sub-blocks under each complexity category into regions according to the distance from each sub-block to the centroid of the cluster; Optionally, the distance from each sub-block to the centroid of the cluster is calculated. Specifically, the Euclidean distance between the feature vector of each sub-block and the center of the cluster is calculated; the distance distribution from all sub-blocks in the category to the center is obtained.

[0043] Divide the area by distance, and divide all sub-blocks into several intervals according to the distance, including the central area, the middle area and the boundary area; Optionally, the top 30% with the smallest distance can be assigned to the central region, the 40% with the middle distance to the middle region, and the remaining proportion to the boundary region. Samples can be drawn proportionally from each region to avoid bias.

[0044] Step 312: Samples are drawn from each interval according to the proportion, and during the sampling process, sub-blocks that are spatially dispersed are selected. The goal is to avoid concentrating all representative samples in the same terrain area and to increase spatial diversity.

[0045] Furthermore, it also includes a sample supplementation mechanism. If the number of terrain sub-blocks selected according to the above rules is less than the set minimum sample number, such as less than 15, then: check if there are any uncovered distance intervals; or find spatially missing sub-block areas; supplement samples from these areas until the minimum sample number requirement is met.

[0046] In step 32, cloth simulation is performed on a representative sample sub-block, including the following steps: Step 321: Determine the key parameter dimensions for fabric simulation and simplify the fabric parameters; Step 322: Based on the reduced fabric parameters, perform adaptive parameter combination and execute the fabric simulation process; Given the high dimensionality of fabric simulation algorithm parameters (such as fabric stiffness coefficient k, friction coefficient μ, damping coefficient d, fabric density ρ, etc.), directly conducting full factorial combination tests would incur exponential experimental costs. Therefore, it is necessary to reduce the fabric parameters and construct lightweight parameter combinations. Step 321, determining the key parameter dimensions for fabric simulation and the method for reducing fabric parameters, includes the following steps: Step 3211: Obtain the parameter ranges and constraints for fabric simulation, and automatically generate multiple sets of parameter samples using low-difference sequences (such as Sobol sequences); Step 3212: For the selected number of sample sub-blocks, allocate the sample sub-blocks to the parallel computing thread pool. While keeping other parameters fixed, change the target parameter values ​​one by one, and automatically call the cloth simulation algorithm to perform ground point extraction calculations, and calculate the error rate, recall rate, F1 score and other index values. Step 3213: Calculate the standard deviation and p-value of the performance index corresponding to each fabric parameter; Step 3214: Construct a sensitivity calculation function based on the principle that the greater the fluctuation and the smaller the p-value, the higher the sensitivity. Calculate the sensitivity score for each fabric parameter based on the obtained standard deviation and p-value. S i = f(standard deviation, p-value); Step 3215, Threshold Determination and Dimensional Adaptive Reduction: Based on the maximum sensitivity ranking, select a set number of fabric parameters as core fabric parameter indicators; In this embodiment, by using fewer sample experiments, the parameters that have a significant impact on the simulation results when the fabric parameters are changed are determined. This achieves dimensionality reduction of the fabric parameters, greatly reducing the number of parameter combinations in the later fabric simulation process, reducing the number of times the fabric simulation algorithm is run, and improving the system's operating efficiency.

[0047] Step 322, the method for adaptive parameter combination based on the reduced fabric parameters, includes the following steps: Step 3221: After obtaining the parameter dimensions, determine the number of combinations based on the number of parameter dimensions; Step 3222: After determining the number of combinations, use the adaptive Latin hypercube sampling algorithm to generate fabric parameter combinations; Specifically, the initial sampling coverage is uniform, but the sampling density is dynamically adjusted during the sample generation process: if the previous test found that the performance index of a certain fabric parameter area fluctuates greatly (high variance), the sampling density of the corresponding area will be increased; for fabric parameter areas with stable performance, the sampling frequency will be reduced; thereby focusing on highly sensitive areas and reducing the computational burden of low-yield areas.

[0048] In step 32, the cloth simulation creates a task queue for each set of parameters, and the cloth simulation algorithm is executed in parallel. During the operation, the system defines a real-time benefit function U(x,t) for each task, taking into account the performance improvement rate ΔF1 / Δt, the calculation time T(x), and the resource utilization rate R(x) of the current task.

[0049] The task scheduling module aims to maximize the global reward ΣU and dynamically adjusts task priorities using a multi-armed slot machine (UCB) strategy. When a resource bottleneck or performance convergence is detected, the system automatically rearranges the task queue, suspending low-reward tasks and scheduling potentially high-reward tasks, thus achieving adaptive allocation of computing resources and optimal reward scheduling.

[0050] If the rate of change of task revenue is lower than the set threshold (e.g., ΔU<2%) during continuous iterations, the system will automatically trigger early stop control to terminate the task in advance and release resources.

[0051] This embodiment achieves a shift from "fixed empirical values" to "complexity-driven adaptive configuration" in cloth simulation parameters through the aforementioned method, significantly improving the accuracy, stability, and automation level of ground point extraction, and meeting the application requirements for high-quality 3D point cloud processing in complex terrain scenarios. The entire parameter optimization and mapping process forms a self-closing system of "sampling-testing-learning-calling," with the model continuously optimized during operation, constituting an intelligent ground point extraction framework with adaptive feedback characteristics.

[0052] Example 2 Based on Example 1, this example provides a ground filtering system based on dynamic optimization of fabric simulation parameters, including: The complexity calculation module is configured to acquire point cloud data of the area to be processed, preprocess it, construct a complexity index, and calculate the complexity index value based on the preprocessed data. The clustering module is configured to use a clustering method that combines the elbow rule and spatial neighborhood constraints to cluster the obtained complexity indices and obtain the terrain complexity categories. The mapping extraction module is configured to select the cloth parameters of the cloth algorithm based on the dynamic mapping of the constructed complexity category-optimal parameter group based on the terrain complexity category, and then use the cloth algorithm to perform ground point filtering and ground extraction.

[0053] 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.

[0054] Example 3 Based on Embodiment 1, this embodiment provides an electronic device, including a memory and a processor, as well as computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, they complete the steps in the above-described ground filtering method based on dynamic optimization of fabric simulation parameters.

[0055] Example 4 Based on Embodiment 1, this embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, they complete the steps in the above-described ground filtering method based on dynamic optimization of fabric simulation parameters.

[0056] 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.

[0057] 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 filtering method based on dynamic optimization of cloth simulation parameters, characterized in that, The method comprises the following steps: Obtaining point cloud data of a region to be processed for preprocessing, constructing a complexity index, and calculating a complexity index value based on the preprocessed data; Using a clustering method that integrates elbow rule and spatial neighborhood constraint to cluster the obtained complexity index to obtain a terrain complexity category; Based on the terrain complexity category, selecting cloth simulation parameters of a cloth algorithm based on a constructed dynamic mapping of complexity category-optimal parameter group, and then using the cloth algorithm to perform ground point filtering and ground extraction.

2. The ground filtering method based on dynamic optimization of cloth simulation parameters according to claim 1, wherein: Constructing a complexity index and calculating a complexity index value based on preprocessed data comprises the following steps: Setting a spatial voxel length based on point cloud density for the preprocessed point cloud data, setting a division region overlap boundary width, dividing the point cloud data to obtain a plurality of sub-block point cloud data; Calculating a complexity index value for each sub-block data respectively, and performing standardization processing on the complexity index value.

3. The method of claim 1, wherein: The constructed complexity index comprises: Elevation variation coefficient: constructing a ratio of standard deviation to mean value as the elevation variation coefficient, Roughness: in each sub-block, the average value of the square of the Euclidean distance of all points to the center point of the sub-block as the roughness; Terrain curvature mean value: in each sub-block, the mean value of the Gaussian curvature values of all points as the terrain curvature mean value; Slope standard deviation: defined as the standard deviation of the single-point slopes of all points in the sub-block the angle with the vertical vector; in the sub-block, the standard deviation of the single-point slopes of all points is the slope standard deviation.

4. The method of claim 1, wherein: Using a clustering method that integrates elbow rule and spatial neighborhood constraint to cluster the obtained complexity index to obtain a terrain complexity category comprises the following steps: Constructing the calculated complexity index value into a one-dimensional feature vector; Using elbow rule to automatically determine the optimal clustering number, and using K-means clustering on the obtained one-dimensional feature vector to obtain an initial classification result; Fusing spatial constraint for clustering optimization, and based on a spatial neighborhood constraint mechanism, optimizing the initial classification result to obtain an adjusted clustering result.

5. The method of claim 4, wherein: The method of using elbow rule to automatically determine the optimal clustering number comprises the following steps: Iterating through candidate clustering numbers to perform clustering; For the clustering result under each k value, calculating the sum of the square of the Euclidean distance of each data point to the center of the class it belongs to, i.e., the within-group sum of squares of errors; Drawing a curve of the within-group sum of squares of errors WCSS changing with k, identifying the elbow inflection point, and when the SSE decreases slowly, the k value corresponding to the elbow inflection point is the optimal clustering number.

6. The method of claim 4, wherein: Fusing spatial constraint for clustering optimization, and based on a spatial neighborhood constraint mechanism, optimizing the initial classification result to obtain an adjusted clustering result comprises the following steps: Obtaining the center coordinates of each terrain sub-block, and establishing a spatial index structure; For each sub-block, searching for adjacent sub-blocks within its neighborhood range based on the constructed spatial index structure; Performing neighborhood majority correction rule: if a certain class in the neighborhood occupies more than a certain proportion, and the current sub-block belongs to a different class, then automatically modifying the class of the current sub-block to the dominant class in the neighborhood.

7. The method of claim 1, wherein: The dynamic mapping of complexity category-optimal parameter group is constructed as a supervised learning regression model, the input of which is the clustering category of the complexity of each terrain sub-block, and the output is the corresponding optimal cloth simulation parameter group; The training process of the supervised learning regression model comprises the following steps: After obtaining the to-be-trained sample, performing complexity calculation, sub-block division, and clustering, a plurality of representative sample sub-blocks under a plurality of complexity categories are obtained; For each parameter combination, cloth simulation is performed on the representative sample sub-blocks to obtain the precision index and the stability index; For the cloth simulation results of the samples in each complexity category, a comprehensive score is calculated based on the precision index and the stability index, and the best parameter combination for cloth simulation of each complexity category is selected; The complexity category is input, and the best parameter combination for cloth simulation is output, and is input into a supervised learning regression model for training to obtain a trained supervised learning regression model, which is used to map the complexity category to output the parameters of cloth simulation.

8. A ground filtering system based on dynamic optimization of cloth simulation parameters, characterized in that, The method comprises the following steps: The complexity calculation module is configured to obtain point cloud data of a to-be-processed region for preprocessing, construct a complexity index, and calculate a complexity index value based on the preprocessed data; The clustering module is configured to adopt a clustering method that fuses elbow rule and spatial neighborhood constraint to cluster the obtained complexity index to obtain a terrain complexity category; The mapping extraction module is configured to select cloth parameters of a cloth algorithm based on the terrain complexity category and the constructed dynamic mapping of the complexity category- optimal parameter combination, and then perform ground point filtering and ground extraction using the cloth algorithm.

9. An electronic device, comprising: The computer program product comprises a memory and a processor, and computer instructions stored in the memory and run on the processor, and when the computer instructions are run by the processor, the steps in the ground filtering method based on dynamic optimization of cloth simulation parameters in any one of claims 1-7 are completed.

10. A computer-readable storage medium, characterized in that, The computer program product is used to store computer instructions, and when the computer instructions are executed by the processor, the steps in the ground filtering method based on dynamic optimization of cloth simulation parameters in any one of claims 1-7 are completed.

Images