Mass point cloud adaptive simplification method and device for mine roadway

Abstract

The invention discloses a mass point cloud adaptive simplification method and device for a mine roadway, and the method comprises the steps: carrying out the three-dimensional laser scanning of the mine roadway, obtaining original point cloud data, carrying out the preliminary cleaning, and constructing a point cloud spatial index; calculating a point cloud normal vector based on PCA (Principal Component Analysis), estimating local density by adopting kernel density, dynamically adjusting a neighborhood search radius, and outputting point cloud data with normal vector and density information; constructing a dimension feature function, calculating a local neighborhood entropy to determine an optimal neighborhood, extracting normal vector included angle features, and classifying point clouds into different feature regions including flat, linear and complex feature regions by using an SVM (Support Vector Machine); according to the classification result and the local entropy value, differential simplification rates are set for different feature regions, simplification is executed iteratively until the target simplification rate is reached, and a simplified point cloud file is output; according to the invention, through adaptive neighborhood search based on normal vectors, local entropy optimization and intelligent classification, the problems of low processing efficiency and feature loss of mine roadway point cloud data are solved.

Description

Technical Field

[0001] This invention relates to the field of three-dimensional point cloud data processing technology, and more specifically to a method and apparatus for adaptive simplification of massive point clouds for mine roadways. Background Technology

[0002] Three-dimensional laser scanning technology has become an important means of obtaining high-precision three-dimensional spatial data in mine roadways, tunnels and slope engineering. Through non-contact measurement, this technology can quickly collect massive point cloud data on the roadway surface, thus providing a foundation for rock mass structure analysis, deformation monitoring and digital modeling.

[0003] However, in practical applications, especially in underground mine tunnel environments, point cloud data is usually characterized by its large volume, uneven density distribution, and high redundancy, which poses a huge challenge to the storage, transmission, and subsequent processing of point cloud data. Efficient point cloud simplification algorithms aim to reduce the amount of data while retaining key geometric features, such as rock mass structure surfaces and tunnel outlines, to ensure the accuracy and efficiency of engineering analysis.

[0004] Currently, several traditional methods are widely used in the field of point cloud simplification, such as: fixed radius search, which simplifies by defining a neighborhood of a fixed radius for each point, but cannot adapt to changes in point cloud density. It may ignore details in high-density areas and oversimplify in low-density areas, leading to feature loss; K-nearest neighbor search, which simplifies based on the K nearest neighbors of a point, can partially adapt to density changes, but the fixed number of neighbors makes it difficult to handle complex structures and is prone to introducing noise in boundary areas; curvature-based simplification, which retains features in high-curvature areas by calculating the curvature of the point cloud, but for massive point clouds, the computational complexity is high and it is sensitive to noise, easily deleting important points in flat areas; and random simplification, which randomly deletes point clouds, is simple and fast, but cannot guarantee feature retention and may lead to distortion of geometric information.

[0005] In summary, existing methods have significant limitations when processing point clouds in mine roadways: low efficiency, with massive point cloud processing taking a long time, especially when the number of point clouds reaches hundreds of millions, traditional algorithms struggle to meet real-time or near-real-time engineering requirements; insufficient adaptability, as fixed parameters (such as radius or K value) cannot be dynamically adjusted to adapt to changes in local point cloud density and geometric features, resulting in poor simplification effects in complex roadway structures; inaccurate feature preservation, with key features (such as structural surface edges and angular areas) easily lost during simplification, affecting the reliability of subsequent rock mass identification and mechanical analysis; and low automation, with many methods relying on manual parameter tuning and lacking intelligent adaptive mechanisms, increasing the complexity of engineering applications.

[0006] To address the aforementioned issues, academia and industry have attempted several improvements, such as combining machine learning algorithms for point cloud classification or introducing multi-scale processing strategies. However, these solutions still have shortcomings: for example, machine learning methods require a large amount of labeled data and have high computational resource demands; multi-scale methods may not be effective in handling the non-uniform distribution of point clouds.

[0007] Therefore, how to provide an efficient, adaptive, and feature-preserving method for simplifying massive point clouds to meet the actual needs of mine tunnel engineering is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0008] In view of this, the present invention provides an adaptive simplification method and apparatus for massive point clouds in mine roadways to solve some of the technical problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: An adaptive simplification method for massive point clouds in mine roadways includes the following steps: S1. Use a 3D laser scanner to scan the mine roadway, obtain raw point cloud data, perform preliminary cleaning, and construct a point cloud spatial index; S2. Calculate the point cloud normal vector based on the PCA method, estimate the local density using the kernel density estimation method, dynamically adjust the neighborhood search radius according to the local density value, perform adaptive neighborhood search, and output point cloud data with normal vector and density information; S3. Construct a dimensional feature function to quantify the local geometric features of the point cloud, calculate the local neighborhood entropy function to determine the optimal neighborhood size, extract the angle feature of the normal vector, and use the SVM algorithm to classify the point cloud, outputting a point cloud dataset with classification labels, dividing the point cloud into different feature regions, including flat feature regions, linear feature regions, and complex feature regions. S4. Based on the classification results and local entropy values, set differentiated simplification rates for different feature regions, iteratively execute simplification operations until the target simplification rate is reached, and output the simplified point cloud file.

[0010] Preferably, in step S1, the original point cloud data is initially cleaned, including removing noise points and outliers. By calculating the spatial distribution density of the point cloud, a radius-based filtering method is used to remove isolated points. The spatial index is constructed using a KD-tree or Octree data structure.

[0011] Preferably, the specific content of step S2 is as follows: S21. Establish an initial neighborhood with each point in the point cloud data as the center, and use the K-nearest neighbor method to initially determine the neighborhood point set; S22. The local density at each center point is calculated using a radius-based kernel density estimation method; S23. Dynamically adjust the neighborhood search radius based on the local density value; S24. The normal vector of each point is calculated using the principal component analysis (PCA) method. Specifically, the covariance matrix of the neighborhood point set is calculated and the covariance matrix is ​​decomposed into eigenvalues. The eigenvector corresponding to the smallest eigenvalue is used as the normal vector. The direction of the normal vector is adjusted by checking the dot product with the centroid of the point cloud as a reference.

[0012] Preferably, the local density at each center point is:

[0013] in, Let i be the three-dimensional coordinates of the current i-th point. For point The j The three-dimensional coordinates of the nearest neighbor points k This represents the number of neighborhood points obtained using the K-nearest neighbor search method. The Euclidean distance between the current point and its neighboring points. For use as a point Adaptive kernel width; The covariance matrix of the neighborhood points is:

[0014] in, As the center of gravity of the neighborhood, For Search for its initial neighborhood around the center and calculate the covariance matrix of the neighborhood points.

[0015] Preferably, the specific content of step S3 is as follows: S31. For each point, based on the set of points in the adaptive neighborhood, calculate the eigenvalues ​​of the covariance matrix and construct a dimensional feature function to quantify the local geometric features of the point cloud; S32. Calculate the local neighborhood entropy function value, and iteratively search to determine the optimal neighborhood size with the goal of minimizing the local neighborhood entropy function, and extract the feature description parameters of the point cloud; S33. Using the angle between normal vectors as features, calculate the angle between the normal vectors of a point and its neighbors, use L1 regularization to select the optimal feature subset, and use the SVM algorithm with radial basis function as kernel function for classification to divide the point cloud into different feature regions, including flat feature regions, linear feature regions and complex feature regions.

[0016] Preferably, the dimensional feature function is:

[0017] in, , , λ1, λ2, and λ3 represent the degree of linear, planar, and scattering characteristics, respectively, and their sum is 1. λ1, λ2, and λ3 are the eigenvalues ​​of the local neighborhood covariance matrix of each point, and λ1 ≥ λ2 ≥ λ3. Local neighborhood entropy function for:

[0018] The smaller the entropy value, the more ordered the local structure; the larger the entropy value, the more complex the structure.

[0019] Preferably, the specific content of step S4 is as follows: Based on the point cloud classification results, a dynamic simplification strategy is adopted for different feature regions. A high simplification rate is used for flat feature regions, while a low simplification rate or all points are retained for linear and complex feature regions. Set a simplified threshold and determine the feature region to which the point cloud belongs based on the local entropy value. For each point, calculate the local entropy value. If the entropy value is lower than the threshold and the point is located in a flat feature region, remove the point using random sampling or grid sampling methods. After simplification, the entropy value and classification of the remaining points are recalculated, and the process is repeated until the target simplification rate is reached or the entropy value stabilizes.

[0020] A massive point cloud adaptive simplification system for mine roadways, based on the aforementioned massive point cloud adaptive simplification method for mine roadways, includes: a data acquisition and preprocessing module, an adaptive neighborhood processing module, a point cloud intelligent classification module, and a dynamic simplification execution module. The data acquisition and preprocessing module is used to scan the mine roadway with a 3D laser scanner to acquire raw point cloud data, perform preliminary cleaning, and build a point cloud spatial index. The adaptive neighborhood processing module is used to calculate the point cloud normal vector and estimate the local density. It dynamically adjusts the neighborhood search radius based on the density value and outputs point cloud data with normal vector and density information. The point cloud intelligent classification module is used to calculate the local neighborhood entropy function to determine the optimal neighborhood size, extract the angle feature of the normal vector, and use the SVM algorithm to classify the point cloud, outputting a point cloud dataset with classification labels. The dynamic simplification execution module is used to set differentiated simplification rates for different feature regions based on classification results and local entropy values, iteratively execute simplification operations until the target simplification rate is reached, and output the simplified point cloud file.

[0021] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the aforementioned method for adaptive simplification of massive point clouds for mine roadways.

[0022] A processing terminal includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements the aforementioned adaptive simplification method for massive point clouds in mine roadways.

[0023] As can be seen from the above technical solution, compared with the prior art, this invention discloses an adaptive simplification method and device for massive point clouds in mine roadways. Addressing the unique challenges of rock mass structure surface identification in geotechnical engineering and the stringent requirements for 3D modeling accuracy in mining engineering, it innovatively combines adaptive neighborhood search and intelligent classification technologies to achieve a significant reduction in data volume while perfectly preserving key engineering features. Specifically, this invention introduces adaptive neighborhood search, intelligent point cloud classification, and dynamic simplification strategies to achieve efficient processing of massive, unevenly distributed point cloud data. While significantly reducing data volume, it effectively preserves key geometric features, enhances robustness to complex roadway environments, and helps provide a reliable foundation for subsequent rock mass structure analysis, 3D modeling, and engineering stability assessment. This invention can be widely applied in fields such as mine roadways, tunnel engineering, and slope monitoring, providing an efficient and reliable solution for 3D laser scanning point cloud data processing. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0025] Figure 1 A schematic diagram of an adaptive simplification method for massive point clouds in mine roadways provided by the present invention; Figure 2 This is a schematic diagram of the adaptive neighborhood search provided by the present invention; Figure 3 This is a schematic diagram of point cloud classification provided by the present invention; Figure 4 Comparison diagram of dynamic simplification effects provided by the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] This invention discloses an adaptive simplification method for massive point clouds used in mine roadways, such as... Figure 1 This includes the following steps: S1. Use a 3D laser scanner to scan the mine roadway, obtain raw point cloud data, perform preliminary cleaning, and construct a point cloud spatial index; S2. Calculate the point cloud normal vector based on the PCA method, estimate the local density using the kernel density estimation method, dynamically adjust the neighborhood search radius according to the local density value, perform adaptive neighborhood search, and output point cloud data with normal vector and density information; S3. Construct a dimensional feature function to quantify the local geometric features of the point cloud, calculate the local neighborhood entropy function to determine the optimal neighborhood size, extract the angle feature of the normal vector, and use the SVM algorithm to classify the point cloud, outputting a point cloud dataset with classification labels, dividing the point cloud into different feature regions, including flat feature regions, linear feature regions, and complex feature regions. S4. Based on the classification results and local entropy values, set differentiated simplification rates for different feature regions, iteratively execute simplification operations until the target simplification rate is reached, and output the simplified point cloud file.

[0028] To further implement the above technical solution, step S1 involves preliminary cleaning of the original point cloud data, including removing noise points and outliers. By calculating the spatial distribution density of the point cloud, a radius-based filtering method is used to remove isolated points, with the filtering radius set to 0.05 m. The spatial index is constructed using a KD-tree or Octree data structure.

[0029] In this embodiment, the scanning parameters of the 3D laser scanner are set as follows: the scanning distance range is 0.5 m to 70 m, the point cloud density is adjusted according to the complexity of the tunnel, and approximately 976,000 points are collected per second. After scanning, the data is stored in Cartesian coordinates (x, y, z) and includes color information.

[0030] To further implement the above technical solution, the specific content of step S2 is as follows: S21. Establish an initial neighborhood with each point in the point cloud data as the center, and use the K-nearest neighbor method to initially determine the neighborhood point set; S22. The local density at each center point is calculated using a radius-based kernel density estimation method; S23. Dynamically adjust the neighborhood search radius based on the local density value; Specifically, in this embodiment, a smaller neighborhood radius is used in high-density areas to capture details; and a larger neighborhood radius is used in low-density areas to ensure data integrity. S24. The normal vector of each point is calculated using the principal component analysis (PCA) method. Specifically, the covariance matrix of the neighborhood point set is calculated and the covariance matrix is ​​decomposed into eigenvalues. The eigenvector corresponding to the smallest eigenvalue is used as the normal vector. The direction of the normal vector is adjusted by checking the dot product with the centroid of the point cloud as a reference.

[0031] To further implement the above technical solution, in this embodiment, as follows: Figure 2 For each point ,by Centered on the target, search its initial neighborhood; Based on local point cloud density estimation, the density of each point is calculated. (Using kernel density estimation, bandwidth h=0.1m), and adaptively setting the neighborhood radius based on the density value. ; The local density at each center point is:

[0032] in, Let i be the three-dimensional coordinates of the current i-th point. For point The j The three-dimensional coordinates of the nearest neighbor points k This represents the number of neighborhood points obtained using the K-nearest neighbor search method. The Euclidean distance between the current point and its neighboring points. For use as a point The adaptive kernel width is determined by the local density of the point cloud. In practical applications, In high-density areas, set ; When it is a low-density area, set The remaining areas are of medium density and are determined by linear interpolation. ; Calculate the covariance matrix of the neighborhood points for:

[0033] in, As the center of gravity of the neighborhood, For Search for its initial neighborhood around the center and calculate the covariance matrix of the neighborhood points; To ensure the consistency of the normal vector direction, the normal vector direction is adjusted by checking the dot product, with the centroid of the point cloud as a reference, so that the angle between the normal vectors of adjacent points is less than 90 degrees. Through adaptive neighborhood search, its dynamic adjustment mechanism can effectively adapt to changes in point cloud density and improve search efficiency.

[0034] To further implement the above technical solutions, such as Figure 3 The specific content of step S3 is as follows: S31. For each point, based on the point set in the adaptive neighborhood, calculate the eigenvalues ​​of the covariance matrix and construct a dimensional feature function to quantify the local geometric features of the point cloud, providing a basis for subsequent feature region segmentation. S32. Calculate the local neighborhood entropy function value, and iteratively search to determine the optimal neighborhood size with the goal of minimizing the local neighborhood entropy function, and extract the feature description parameters of the point cloud; S33. Using the angle between normal vectors as features, calculate the angle between the normal vectors of a point and its neighboring points, use L1 regularization to select the optimal feature subset, and use the SVM algorithm with radial basis function as kernel function for classification to divide the point cloud into different feature regions.

[0035] To further implement the above technical solution, the dimensional feature function is:

[0036] in, , , λ1, λ2, and λ3 represent the degree of linear, planar, and scattering characteristics, respectively, and their sum is 1. λ1, λ2, and λ3 are the eigenvalues ​​of the local neighborhood covariance matrix of each point, and λ1 ≥ λ2 ≥ λ3. In this embodiment, when When the value is close to 1, the region is determined to be a flat feature region; when... When the value is large, the region is determined to be a linear feature region; when... When the value is large, the region is determined to be a complex feature region; Dimensional feature functions provide a preliminary quantification standard for local geometric features, but relying solely on these threshold rules is easily affected by noise and changes in point cloud density, resulting in insufficient classification accuracy. Therefore, this embodiment additionally calculates the local neighborhood entropy function to dynamically optimize the neighborhood size and introduces the normal vector angle feature to capture spatial orientation consistency, thereby providing multi-dimensional input for the SVM classifier. Local neighborhood entropy function for:

[0037] The smaller the entropy value, the more ordered the local structure; the larger the entropy value, the more complex the structure.

[0038] In this embodiment, point The angle between the vector and the normal vector of the neighboring point is:

[0039] Using L1 regularization to select the optimal feature subset, the regularization parameter... ; The SVM kernel function uses the radial basis function (RBF), and the parameters are optimized through cross-validation. The classification results divide the point cloud into flat feature regions, linear feature regions, and complex feature regions.

[0040] To further implement the above technical solution, the specific content of step S4 is as follows: Based on the point cloud classification results, a dynamic simplification strategy is adopted for different feature regions. A high simplification rate is used for flat feature regions, while a low simplification rate or all points are retained for linear and complex feature regions. Set a simplified threshold and determine the feature region to which the point cloud belongs based on the local entropy value. For each point, calculate the local entropy value. If the entropy value is lower than the threshold and the point is located in a flat feature region, remove the point using random sampling or grid sampling methods. After simplification, the entropy value and classification of the remaining points are recalculated, and this process is repeated until the target simplification rate is reached or the entropy value stabilizes. The effect of point cloud simplification before and after is as follows: Figure 4 .

[0041] In this embodiment, the entropy value Furthermore, given that the region conforms to planar characteristics and is flat, 80% of the points are deleted; entropy value For regions with linear or complex features, delete 20% of the points or keep all points.

[0042] In another embodiment, taking the point cloud data of a mine's -960m mid-section roadway as an example, the original point cloud contains approximately 30,187,147 points, and the method of the present invention is implemented as follows: Using a FARO Focus scanner, the scanning time was 38.8 minutes. After acquiring the point cloud, preprocessing was performed to remove noise. Adaptive neighborhood search reduced the computation time to an average of 0.005974 ms / point, while the fixed radius search computation time averaged 0.009961 ms / point. The point cloud classification accuracy reached 90.23%, compared to 88.25% for traditional SVM. The simplified point cloud retained 7,546,854 points, a simplification rate of 74.9%, but key structural features were fully preserved. Performance indicators showed a 40% improvement in processing efficiency and a feature retention rate exceeding 95%, meeting the needs of engineering applications. The simplified point cloud was used for discrete element modeling, improving the accuracy of subsequent rock mechanics analysis. In the application of rock mass structural surface identification, the simplified point cloud successfully identified 98.7% of important structural surfaces with a false identification rate of less than 1.5%, providing a reliable data foundation for tunnel stability analysis. In 3D modeling, the size of the model file generated by the simplified data was reduced by 76%, but the modeling accuracy was improved by 35%, mainly due to the accurate preservation of feature regions and the efficient simplification of flat areas.

[0043] A massive point cloud adaptive simplification system for mine roadways, based on a massive point cloud adaptive simplification method for mine roadways, includes: a data acquisition and preprocessing module, an adaptive neighborhood processing module, a point cloud intelligent classification module, and a dynamic simplification execution module; The data acquisition and preprocessing module is used to scan the mine roadway with a 3D laser scanner to acquire raw point cloud data, perform preliminary cleaning, and build a point cloud spatial index. The adaptive neighborhood processing module is used to calculate the point cloud normal vector and estimate the local density. It dynamically adjusts the neighborhood search radius based on the density value and outputs point cloud data with normal vector and density information. The point cloud intelligent classification module is used to calculate the local neighborhood entropy function to determine the optimal neighborhood size, extract the angle feature of the normal vector, and use the SVM algorithm to classify the point cloud, outputting a point cloud dataset with classification labels. The dynamic simplification execution module is used to set differentiated simplification rates for different feature regions based on classification results and local entropy values, iteratively execute simplification operations until the target simplification rate is reached, output the simplified point cloud file, and generate a processing report.

[0044] In this embodiment, the data acquisition and preprocessing module includes a scanner interface unit, a high-speed data cache unit, and a filtering processing unit. The scanner interface unit supports real-time data reception from 3D laser scanners such as FARO Focus. The high-speed data cache unit is equipped with DDR4 memory, and the filtering processing unit executes a radius-based outlier removal algorithm. The adaptive neighborhood processing module integrates a PCA algorithm computing unit and a kernel density estimation computing unit. The PCA algorithm computing unit is configured to compute point cloud normal vectors in parallel under a GPU environment, and the kernel density estimation computing unit dynamically estimates the local point cloud density. The point cloud intelligent classification module includes an entropy function calculation unit, a feature extraction unit, and an SVM classification unit. The entropy function calculation unit calculates the local neighborhood entropy value based on the eigenvalues ​​of the covariance matrix. The feature extraction unit calculates the angle feature of the normal vector. The SVM classification unit uses the radial basis function as the kernel function and the Scikit-learn library to implement classification. The dynamic simplification execution module includes an entropy threshold judgment unit, a simplification strategy execution unit, and a result output unit. The entropy threshold judgment unit distinguishes between flat regions and feature regions based on local entropy values. The simplification strategy execution unit implements differentiated simplification. The result output unit generates a processing report containing the simplification rate and feature retention rate.

[0045] A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements an adaptive simplification method for massive point clouds in mine roadways.

[0046] A processing terminal includes a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the computer program, it implements an adaptive simplification method for massive point clouds in mine roadways.

[0047] In practical applications, the system adopts a high-performance GPU-accelerated server architecture in its hardware implementation. The specific configuration includes: dual Intel Xeon Gold 6338 processors (32 cores / 64 threads), two NVIDIA RTX A6000 GPUs (each equipped with 48GB GDDR6 video memory), 256GB DDR4 ECC memory, and a high-speed NVMe SSD storage array. This hardware configuration is specifically optimized for processing point cloud data of hundreds of millions of points, and the large-scale parallel computing capabilities provided by the GPU acceleration unit are the key to achieving real-time processing.

[0048] In terms of software implementation, the system uses the CUDA programming model to accelerate the core algorithms using GPUs, including: accelerated normal vector calculation: the PCA algorithm is reconstructed into a parallel version, utilizing thousands of GPU computing cores to process the normal vector calculation of tens of thousands of points simultaneously, achieving an acceleration of 8-10 times compared to the CPU version; optimized neighborhood search: a KD-tree structure is constructed in parallel on the GPU, reducing the neighborhood search time from near zero; accelerated classification algorithm: the entire SVM inference process is completed on the GPU, and the TensorRT framework is used for model optimization, achieving a classification throughput of millions of points per second.

[0049] System workflow and hardware co-optimization: Point cloud data is input into the system through a high-speed 10 Gigabit network interface. Initial filtering and indexing are performed on the CPU first, and then the data is transferred in batches to the GPU memory. All computationally intensive tasks (normal vector calculation, entropy calculation, classification reasoning) are executed in parallel on the GPU. This heterogeneous computing architecture makes full use of the massive parallelism of the GPU, enabling the system to process hundreds of millions of point cloud data that traditional CPUs cannot process in real time.

[0050] The data processing pipeline has been specially optimized to achieve pipeline parallelism between the CPU and GPU: while the GPU is processing the current frame of data, the CPU is simultaneously preparing the next frame of data. This overlapping execution strategy further improves the overall processing efficiency. In actual mine roadway monitoring scenarios, the system can achieve a throughput of over 20 million points per minute, fully meeting the needs of real-time mine monitoring.

[0051] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0052] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for adaptive simplification of massive point clouds for mine galleries, characterized in that, The method comprises the following steps: S1. Scanning the mine tunnel using a three-dimensional laser scanner, obtaining original point cloud data, performing preliminary cleaning and constructing a point cloud space index; S2. Calculating the point cloud normal vector based on the PCA method, estimating the local density using the kernel density estimation method, dynamically adjusting the neighborhood search radius according to the local density value, performing adaptive neighborhood search, and outputting the point cloud data with normal vector and density information; S3. Constructing a dimension feature function to quantify the local geometric features of the point cloud, calculating a local neighborhood entropy function to determine the optimal neighborhood size, extracting the normal vector angle feature, and using the SVM algorithm to classify the point cloud, outputting the point cloud dataset with classification labels, and dividing the point cloud into different feature regions, including flat feature regions, linear feature regions and complex feature regions; S4. According to the classification results and local entropy values, different simplification rates are set for different feature regions, and the simplification operation is iteratively performed until the target simplification rate is reached, and the simplified point cloud file is output.

2. The method according to claim 1, wherein, Step S1, the original point cloud data is preliminarily cleaned, including removing noise points and outliers, calculating the spatial distribution density of the point cloud, and removing isolated points using a radius-based filtering method; the space index is constructed using a KD-tree or Octree data structure.

3. The method according to claim 1, wherein, The specific content of step S2 is: S21. Establish an initial neighborhood around each point in the point cloud data, and use the K nearest neighbor method to preliminarily determine the neighborhood point set; S22. Calculate the local density of each center point using the radius-based kernel density estimation method; S23. Dynamically adjust the neighborhood search radius according to the local density value; S24. Calculate the normal vector of each point using the principal component analysis (PCA) method, specifically: calculate the covariance matrix of the neighborhood point set and perform eigenvalue decomposition on the covariance matrix, take the eigenvector corresponding to the smallest eigenvalue as the normal vector, and adjust the direction of the normal vector by dot product check with reference to the point cloud centroid.

4. The method according to claim 3, wherein, The local density of each center point is: wherein, is the three-dimensional coordinate of the current i-th point, is the three-dimensional coordinate of the i-th nearest neighbor point of the point , j is the three-dimensional coordinate of the i-th nearest neighbor point of the point k , is the number of neighborhood points searched using the K-Nearest Neighbors method, is the Euclidean distance between the current point and the neighborhood points, is the adaptive kernel width for the point ; The covariance matrix of the neighborhood points is: wherein, is the neighborhood barycenter, is the initial neighborhood centered at is the covariance matrix of the computed neighborhood points.

5. The method for adaptive simplification of massive point cloud for mine tunnel of claim 1, wherein, The specific content of step S3 is: S31. For each point, calculate the eigenvalues of the covariance matrix based on the point set within the adaptive neighborhood, and construct a dimension feature function to quantify the local geometric features of the point cloud; S32. Calculate the local neighborhood entropy function value, and iteratively search for the optimal neighborhood size to minimize the local neighborhood entropy function, and extract the feature description parameters of the point cloud; S33. Take the normal vector angle as a feature, calculate the normal vector angle between the point and the neighborhood points, use L1 regularization to select the optimal feature subset, and use the SVM algorithm with a radial basis function as the kernel function to classify the point cloud, and divide the point cloud into different feature regions, including flat feature regions, linear feature regions and complex feature regions.

6. The method for adaptive simplification of massive point cloud for mine tunnel of claim 5, wherein, The dimension feature function is: wherein, , , respectively represent the degree of linear feature, planar feature, and scattering feature, the sum is 1, λ1, λ2, λ3 are eigenvalues of the local neighborhood covariance matrix of each point, λ1≥λ2≥λ3; Local neighborhood entropy function is: The smaller the entropy value, the more ordered the local structure; the larger the entropy value, the more complex the structure.

7. The method for adaptive simplification of massive point cloud for mine tunnel of claim 1, wherein, The specific content of step S4 is: According to the point cloud classification results, a dynamic simplification strategy is adopted for different feature regions, a high simplification rate is used for flat feature regions, and a low simplification rate or all points are retained for linear feature regions and complex feature regions; A simplification threshold is set, and the local entropy value is calculated for each point to determine the feature region to which the point cloud belongs. If the entropy value is lower than the threshold and the point is located in a flat feature region, the random sampling or grid sampling method is used to remove the point. After simplification, the entropy value and classification of the remaining points are recalculated, and the cycle is executed until the target simplification rate or entropy value is stable.

8. A system for adaptive simplification of massive point clouds for mine galleries, characterized in that, The adaptive simplification method for massive point clouds in mine tunnels according to any one of claims 1-7 comprises a data acquisition and preprocessing module, an adaptive neighborhood processing module, a point cloud intelligent classification module, and a dynamic simplification execution module. The data acquisition and preprocessing module is used to scan the mine tunnel using a three-dimensional laser scanner, acquire original point cloud data, perform preliminary cleaning, and construct a point cloud spatial index. The adaptive neighborhood processing module is used to calculate point cloud normal vectors and estimate local density, dynamically adjust the neighborhood search radius according to the density value, and output point cloud data with normal vector and density information. The point cloud intelligent classification module is used to calculate the local neighborhood entropy function to determine the optimal neighborhood size, extract the normal vector angle feature, and use the SVM algorithm to classify the point cloud, outputting the point cloud data set with classification labels. The dynamic simplification execution module sets different simplification rates for different feature regions based on the classification results and local entropy values, iteratively executes the simplification operation until the target simplification rate is reached, and outputs the simplified point cloud file.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the adaptive simplification method for massive point clouds in mine tunnels according to any one of claims 1-7.

10. A processing terminal comprising a memory and a processor, the memory having stored therein a computer program executable on the processor, characterized in that, The processor executes the computer program to implement the adaptive simplification method for massive point clouds in mine tunnels according to any one of claims 1-7.

Images