A mine three-dimensional model point cloud data combination filtering method and medium

Abstract

This invention discloses a combined filtering method and medium for point cloud data of a 3D mine model. The method includes acquiring a 3D point cloud dataset A of mine roadways using a 3D laser scanning device; importing the original point cloud data set A into a "profile-axial-random" composite data simplification model for data simplification; and importing the simplified dataset D into a composite data denoising model to execute data denoising instructions. This combined filtering method consists of two parts: a composite filtering data simplification model and a composite denoising model. The composite simplification algorithm can significantly reduce the data volume while preserving the semi-circular arch contour features and detailed structure, thus reducing storage space usage. The composite denoising algorithm can effectively filter out various outliers such as salt-and-pepper noise and Gaussian noise to the greatest extent possible while ensuring model quality.

Description

Technical Field

[0001] This invention relates to the fields of computer vision technology and 3D point cloud data processing, specifically to a coarsening method for 3D mine models based on composite algorithms. Background Technology

[0002] With the advancement of modern measurement technology, 3D laser scanning technology has flourished in various fields due to its advantages such as high scanning speed, high efficiency, high accuracy, relatively low cost, and non-contact measurement. However, this method also has unavoidable drawbacks, such as the large space requirements for data storage, slow transmission speed, and a large workload for reverse model reconstruction. Furthermore, factors such as scanning accuracy contribute to high noise levels. Therefore, utilizing post-processing algorithms to simplify and filter point cloud data for noise reduction is crucial.

[0003] Although scholars both domestically and internationally have conducted extensive research on point cloud filtering, most existing studies only use one or two algorithms to process point cloud data and do not clearly distinguish between various filtering algorithms. The following problems exist when using various filtering algorithms to filter 3D point cloud data: ① Using a single filtering algorithm can reduce the massive amount of data to the target quantity, but it cannot remove noise from the data, or it solves the noise problem but cannot achieve a significant reduction in data volume; ② Using a combination of two or more algorithms can lead to problems with the selection of inappropriate combination algorithms, resulting in poor combined filtering effects; ③ During the filtering process, due to factors such as the incompatibility of the selected algorithm with the problem being solved, there are problems such as overly smoothed data or blurred data edges. This causes filtering algorithms to fail to achieve ideal results in different scenarios.

[0004] Currently, there is no method for combining and filtering three-dimensional point clouds of mine three-dimensional models. Filtering algorithms provide an effective means for simplifying and reducing noise in point cloud data in practical engineering. Summary of the Invention

[0005] To address the shortcomings and defects of existing point cloud filtering methods in mine roadway applications, this invention provides a combined point cloud data filtering method. The combined filtering method consists of two parts: a composite filtering data simplification model and a composite denoising model. The composite simplification algorithm can significantly reduce the amount of data while preserving the semi-circular arch contour features and detailed structure, thus reducing the storage space occupied. The composite denoising algorithm can effectively filter out various outliers such as salt-and-pepper noise and Gaussian noise to the greatest extent while ensuring model quality.

[0006] The technical solution adopted by this invention to solve its technical problem is as follows:

[0007] A method for combined filtering of point cloud data from a 3D mine model, based on the step-by-step characteristics of the data, divides the data filtering into two parts: data simplification and data noise reduction, specifically including the following steps:

[0008] Step 1: Obtain a 3D point cloud dataset A of the mine roadway. The dataset should contain at least one data point.

[0009] Step 2: Import the original point cloud data set A into the "profile-axis-random" composite data simplification model for data simplification;

[0010] Step 3: Import the simplified dataset D into the composite data denoising model and execute the data denoising command;

[0011] In step 2, data simplification includes:

[0012] Step 2.1: Taking the original 3D scan point cloud dataset A as the object and the cross-section of the roadway along its vertical major axis as the target, principal component analysis is used to calculate the normal vector of the cross-section, and then a rectangular coordinate system is established with the normal vector as the vertical axis; the curvature of the point to be determined is obtained by fitting the curve parameters; then, a conditional filter is created using the obtained normal and curvature data, and the curvature threshold is set according to the roadway model construction requirements, thereby obtaining the initial simplified dataset B of the roadway model "profile" data;

[0013] Step 2.2: Taking the coarsened dataset B of the "profile" as the object and the long axis direction of the roadway as the target, import the first coarsened dataset into the speed-up filtering model, set the grid parameters according to the actual conditions of the roadway, select the point cloud closest to the grid center to replace all the point clouds in the grid, and then obtain the model dataset C after the second coarsening.

[0014] Step 2.3: If the secondary coarsening data meets the requirements, proceed directly to step 3; otherwise, proceed to step 2.4.

[0015] Step 2.4: Taking the dataset C after secondary coarsening as the object, use a random function to generate random numbers that encompass all point cloud data of the model, and perform point cloud coarsening according to an equal probability pattern to obtain the model dataset D after tertiary coarsening.

[0016] In step 3, the data noise reduction instructions include:

[0017] Step 3.1 If the imported dataset is ordered point cloud data, then proceed to step 3.2. If the imported data is unordered point cloud data, then first perform ordering processing on the data to obtain the initial ordered point cloud dataset D′.

[0018] Step 3.2: Taking the simplified point cloud dataset D′ as the object, set the median radius m of the point cloud neighborhood and the attribute parameters. The point cloud data sequence within the neighborhood is f1, f2…f nTake m point cloud data from the ordered point cloud sequence, sort them by coordinate size, use the median of the sequence number as the calculation result, and then filter out outliers that are far away in the model to obtain the dataset d.

[0019] Step 3.3: Based on the characteristics of the model region, the model arch and the middle and lower part are divided into datasets. The part with many curved surface shapes and large curvature changes is divided into a dataset d1, while the part with clear edges and complex geometric structures is divided into dataset d2.

[0020] Step 3.4: Using dataset d1 as the processing object, define the curvature neighborhood range and Gaussian function standard deviation of the calculation points, calculate the normalized Gaussian curvature, and then use the normalized Gaussian curvature as the filter weight. For all input point cloud data, multiply it with the weight of the corresponding position, and then perform a weighted average of adjacent point clouds to obtain the final output value. Repeat this process until the complete dataset d1 is traversed, that is, multiply dataset d1 with the Gaussian curvature filter mask to obtain the Gaussian curvature filtered point cloud dataset d1′;

[0021] Step 3.5: Taking dataset d2 as the processing object, utilize the edge-preserving property of bilateral filtering, calculate the spatial proximity of each point to the center point, multiply it by the weight value of the data intensity, and then perform a convolution operation on the optimized weight and intensity. Multiply dataset d2 with the bilateral filtering mask to obtain the bilaterally filtered point cloud dataset d2′.

[0022] Step 3.6: Merge the processed point cloud datasets d1′ and d2′ to obtain the composite denoised model dataset E.

[0023] Furthermore, in step 2.1, the method for calculating the normal curvature is as follows:

[0024]

[0025] In the formula, Kn is the normal curvature, Ⅰ and Ⅱ are the first and second basic forms of the surface, respectively, E, F and G are the coefficients of the first basic form of the surface, and L, M and N are the coefficients of the second basic form of the surface.

[0026] Furthermore, in step 2.2, the size of the voxel grid in the X, Y, and Z directions is set according to the actual conditions of the roadway to control the size of the divided voxels, and the amount of data retained decreases as the set voxel size increases.

[0027] Furthermore, the calculation method for physical fitness is as follows:

[0028]

[0029] Among them, X centroid Y centroidZ centroid V represents the centroid in the X, Y, and Z directions after being sorted by coordinates. centroid For a non-empty voxel center, there are n point clouds in voxel V.

[0030] Furthermore, in step 3.2, the method for calculating the median of the point cloud data is as follows:

[0031]

[0032] In the formula, f1, f2…f n Given an ordered point cloud sequence, where m is the set window size, take m point cloud data points from the ordered point cloud sequence, sort them by coordinate size, and use the median of the sequence number as the calculation result.

[0033] Furthermore, in step 3.5, the weight calculation method for the bilateral filter is as follows:

[0034]

[0035] In the formula, w is the weight after bilateral filtering optimization, (k, l) is the coordinate of the center point of the window, and (i, j) is the coordinate of the other coefficients of the window.

[0036] The present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the aforementioned method for combining and filtering point cloud data of a three-dimensional mine model.

[0037] Compared with the prior art, the present invention has the following advantages:

[0038] Composite data simplification algorithms effectively avoid the problems of fine structure, over-smoothing in low-density areas, and unbalanced sampling rates inherent in single algorithms. They possess the unique advantages of maintaining distortion-free low-density areas and avoiding coarsening in high-density areas. Composite data denoising algorithms can effectively filter out various outliers such as salt-and-pepper noise and Gaussian noise without introducing new noise, achieving better denoising results. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of the algorithm of the present invention;

[0041] Figure 2 This is a structural diagram of the location of the sample section collected in Example 1.

[0042] Figure 3 This is a diagram showing the process and results of data simplification for the model in Example 1.

[0043] Figure 4 This is a flowchart and result diagram of the composite noise reduction process in Example 1. Detailed Implementation

[0044] 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.

[0045] Reference Figure 1 The above is a flowchart of a point cloud dataset processing method based on the present invention, illustrating the entire process from data simplification to data noise reduction.

[0046] A method for filtering three-dimensional laser scanning point cloud data includes the following steps:

[0047] Step 1: Obtain a 3D point cloud dataset. The dataset should contain at least one data point.

[0048] Step 2: Perform curvature sampling on the original dataset. Calculate the normals of the point cloud data, establish a coordinate system using the normal vectors, and perform a second-order fitting. Obtain the curvature of each point by fitting the surface, set a curvature threshold, and remove points with curvature less than the threshold.

[0049] Step 3: Perform voxel sampling on the point cloud dataset after curvature sampling. Set the voxel size, divide the acquired 3D point cloud dataset into 3D voxel grids, and obtain the index h of each point cloud data within the voxel grid. After sorting the indices, take the centroid to replace all points in the voxel.

[0050] Step four: If the number of point clouds after voxel sampling meets the requirement, proceed to step five; otherwise, randomly sample the point cloud dataset after voxel sampling. Set the final number of point clouds to be retained, and use a random function to generate random numbers to delete point clouds with equal probability, retaining the set number of point clouds.

[0051] Step 5: If the obtained original point cloud dataset is an ordered point cloud, proceed to step 6; otherwise, sort the unordered point cloud data, i.e., make the unordered point cloud ordered, and obtain ordered point cloud data.

[0052] Step 6: Perform median filtering on the ordered point cloud dataset. Set the neighborhood size for each point, sort all points within the defined neighborhood by their coordinates, and use the median of the sorted values ​​to replace all points in the calculated neighborhood.

[0053] Step 7: Segment the median-filtered dataset. The part of the data with many curved shapes and large curvature changes is divided into a dataset d1, while the part with clear edges and complex geometric structures is divided into dataset d2.

[0054] Step 8: Perform Gaussian curvature filtering on dataset d1. Define the curvature neighborhood range and standard deviation of the Gaussian function for each point, calculate the normalized Gaussian curvature, and then use the normalized Gaussian curvature as the filter weight. For all input point cloud data, multiply it by the corresponding weight, and then perform a weighted average of adjacent point clouds to obtain the final output value. Repeat this process until the entire dataset d1 is traversed, that is, multiply the dataset d1 by the Gaussian curvature filtering mask to obtain the Gaussian curvature filtered point cloud data d1′.

[0055] Step nine: Perform bilateral filtering on dataset d2. Bilateral filtering combines two Gaussian filters, one for calculating spatial proximity and the other for calculating pixel value similarity. σs and σr are set. By calculating the spatial proximity of each point to the center point, multiplying it by the weights calculated from pixel value similarity, and then convolving the optimized weights with the image, i.e., multiplying dataset d2 with the bilateral filtering mask to obtain the bilaterally filtered point cloud data d2′.

[0056] Step 10: Combine the point cloud data d1′ processed in Step 8 and the point cloud data d2′ processed in Step 9 to obtain the combined filtered point cloud data.

[0057] To make the above-mentioned objectives, features and advantages of the present invention more apparent and understandable, and in conjunction with specific application scenarios, the present invention provides an embodiment of a method for addressing the shortcomings and defects of existing point cloud filtering methods in mine roadway applications. The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] Example 1, referring to Figure 2 Taking the most typical semi-circular arch roadway in a mine as the analysis object, with refined structures such as air doors and windows, the model point cloud dataset (referred to as dataset A) is approximately 2.24 × 10⁻⁶. 7 Data points.

[0059] First, the dataset A is simplified, referring to... Figure 1 , Figure 3 The specific steps are as follows:

[0060] (1-1) Taking dataset A as the object and the cross-section perpendicular to the major axis of the roadway as the target, principal component analysis is used to calculate the normal vector of this cross-section, and then a rectangular coordinate system is established with the normal vector as the vertical axis; the curvature of the point to be determined is obtained by fitting the curve parameters; the method for calculating the normal curvature is as follows:

[0061]

[0062] In the formula, K n Let be the normal curvature, Ⅰ and Ⅱ be the first and second fundamental forms of the surface respectively, E, F, and G be the coefficients of the first fundamental form of the surface, and L, M, and N be the coefficients of the second fundamental form of the surface.

[0063] Then, a conditional filter was created using the obtained normal and curvature data. A curvature threshold was set according to the needs of the tunnel model construction, and point clouds below the curvature threshold were removed. The larger the threshold, the higher the data simplification rate, thus obtaining the initial simplified dataset B of the tunnel model "profile" data. At this time, the dataset size was 9.36 × 10⁻⁶. 6 The data reduction rate was 58.2%. This filtering method causes the least "damage" to the tunnel surface, can maximize the protection of the weak geometric structure of the arch surface, and is more suitable for smoothing point cloud data with high curvature changes, which is the key to data coarsening.

[0064] (1-2) Taking the coarsened dataset B of the "profile" as the object and the long axis direction of the roadway as the target, the coarsened dataset is imported into the speed-up filtering model. The size of the voxel grid in the X, Y, and Z directions is set according to the actual roadway conditions to control the size of the divided voxels. The amount of data retained decreases as the set voxel size increases, which can quickly and efficiently reduce the amount of point cloud data and perform some smoothing correction on the data. The smoothing effect is better for point cloud data with low curvature changes. The calculation method for the center of gravity of each voxel is as follows:

[0065]

[0066] Among them, X centroid Y centroid Z centroid V represents the centroid in the X, Y, and Z directions after being sorted by coordinates. centroid For a non-empty voxel center, there are n point clouds in voxel V.

[0067] The point cloud closest to the grid center is selected to replace all point clouds in the grid, thus obtaining the model dataset C after secondary coarsening. At this point, the point cloud data size of dataset C is 1.9 × 10⁻⁶. 5 Compared to the first data simplification, the second data simplification achieved a simplification rate of 97.8%. The second data coarsening focuses on the "axial" cross-sectional data, which can usually achieve large-scale simplification of the point cloud dataset. The first coarsening based on curvature filtering is used as the control, thereby achieving maximum data simplification while ensuring model quality.

[0068] (1-3) is the optimal dataset for testing the model, and a third random data simplification is performed.

[0069] (1-4) Taking the dataset C after the second coarsening as the object, a random number generator is used to generate random numbers encompassing all point cloud data of the model. Point cloud coarsening is then performed according to an equal probability pattern to obtain the model dataset D after the third coarsening. At this point, the point cloud data size of dataset D is 1.5 × 10⁻⁶. 5 The data volume was reduced by 21.1% compared to the second data set. Random filtering is a proportional random reduction mode. Because it cannot consider the geometric features of the model and the reduction process is random, its accuracy is not controllable. However, under the premise of controlling the first and second model coarsening, the third coarsening can play a supplementary role, thereby achieving the goal of maximizing the model coarsening effect while ensuring quality.

[0070] Secondly, the simplified dataset D is imported into the composite denoising model for data denoising, referring to... Figure 1 , Figure 4 The specific steps are as follows:

[0071] (2-1) Perform point cloud detection on dataset D to determine that the dataset is unordered point cloud data. Since the composite denoising model is designed for ordered point cloud data, the dataset D is transformed into an ordered point cloud dataset D′ through data editing.

[0072] (2-2) Taking the simplified point cloud dataset D′ as the object, a median-based filtering method is selected. The median radius m of the point cloud neighborhood and attribute parameters are set, and the point cloud data sequence within the neighborhood is f1, f2…f n Take m point cloud data points from an ordered point cloud sequence. The method for calculating the median of the point cloud data is as follows:

[0073]

[0074] In the formula, f1, f2…f n Given an ordered point cloud sequence, where m is the set window size, take m point cloud data points from the ordered point cloud sequence, sort them by coordinate size, and use the median of the sequence number as the calculation result.

[0075] The data is sorted by coordinate size, and the median of the index is used as the calculation result. This process then filters out outliers that are far away from the model, resulting in dataset d. This process can remove noise and outliers from point clouds, and is mainly used to remove salt-and-pepper noise and impulse noise.

[0076] (2-3) Based on the characteristics of the model region, the model arch and the middle and lower part are divided into datasets. The part with many curved surface shapes and large curvature changes is divided into a dataset d1, while the part with clear edges and complex geometric structures is divided into dataset d2.

[0077] (2-4) Taking dataset d1 as the processing object, the curvature neighborhood range and Gaussian function standard deviation of the calculation point are set according to the scenario. A larger radius can take more points into account. In differential geometry, Gaussian curvature is the product of the principal curvatures k1 and k2 at that point.

[0078]

[0079] In the formula, K is the Gaussian curvature, Ⅰ and Ⅱ are the first and second basic forms of the surface, respectively, E, F, and G are the coefficients of the first basic form of the surface, and L, M, and N are the coefficients of the second basic form of the surface.

[0080] Calculate the normalized Gaussian curvature, and then use the normalized Gaussian curvature as the weight of the filter. For all input point cloud data, multiply it by the corresponding weight, and then perform a weighted average of adjacent point clouds to obtain the final output value. Repeat this process until the complete dataset d1 is traversed, that is, multiply the dataset d1 by the Gaussian curvature filter mask to obtain the Gaussian curvature filtered point cloud dataset d1′.

[0081] (2-5) Taking dataset d2 as the processing object, and utilizing the edge-preserving property of bilateral filtering, the spatial proximity of each point to the center point is calculated, multiplied by the weight value of the data intensity, and then the optimized weights and intensity are convolved. The weight calculation method of bilateral filtering is as follows:

[0082]

[0083] In the formula, w is the weight after bilateral filtering optimization, (k, l) is the coordinate of the center point of the window, and (i, j) is the coordinate of the other coefficients of the window.

[0084] Multiplying the dataset d2 with the bilateral filter mask yields the bilaterally filtered point cloud dataset d2′.

[0085] (2-6) Merge the processed point cloud datasets d1′ and d2′ to obtain the composite denoising model dataset E, which is the final model body after processing by the combined filtering method.

[0086] By using the method described in Example 1 above, 1. data filtering is finely divided into data simplification and data noise reduction, making the operation process more targeted and improving the accuracy of data coarsening. This effectively avoids the "collateral loss" of the dataset caused by the insufficient applicability of the method and makes it easier to ensure the quality of the coarsened model.

[0087] 2. The composite filtering data simplification model adopts a combination of "profile-axis-random" construction method, which can effectively avoid the problems of fine structure, excessive smoothing in low-density areas and unbalanced sampling rate that exist in single algorithms. It has the unique advantages of no distortion in low-density areas and no coarsening in high-density areas.

[0088] 3. The composite noise reduction model constructed based on the environmental characteristics of mine roadways can effectively filter out various outliers such as salt-and-pepper noise and Gaussian noise without introducing new noise, thus achieving better noise reduction results.

[0089] Example 2

[0090] This embodiment provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the mine 3D model point cloud data combination filtering method described in Embodiment 1.

[0091] This embodiment also provides an electronic device, including a memory and a processor. The memory is used to store computer programs, and the processor runs the computer programs to enable the electronic device to perform the aforementioned method for combining and filtering point cloud data of a three-dimensional mine model.

[0092] Alternatively, the aforementioned electronic device may be a server.

[0093] Embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0094] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0095] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0096] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0097] 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 systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0098] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for combined filtering of point cloud data from a 3D mine model, characterized in that, The method includes: Step 1: Obtain a 3D point cloud dataset A of the mine roadway. The dataset should contain at least one data point. Step 2: Import the original point cloud data set A into the "profile-axis-random" composite data simplification model for data simplification; Step 3: Import the simplified dataset D into the composite data denoising model and execute the data denoising command; In step 2, data simplification includes: Step 2.1: Taking the original 3D scan point cloud dataset A as the object and the cross-section of the roadway along its vertical major axis as the target, the principal component analysis method is used to calculate the normal vector of the cross-section, and then a rectangular coordinate system is established with the normal vector as the vertical axis; the curvature of the point to be determined is obtained by fitting the curve parameters; then, a conditional filter is created using the obtained normal and curvature data, and the curvature threshold is set according to the roadway model construction requirements, thereby obtaining the initial simplified dataset B of the roadway model "profile" data; Step 2.2: Taking the coarsened dataset B of the "profile" as the object and the long axis direction of the roadway as the target, import the first coarsened dataset into the speed-up filtering model, set the grid parameters according to the actual conditions of the roadway, select the point cloud closest to the grid center to replace all the point clouds in the grid, and then obtain the model dataset C after the second coarsening. Step 2.3: If the secondary coarsening data meets the requirements, proceed directly to step 3; otherwise, proceed to step 2.

4. Step 2.4: Using the dataset C after secondary coarsening as the object, generate random numbers encompassing all point cloud data of the model using a random function, and perform point cloud coarsening according to an equal probability pattern to obtain the model dataset D after tertiary coarsening. In step 3, the data noise reduction instructions include: Step 3.1 If the imported dataset is ordered point cloud data, proceed to step 3.

2. If the imported data is unordered point cloud data, first perform ordering processing on the data to obtain the initial ordered point cloud dataset D′. Step 3.2: Taking the simplified point cloud dataset D′ as the object, set the median radius m of the point cloud neighborhood and the attribute parameters. The point cloud data sequence within the neighborhood is f1, f2…f n Take m point cloud data from the ordered point cloud sequence, sort them by coordinate size, use the median of the sequence number as the calculation result, and then filter out outliers that are far away in the model to obtain the dataset d. Step 3.3: Based on the characteristics of the model region, the model arch and the middle and lower part are divided into datasets. The part with many curved surface shapes and large curvature changes is divided into a dataset d1, while the part with clear edges and complex geometric structures is divided into dataset d2. Step 3.4: Taking dataset d1 as the processing object, define the curvature neighborhood range and standard deviation of the Gaussian function for the calculation points, calculate the normalized Gaussian curvature, and then use the normalized Gaussian curvature as the weight of the filter. For all input point cloud data, multiply it with the weight of the corresponding position, and then perform a weighted average of adjacent point clouds together to obtain the final output value. Repeat this process until the complete dataset d1 is traversed, that is, multiply the dataset d1 with the Gaussian curvature filter mask to obtain the Gaussian curvature filtered point cloud dataset d1′. Step 3.5: Taking dataset d2 as the processing object, the edge-preserving property of bilateral filtering is utilized. The spatial proximity of each point to the center point is calculated and multiplied by the weight value of the data intensity. Then, the optimized weight and intensity are convolved. Dataset d2 is multiplied by the bilateral filtering mask to obtain the bilaterally filtered point cloud dataset d2′. Step 3.6: Merge the processed point cloud datasets d1′ and d2′ to obtain the composite denoised model dataset E.

2. The method according to claim 1, characterized in that, In step 2.1, the method for calculating the normal curvature is as follows: In the formula, Kn is the normal curvature, Ⅰ and Ⅱ are the first and second basic forms of the surface, respectively, E, F and G are the coefficients of the first basic form of the surface, and L, M and N are the coefficients of the second basic form of the surface.

3. The method according to claim 1, characterized in that, In step 2.2, the size of the voxel grid in the X, Y, and Z directions is set according to the actual conditions of the roadway to control the size of the divided voxels, and the amount of data retained decreases as the set voxel size increases.

4. The method according to claim 3, characterized in that, The method for calculating physical fitness is as follows: Among them, X centroid Y centroid Z centroid V represents the centroid in the X, Y, and Z directions after being sorted by coordinates. centroid Let V be a non-empty voxel center with n point clouds.

5. The method according to claim 1, characterized in that, In step 3.2, the method for calculating the median of the point cloud data is as follows: y i =Med{f i-v ,…,f i ,…,f i+v },i∈N, In the formula, f1, f2…f n Given an ordered point cloud sequence, where m is the set window size, take m point cloud data points from the ordered point cloud sequence, sort them by coordinate size, and use the median of the sequence number as the calculation result.

6. The method according to claim 1, characterized in that, In step 3.5, the weight calculation method for bilateral filtering is as follows: In the formula, w is the weight after bilateral filtering optimization, (k, l) is the coordinate of the center point of the window, and (i, j) is the coordinate of the other coefficients of the window.

7. A computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for combined filtering of point cloud data of a three-dimensional mine model as described in any one of claims 1-6.

Images