Coal mine deep roadway surface reconstruction method based on triangular prism voxel and spatial positioning

Abstract

This application relates to the field of digital modeling technology, and specifically discloses a method for reconstructing the surface of deep coal mine roadways based on triangular prism voxels and spatial positioning. An electronic device acquires the surrounding rock point cloud of a target coal mine roadway segment and divides the surrounding rock point cloud into multiple slices. For each slice, triangular prism voxelization downsampling is performed to obtain triangular prism voxels that conform to the roadway cross-sectional shape. Surrounding rock points in the surrounding rock point cloud are assigned to triangular prism voxels to determine non-empty voxels and empty voxels. A first centroid list of non-empty voxels is constructed. The first centroid list indicates the mapping relationship between the non-empty voxel number and the centroid coordinates of the non-empty voxels. Based on the first centroid list, the centroids of the empty voxels are filled in to determine a second centroid list. Finally, based on the second centroid list, a triangular mesh model is constructed to obtain the target model of the target coal mine roadway segment, thereby effectively improving the accuracy and efficiency of deep coal mine roadway modeling.

Description

Technical Field

[0001] This application relates to the field of digital modeling technology, specifically to a method for reconstructing the surface of deep coal mine roadways based on triangular prism voxels and spatial positioning. Background Technology

[0002] In the construction and safety monitoring of coal mine roadways, deep roadways are prone to geological disasters such as deformation and convergence due to high ground stress. Rapid and accurate reconstruction of the three-dimensional model of the roadway surface is crucial for deformation monitoring and disaster early warning.

[0003] Current modeling methods, such as manual measurement and total station measurement, suffer from low efficiency and limited coverage, making them unsuitable for large-scale, high-precision modeling. While 3D laser scanning technology can acquire high-density point cloud data and has become the mainstream technology for tunnel modeling, it faces several core challenges in data processing: First, the use of cubic voxels for downsampling results in poor compatibility between cubic voxels and the geometric shapes of long cylindrical or arched roofs in tunnels, leading to insufficient fitting accuracy for tunnel cross-sectional contours and significant residual noise in the filtered point cloud. Second, deep tunnels are prone to large-area point cloud gaps due to equipment obstruction and scanning blind spots. Existing symmetry-based completion methods are only suitable for small-scale gaps. Deep learning-based methods offer a more comprehensive solution. The completion methods of DL require a large amount of labeled training data, have high computational costs and weak generalization ability, and cannot achieve efficient completion of large-area point cloud missing data; thirdly, the surface reconstruction algorithms from point cloud to mesh model (such as Poisson reconstruction) are limited by point cloud accuracy, the effect of ordinary voxel filtering and large-area point cloud missing data. The reconstructed model surface is prone to unevenness and distortion at the point cloud missing areas. Additional filtering algorithms are required to ensure the effect.

[0004] It is evident that the aforementioned problems collectively restrict the accuracy and efficiency of deep coal mine roadway modeling. There is an urgent need for a coal mine roadway surface reconstruction method that adapts to the roadway geometry, can efficiently handle point cloud deficiencies, and has a simple reconstruction process, in order to support the dynamic monitoring and safety assessment of deep roadways. Summary of the Invention

[0005] This application provides a method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning, which can effectively improve the accuracy and efficiency of deep coal mine roadway surface reconstruction.

[0006] To achieve the above objectives, this application provides the following technical solution:

[0007] In a first aspect, embodiments of this application provide a method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning, the method comprising:

[0008] The surrounding rock point cloud of the target coal mine roadway section is obtained and divided into multiple slices; the surrounding rock point cloud represents the point cloud corresponding to the surrounding rock area within the target coal mine roadway section.

[0009] For each slice, triangular prism prismatic downsampling is performed to obtain triangular prism pixels; the triangular prism pixels conform to the cross-sectional morphology of the target coal mine roadway section;

[0010] The surrounding rock points in the surrounding rock point cloud are assigned to triangular prism voxels to determine non-empty voxels and empty voxels; non-empty voxels represent voxels containing surrounding rock points, and empty voxels represent voxels without surrounding rock points.

[0011] Construct a first list of centroids for non-empty voxels; the first list of centroids is used to indicate the mapping relationship between the numbers of non-empty voxels and the coordinates of their centroids.

[0012] The centroids of empty voxels are filled in based on the first centroid list to determine the second centroid list; the second centroid list includes the mapping relationship between the number of non-empty voxels and the centroid coordinates of non-empty voxels, as well as the mapping relationship between the number of empty voxels and the centroid coordinates of empty voxels.

[0013] Based on the second centroid list, a triangular mesh model is constructed to obtain the target model of the target coal mine roadway segment.

[0014] In some embodiments of this application, triangular prism voxelization downsampling is performed to obtain triangular prism voxels, including:

[0015] Within a cross-section perpendicular to the tunnel excavation direction, construct the axial bounding box of the slice; the axial bounding box represents the smallest cube that encloses the slice.

[0016] Divide the four sides of the axial bounding box into four equal parts according to a preset number of equal parts to obtain the divided line segments, and determine the dividing points of the axial bounding box.

[0017] Connect the dividing point to the two endpoints of the line segments after the division on each side to form multiple triangular bases;

[0018] A stretching operation is performed on the base of multiple triangles in the opposite direction of the tunnel excavation to obtain multiple triangular prism elements; wherein the stretching thickness of the stretching operation is the same as the thickness of the slice.

[0019] In some embodiments of this application, constructing a first list of centroids for non-empty voxels includes:

[0020] The non-empty voxels are assigned numbers based on the slices, edges, and segmented line segments corresponding to them, thus obtaining the non-empty voxel numbers; where the non-empty voxel numbers include the slice number, edge number, and line segment number.

[0021] The average value of the three-dimensional coordinates of all surrounding rock points within the non-hollow voxel is determined as the centroid coordinates of the non-hollow voxel.

[0022] Establish a mapping relationship between the numbers of non-empty voxels and their corresponding centroid coordinates to obtain the first centroid list.

[0023] In some embodiments of this application, the centroids of empty voxels are filled in based on a first centroid list to determine a second centroid list, including:

[0024] Based on the geometric feature region to which the hollow voxel belongs in the cross section of the surrounding rock of the roadway, the target non-hollow voxel corresponding to the hollow voxel is determined in the first centroid list;

[0025] Determine the centroid coordinates of the empty voxels based on the centroid coordinates of the target non-empty voxels;

[0026] The mapping relationship between the empty voxel number and the corresponding empty voxel centroid coordinates is added to the first centroid list to obtain the second centroid list.

[0027] In some embodiments of this application, based on the geometric feature region to which the empty voxel belongs in the cross-section of the surrounding rock of the roadway, the target non-empty voxel corresponding to the empty voxel is determined in the first centroid list, including:

[0028] When the geometric feature region is a straight line feature region, determine whether there is a first non-empty voxel in the first centroid list that is greater than or equal to a preset number and has the same slice number and edge number as the empty voxel;

[0029] If a preset number of first non-empty voxels exist in the first centroid list, then the target non-empty voxel is determined based on the first non-empty voxels; otherwise, it is determined whether there is a preset number of second non-empty voxels in the first centroid list that have the same edge number and line segment number as the empty voxel.

[0030] If a predetermined number of second non-empty voxels exist in the first centroid list, the target non-empty voxel is determined based on the second non-empty voxels; otherwise, the target non-empty voxel is determined based on the third non-empty voxel in the first centroid list that is symmetrical to the position of the empty voxel.

[0031] In some embodiments of this application, based on the geometric feature region to which the empty voxel belongs in the cross-section of the surrounding rock of the roadway, the target non-empty voxel corresponding to the empty voxel is determined in the first centroid list, including:

[0032] When the geometric feature region is a curved feature region, determine whether there is a second non-empty voxel in the first centroid list that is greater than or equal to a preset number;

[0033] If a predetermined number of second non-empty voxels exist in the first centroid list, the target non-empty voxel is determined based on the second non-empty voxels; otherwise, the target non-empty voxel is determined based on the third non-empty voxel in the first centroid list that is symmetrical to the position of the empty voxel.

[0034] In some embodiments of this application, a triangular mesh model is constructed based on a second centroid list to obtain a target model for the target coal mine roadway segment, including:

[0035] Within the same slice, adjacent bulk centroids are connected based on the second centroid list to obtain the tunnel cross-sectional outline.

[0036] Connect the centroids of voxels with the same edge and line segment numbers in adjacent slices to obtain multiple spatial quadrilaterals;

[0037] Each spatial quadrilateral is divided into two triangular faces, and the coordinates of all the vertices of the triangular faces and their connection relationships are summarized to obtain the target model.

[0038] In some embodiments of this application, the surrounding rock point cloud is segmented into multiple slices, including:

[0039] Along the tunnel excavation direction, the surrounding rock point cloud is divided into multiple slices of equal thickness according to the preset length.

[0040] In some embodiments of this application, obtaining the surrounding rock point cloud of a target coal mine roadway section includes:

[0041] Obtain the first point cloud of the target coal mine roadway section;

[0042] The trained model is used to perform semantic segmentation on the first point cloud to obtain semantic labels;

[0043] Remove points in the first point cloud whose semantic label is not surrounding rock, and determine the surrounding rock point cloud based on the remaining points whose semantic label is surrounding rock.

[0044] Secondly, embodiments of this application provide an electronic device, including a processor and a memory storing processor-executable instructions; when the instructions are executed by the processor, the above-mentioned method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning is implemented. Attached Figure Description

[0045] To more intuitively illustrate the prior art and this application, several exemplary figures are provided below. It should be understood that the specific shapes and structures shown in the figures should not generally be regarded as limiting conditions for implementing this application; for example, based on the technical concept disclosed in this application and the exemplary figures, those skilled in the art are able to easily make conventional adjustments or further optimizations to the addition / reduction / classification, specific shapes, positional relationships, connection methods, size ratios, etc. of certain units (components).

[0046] Figure 1 A schematic diagram illustrating the implementation process of the method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning, provided in this application embodiment;

[0047] Figure 2 A schematic diagram illustrating the visualization result of point cloud after semantic segmentation, provided in an embodiment of this application;

[0048] Figure 3 A schematic diagram of the surrounding rock point cloud provided in the embodiments of this application;

[0049] Figure 4 A schematic diagram of the element numbering of the triangular prism provided in the embodiments of this application;

[0050] Figure 5 This is a schematic diagram of downsampled point cloud data provided in an embodiment of this application;

[0051] Figure 6 A schematic diagram of the completed point cloud data provided for an embodiment of this application;

[0052] Figure 7 A schematic diagram illustrating the spatial quadrilateral partitioning rules provided in an embodiment of this application;

[0053] Figure 8 A schematic diagram showing the result of triangular facet creation provided in an embodiment of this application;

[0054] Figure 9 A schematic diagram of the triangular mesh model provided in the embodiments of this application;

[0055] Figure 10 This is a schematic diagram of the composition structure of the electronic device provided in the embodiments of this application.

[0056] Figure label:

[0057] Electronic device 1, processor 11, memory 12, communication interface 13, bus 14. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application. Any combination of different embodiments is possible.

[0059] In the description of this application: unless otherwise stated, "a plurality of" means two or more. The terms "first," "second," "third," etc., in this application are intended to distinguish the objects referred to and do not have any special meaning in terms of technical connotation (e.g., they should not be construed as an emphasis on importance or order). Expressions such as "including," "comprising," and "having" also mean "not limited to" (certain units, components, materials, steps, etc.).

[0060] To address the problems existing in current coal mine roadway surface reconstruction methods, this application provides a method for deep coal mine roadway surface reconstruction based on triangular prism voxels and spatial positioning. This method features simple equipment, rapid modeling, and support for data completion. Point cloud processing is optimized through triangular prism voxel structure, and downsampling accuracy is improved and outliers are reduced using voxel meshes adapted to roadway morphology. A numbering system enables rapid completion of large-area point cloud gaps, avoiding reliance on deep learning (DL) data. Furthermore, direct connection of voxel cores to construct triangular meshes simplifies the process while ensuring model integrity. Ultimately, this method aims to provide an efficient and complete solution for digital modeling and deformation monitoring of deep coal mine roadways.

[0061] This application provides a method for reconstructing the surface of deep coal mine roadways based on triangular prism voxels and spatial positioning, such as... Figure 1 As shown, the method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning may include the following steps:

[0062] Step 101: Obtain the surrounding rock point cloud of the target coal mine roadway section and divide the surrounding rock point cloud into multiple slices; the surrounding rock point cloud represents the point cloud corresponding to the surrounding rock area within the target coal mine roadway section.

[0063] In the embodiments of this application, the electronic device can first acquire the surrounding rock point cloud of the target coal mine roadway section and divide the surrounding rock point cloud into multiple slices; the surrounding rock point cloud represents the point cloud corresponding to the surrounding rock area within the target coal mine roadway section.

[0064] In the embodiments of this application, the target coal mine roadway segment refers to a coal mine roadway area of ​​a specified length that needs to be three-dimensionally modeled, and is the modeling object of the deep coal mine roadway surface reconstruction method based on triangular prism voxels and spatial positioning.

[0065] In the embodiments of this application, the surrounding rock point cloud represents the point cloud corresponding to the surrounding rock area within the target coal mine roadway section, that is, it only contains the point cloud data of the roadway surrounding rock, and removes non-surrounding rock interference point clouds such as equipment, personnel, and pipelines.

[0066] In the embodiments of this application, a slice refers to an independent point cloud segment obtained by dividing a continuous surrounding rock point cloud along a specified direction, and each slice is the basic unit for subsequent voxel construction.

[0067] In some embodiments of this application, when the electronic device divides the surrounding rock point cloud into multiple slices, it can divide the surrounding rock point cloud into multiple slices of equal thickness along the tunnel excavation direction according to a preset length.

[0068] In the embodiments of this application, the tunnel excavation direction refers to the advancing direction of coal mine tunnel construction, that is, the length direction of the tunnel, which is the reference direction for the segmentation of the surrounding rock point cloud.

[0069] In the embodiments of this application, the preset length refers to the pre-set length of a single slice along the tunnel excavation direction, which is the basis for determining the slice thickness.

[0070] In some embodiments of this application, the preset length can be set based on the combined requirements of coal mine roadway modeling accuracy and computational efficiency, for example, 0.2m, to ensure that the modeling accuracy reaches the centimeter level while taking into account the computational efficiency of data processing; each slice after segmentation contains a complete roadway cross-sectional structure, ensuring that a single slice can independently represent the roadway cross-sectional shape at the corresponding location, and all slices are arranged continuously along the roadway excavation direction without overlap or gaps, ensuring the continuity of the final generated target model.

[0071] In the embodiments of this application, all slices have the same length along the tunnel excavation direction, that is, the same thickness, to ensure the uniformity of subsequent voxel construction.

[0072] In some embodiments of this application, when the electronic device acquires the surrounding rock point cloud of the target coal mine roadway section, it can acquire the first point cloud of the target coal mine roadway section; use the trained model to perform semantic segmentation on the first point cloud to obtain semantic labels; remove points in the first point cloud whose semantic labels are not surrounding rock, and determine the surrounding rock point cloud based on the remaining points whose semantic labels are surrounding rock.

[0073] In the embodiments of this application, the first point cloud refers to the original point cloud data of the target coal mine roadway section obtained by the scanning device, which includes the point cloud of all scanned objects such as surrounding rock, equipment, personnel, and pipelines, without any filtering processing.

[0074] In the embodiments of this application, the trained model refers to a deep learning model that has been trained on a point cloud dataset and has the ability to perform point cloud semantic segmentation.

[0075] In the embodiments of this application, semantic segmentation refers to the process of classifying each point cloud in the first point cloud, that is, determining the category of the scanned object to which each point cloud belongs.

[0076] In the embodiments of this application, semantic labels refer to labels assigned to each point cloud through semantic segmentation that characterize its category, such as "surrounding rock", "equipment", "personnel", etc.

[0077] In some embodiments of this application, the first point cloud can be acquired by a lidar mounted on the tail of the tunneling machine. Combined with Simultaneous Localization and Mapping (SLAM) algorithms, such as Lidar Odometry and Mapping (LOAM) and Fast-LIO, the point clouds of each frame are stitched together. At the same time, motion distortion is eliminated by synchronizing the timestamps of the laser point cloud with the data from the Inertial Measurement Unit (IMU). The relative error of the stitched point cloud is less than 2 cm, ensuring the accuracy of the original point cloud data.

[0078] In some embodiments of this application, the trained model can be a PointNet++ semantic segmentation model, using a self-built dataset containing labeled point clouds of roadway surrounding rock, pipelines, equipment, personnel, etc. for transfer learning, training until the accuracy on the validation set is stable at over 95%, ensuring the accuracy of semantic segmentation; after removing non-surrounding rock point clouds, the remaining point cloud data with semantic label "surrounding rock" are integrated to form a continuous and interference-free surrounding rock point cloud, providing a high-quality data source for subsequent slice segmentation.

[0079] Step 102: For each slice, perform triangular prism pixelation downsampling to obtain triangular prism pixels; the triangular prism pixels fit the cross-sectional shape of the target coal mine roadway section.

[0080] In the embodiments of this application, the electronic device can acquire the surrounding rock point cloud of the target coal mine roadway section, divide the surrounding rock point cloud into multiple slices, and then perform triangular prism pixelation downsampling on each slice to obtain triangular prism pixels; the triangular prism pixels fit the roadway cross-sectional shape of the target coal mine roadway section.

[0081] In the embodiments of this application, the triangular prism voxelization downsampling is a dimensionality reduction and compression process based on triangular prism voxels for sliced ​​point clouds. By constructing triangular prism voxels that fit the morphology of the tunnel, the discrete point cloud is managed in a structured manner. The core features of the voxel represent all point clouds within the voxel, thereby reducing the amount of point cloud data while preserving the geometric features of the tunnel.

[0082] In the embodiments of this application, the triangular prism refers to a triangular prism spatial grid unit formed by stretching a triangle as the base along the tunnel excavation direction, and it is the core data processing carrier of the method of this application.

[0083] In the embodiments of this application, the roadway cross-sectional shape refers to the cross-sectional geometry perpendicular to the coal mine roadway excavation direction, such as rectangle, arch, semicircle, etc. The triangular prism element of this application can be adapted to the above-mentioned various roadway cross-sectional shapes.

[0084] It should be noted that, in the embodiments of this application, after the surrounding rock point cloud is divided into multiple slices, all subsequent operations are performed on a slice-by-slice basis. That is, subsequent operations are performed on each slice, including triangular prism voxelization downsampling, determination of non-empty voxels and empty voxels, and construction of a centroid list, until the model construction of the target coal mine roadway segment is completed based on the second centroid list of all slices.

[0085] In some embodiments of this application, when the electronic device performs triangular prism voxelization downsampling to obtain triangular prism voxels, it can construct an axial bounding box for the slice within a cross-section perpendicular to the tunnel excavation direction; the axial bounding box represents the smallest cube enclosing the slice; the four sides of the axial bounding box are divided into four equal parts according to a preset number of divisions to obtain the segmented line segments, and the division points of the axial bounding box are determined; the division points are connected to the two endpoints of the segmented line segments on each side to form multiple triangular bases; a stretching operation is performed on the multiple triangular bases in the opposite direction of the tunnel excavation direction to obtain multiple triangular prism voxels; wherein the stretching thickness of the stretching operation is the same as the thickness of the slice.

[0086] In the embodiments of this application, the cross section perpendicular to the tunnel excavation direction refers to the cross section at 90° to the coal mine tunnel excavation direction, i.e., the cross section of the tunnel, which is the construction plane of the base of the triangular prism.

[0087] In the embodiments of this application, the preset number of equal divisions is the number of divisions of the axial bounding box edge that is pre-set based on the accuracy of roadway modeling and calculation efficiency, and can be adjusted according to actual needs.

[0088] In the embodiments of this application, the four edges are the four edges of the axially bounding box on the cross-section perpendicular to the tunnel excavation direction, which are the basis for voxel segmentation.

[0089] In the embodiments of this application, the segmented line segments are small line segments of equal length obtained by dividing the edge of the axial bounding box into a preset number of equal parts.

[0090] In the embodiments of this application, the dividing point is a reference point set within the axial bounding box for constructing the base of the triangle, and is the core of forming the base of the triangular prism.

[0091] For example, the dividing point can be the center point of the axial bounding box on the cross section perpendicular to the tunneling direction.

[0092] In the embodiments of this application, the stretching operation refers to the geometric operation of extending the base of a two-dimensional triangle along a specified direction to form a three-dimensional triangular prism space.

[0093] In the embodiments of this application, the stretching thickness refers to the distance the base of the triangle is stretched, that is, the axial length of the formed triangular prism element.

[0094] In the embodiments of this application, the thickness of a slice refers to the length of a single slice along the tunnel excavation direction.

[0095] In some embodiments of this application, during actual implementation, the four sides of the axial enclosure are numbered clockwise as side 1 (upper horizontal side), side 2 (right vertical side), side 3 (lower horizontal side), and side 4 (left vertical side). The position of the dividing point can be adaptively adjusted according to the cross-sectional shape of the tunnel. The default setting is the center point of the axial enclosure on the cross-section perpendicular to the tunneling direction. If the tunnel has an arched roof, the dividing point can be adjusted to H / 3 from the bottom of the enclosure (H is the height of the enclosure). The preset number of equal divisions can be set for the horizontal and vertical sides respectively, such as 40 equal divisions for the horizontal side and 24 equal divisions for the vertical side.

[0096] Step 103: Assign the surrounding rock points in the surrounding rock point cloud to triangular prism voxels to determine non-empty voxels and empty voxels; non-empty voxels represent voxels containing surrounding rock points, and empty voxels represent voxels without surrounding rock points.

[0097] In the embodiments of this application, the electronic device performs triangular voxelization downsampling for each slice. After obtaining triangular voxels, the surrounding rock points in the surrounding rock point cloud can be assigned to triangular voxels to determine non-empty voxels and empty voxels. Non-empty voxels represent voxels containing surrounding rock points, and empty voxels represent voxels without surrounding rock points.

[0098] In the embodiments of this application, a surrounding rock point refers to a single point cloud data in the surrounding rock point cloud, which contains three-dimensional spatial coordinate information and is the basic unit constituting the surrounding rock point cloud.

[0099] In the embodiments of this application, a non-empty voxel means that at least one surrounding rock point is assigned to the triangular prism voxel. An empty voxel means that no surrounding rock point is assigned to the triangular prism voxel, which is mostly formed by point cloud defects caused by equipment occlusion or scanning blind spots.

[0100] Step 104: Construct a first centroid list for non-empty voxels; the first centroid list is used to indicate the mapping relationship between the numbers of non-empty voxels and the centroid coordinates of non-empty voxels.

[0101] In embodiments of this application, after the electronic device assigns the surrounding rock points in the surrounding rock point cloud to triangular prism voxels to determine non-empty voxels and empty voxels, it can fill in the centroids of the empty voxels based on a first centroid list to determine a second centroid list; the second centroid list includes the mapping relationship between the number of the non-empty voxel and the centroid coordinates of the non-empty voxel, and the mapping relationship between the number of the empty voxel and the centroid coordinates of the empty voxel.

[0102] In the embodiments of this application, the first centroid list refers to the associated list that records key information of non-empty voxels, and is the core data basis for subsequent empty voxel centroid filling.

[0103] In the embodiments of this application, the number refers to the identification information assigned to each triangular prism voxel to uniquely identify the spatial location of the voxel.

[0104] In the embodiments of this application, the centroid coordinates refer to the statistical characteristic values ​​of the three-dimensional spatial coordinates of all surrounding rock points within a voxel, which are used to represent the spatial position of the voxel.

[0105] In the embodiments of this application, when constructing the first centroid list of non-empty voxels, the electronic device can assign numbers to the non-empty voxels according to the slices, edges, and segmented line segments corresponding to the non-empty voxels, thereby obtaining the numbers of the non-empty voxels; wherein, the numbers of the non-empty voxels include slice number, edge number, and line segment number; the average value of the three-dimensional coordinates of all surrounding rock points within the non-empty voxel is determined as the centroid coordinates of the non-empty voxel; and a mapping relationship between the numbers of the non-empty voxels and the corresponding centroid coordinates is established to obtain the first centroid list.

[0106] In the embodiments of this application, the slice number is used to identify the slice to which the voxel belongs, and can characterize the position of the voxel along the tunnel excavation direction.

[0107] In the embodiments of this application, the edge number is used to identify the number of the four edges of the axial bounding box corresponding to the voxel, which can characterize the edge affiliation of the voxel on the tunnel cross section.

[0108] In the embodiments of this application, the line segment number is used to identify the sequence number of the segmented line segment corresponding to the voxel, which can characterize the position of the voxel on the corresponding edge.

[0109] In the embodiments of this application, both non-empty voxels and empty voxels are numbered using a three-level system, which includes slice level, edge level, and line segment level, thereby enabling precise positioning of the three-dimensional spatial location of voxels through numbering.

[0110] For example, the number assigned to the non-empty voxel is a globally unique third-level number V(i, j, k), where i is the slice number, j is the edge number, and k is the line segment number. Before assigning the number, all surrounding rock points in the slice must be assigned to the corresponding triangular prism voxels according to the YOZ plane coordinates (cross-sectional coordinates perpendicular to the excavation direction) of the surrounding rock points, and then the non-empty voxels are numbered.

[0111] In the embodiments of this application, the average value of the three-dimensional coordinates of all surrounding rock points refers to the value obtained by calculating the arithmetic mean of the coordinates of all surrounding rock points in the X, Y, and Z directions within the non-empty voxel, which is a representative value of the spatial position of the point cloud within the voxel.

[0112] For example, the formula for calculating the centroid coordinates is expressed as: , , ,in, This represents the number of surrounding rock points within a non-empty voxel. , as well as This represents the three-dimensional coordinates of the i-th surrounding rock point within the voxel.

[0113] It is understandable that the unique number of each non-empty voxel is matched one-to-one with the calculated centroid coordinates to form a structured association list, namely the first centroid list.

[0114] Step 105: Based on the first centroid list, complete the centroids of the empty voxels to determine the second centroid list; the second centroid list includes the mapping relationship between the numbers of non-empty voxels and the centroid coordinates of non-empty voxels, as well as the mapping relationship between the numbers of empty voxels and the centroid coordinates of empty voxels.

[0115] In embodiments of this application, after constructing a first centroid list of non-empty voxels, the electronic device can fill in the centroids of empty voxels based on the first centroid list to determine a second centroid list; the second centroid list includes the mapping relationship between the numbers of non-empty voxels and the centroid coordinates of non-empty voxels, and the mapping relationship between the numbers of empty voxels and the centroid coordinates of empty voxels.

[0116] In the embodiments of this application, "filling" refers to calculating the centroid coordinates of empty voxels based on the centroid coordinates of non-empty voxels through interpolation, coordinate transformation, and other methods to fill the voxel information gaps caused by the lack of point cloud.

[0117] Understandably, the second centroid list refers to the complete list formed by adding the numbers of empty voxels and the mapping relationship between the completed centroid coordinates on the basis of the first centroid list. It has no missing voxel information and is the basis for subsequent mesh model construction.

[0118] In some embodiments of this application, when the electronic device completes the centroids of empty voxels based on the first centroid list to determine the second centroid list, it can determine the target non-empty voxel corresponding to the empty voxel in the first centroid list based on the geometric feature region to which the empty voxel belongs in the roadway surrounding rock section; determine the centroid coordinates of the empty voxel based on the centroid coordinates of the target non-empty voxel; and add the mapping relationship between the number of the empty voxel and the corresponding centroid coordinates of the empty voxel to the first centroid list to obtain the second centroid list.

[0119] In the embodiments of this application, the geometric feature region is a different region divided according to the geometric shape of the cross-section of the surrounding rock of the tunnel, mainly including the straight feature region and the curved feature region.

[0120] In the embodiments of this application, the target non-empty voxel is a non-empty voxel selected from the first centroid list and used to calculate the centroid coordinates of the empty body, serving as a reference for filling in the centroid of the empty body.

[0121] In the embodiments of this application, the centroid coordinates of the empty voxel can be calculated based on the centroid coordinates of the target non-empty voxel through interpolation, coordinate transformation, and other methods.

[0122] In some embodiments of this application, when the electronic device determines the target non-empty voxel corresponding to the empty voxel in the first centroid list based on the geometric feature region to which the empty voxel belongs in the cross-section of the surrounding rock of the roadway, it can, if the geometric feature region is a straight line feature region, determine whether there is a first non-empty voxel in the first centroid list that is greater than or equal to a preset number and has the same slice number and edge number as the empty voxel; if there is a preset number of first non-empty voxels in the first centroid list, then the target non-empty voxel is determined based on the first non-empty voxels; otherwise, it determines whether there is a second non-empty voxel in the first centroid list that is greater than or equal to a preset number and has the same edge number and line segment number as the empty voxel; if there is a preset number of second non-empty voxels in the first centroid list, then the target non-empty voxel is determined based on the second non-empty voxels; otherwise, the target non-empty voxel is determined based on a third non-empty voxel in the first centroid list that is symmetrical to the position of the empty voxel.

[0123] In the embodiments of this application, the straight line feature area is a region in the cross-section of the surrounding rock of the tunnel that has a straight line shape, such as the sidewall and floor area of ​​a rectangular tunnel.

[0124] In the embodiments of this application, the preset quantity is the minimum number of target non-empty voxels that is pre-set to ensure the accuracy of empty body centroid filling; for example, the preset quantity can be 2.

[0125] In the embodiments of this application, the first non-empty voxel is a non-empty voxel with the same slice number and edge number as the empty voxel, that is, a non-empty voxel that is in the same slice and on the same axial bounding box edge as the empty voxel.

[0126] In the embodiments of this application, the second non-empty voxel is a non-empty voxel with the same edge number and line segment number as the empty voxel, that is, a non-empty voxel that is located on the same axial bounding box edge and the same segmented line segment as the empty voxel.

[0127] In the embodiments of this application, positional symmetry refers to the symmetrical spatial position of voxels on the cross-section of the surrounding rock of the roadway.

[0128] In the embodiments of this application, the third non-empty voxel refers to the non-empty voxel that is symmetrical to the empty voxel on the cross section of the surrounding rock of the roadway, and is the catch-up reference voxel for the core completion of the empty voxel.

[0129] For example, assuming the preset quantity is 4, the first non-empty voxels selected are the 4 non-empty voxels with the smallest difference from the line segment number of the empty voxel; the second non-empty voxels can be the 4 non-empty voxels with the smallest difference from the slice number of the empty voxel; the selection of the third non-empty voxel follows two symmetry rules: one is edge symmetry, that is, selecting the non-empty voxels with the same line segment number on the edge corresponding to the empty voxel j±2; the other is line segment symmetry, that is, selecting the non-empty voxels with the line segment number k'=NY+1-k on the edge corresponding to the empty voxel (NY is the number of equal parts of the horizontal edge, and k represents the line segment number of the empty voxel).

[0130] In some embodiments of this application, when the electronic device determines the target non-empty voxel corresponding to the empty voxel in the first centroid list based on the geometric feature region to which the empty voxel belongs in the cross-section of the surrounding rock of the roadway, it can determine whether there is a second non-empty voxel greater than or equal to a preset number in the first centroid list if the geometric feature region is a curved feature region; if there is a preset number of second non-empty voxels in the first centroid list, the target non-empty voxel is determined based on the second non-empty voxels; otherwise, the target non-empty voxel is determined based on a third non-empty voxel in the first centroid list that is symmetrical to the position of the empty voxel.

[0131] In the embodiments of this application, the curved feature region is a region in the cross-section of the surrounding rock of the roadway that has a curved shape, such as the arched roof or semi-circular roof region of a coal mine roadway.

[0132] In the embodiments of this application, the second non-empty voxel is preferentially selected as the target non-empty voxel in the curve feature region. If there is not a sufficient number of second non-empty voxels, the third non-empty voxel is selected as a fallback.

[0133] In some embodiments of this application, when determining the centroid coordinates of a hollow voxel based on the centroid coordinates of the target non-hollow voxel, firstly, the slice containing the hollow voxel is determined, and the average value of the X-axis coordinates of the centroids of all non-hollow voxels in that slice is calculated. Calculate the average coordinates of the centroids of all non-empty volumes on sides 1 and 3 of the axial bounding box on the Z-axis, and denote them as follows: and Calculate the average Y-axis coordinates of the centroids of all non-empty volumes on edges 2 and 4, and denote them as follows: and Calculate the unit line segment lengths DY and DZ along the Y-axis and Z-axis, respectively. The calculation methods for DY and DZ can be expressed as follows: and Where NY is the number of equal divisions of the horizontal edge and NZ is the number of equal divisions of the vertical edge; then, if the target non-empty voxel is a non-empty voxel on the same slice and the same edge, the centroid coordinates of the empty voxel can be determined using the same-edge completion formula: when j=1 (edge ​​1: upper horizontal edge) or j=3 (edge ​​3: lower horizontal edge), , , When j=2 (edge ​​2: right perpendicular edge) or j=4 (edge ​​4: left perpendicular edge), , , ;in, The line segment number is the empty voxel. For the target number of non-empty voxels, , , Here are the centroid coordinates of the target non-empty voxel; if the target non-empty voxel is an adjacent slice non-empty voxel on the same side and line segment, the slice direction completion formula is used: , , If the target non-empty voxel is a positionally symmetric non-empty voxel, the symmetric completion formula is used: , , .

[0134] Step 106: Construct a triangular mesh model based on the second centroid list to obtain the target model of the target coal mine roadway section.

[0135] In the embodiments of this application, after the electronic device completes the centroids of the empty voxels based on the first centroid list to determine the second centroid list, it can construct a triangular mesh model based on the second centroid list to obtain the target model of the target coal mine roadway segment.

[0136] In the embodiments of this application, the triangular mesh model refers to a continuous three-dimensional curved surface model formed by splicing multiple triangular faces, which can adapt to subsequent deformation monitoring, visualization and other needs.

[0137] Understandably, the target model refers to the final generated three-dimensional triangular mesh model that fully represents the geometry of the target coal mine roadway section.

[0138] In the embodiments of this application, when the electronic device constructs a triangular mesh model based on the second centroid list to obtain the target model of the target coal mine roadway segment, it can connect adjacent voxels within the same slice based on the second centroid list to obtain the roadway cross-sectional outline; connect the centroids of voxels with the same edge number and line segment number in adjacent slices to obtain multiple spatial quadrilaterals; divide each spatial quadrilateral into two triangular faces, and summarize the vertex coordinates and vertex connection relationships of all triangular faces to obtain the target model.

[0139] In the embodiments of this application, adjacent voxel centroids are the centroids of voxels that are spatially adjacent within the same slice, including the voxel centroids of adjacent line segment numbers on the same side and the voxel centroids of the beginning and end of adjacent sides.

[0140] In the embodiments of this application, the tunnel cross-section outline is a closed curve or broken line formed by connecting adjacent volume centers within the same slice, representing the tunnel cross-section geometry corresponding to a single slice.

[0141] In the embodiments of this application, adjacent slices are two slices with consecutive slice numbers along the tunnel excavation direction, such as slice number i and slice number i+1.

[0142] In the embodiments of this application, a spatial quadrilateral is a three-dimensional spatial quadrilateral formed by sequentially connecting the mass centers of two adjacent slice volumes with the same side number and the same line segment number with the mass centers of two adjacent slice volumes with adjacent line segment numbers. It is the basic unit for constructing a triangular mesh model.

[0143] In the embodiments of this application, a triangular facet refers to a two-dimensional triangular facet obtained by dividing a spatial quadrilateral, and is the basic unit constituting a triangular mesh model. When dividing a spatial quadrilateral, the short diagonal division rule can be followed, that is, connecting the centers of mass of a pair of shorter diagonals in the spatial quadrilateral to divide the spatial quadrilateral into two triangular faces.

[0144] In the embodiments of this application, vertex coordinates refer to the centroid coordinates of the three vertices of the triangular face.

[0145] In the embodiments of this application, vertex connection relationship refers to the connection order between the three vertices of a triangle face, representing the topological structure of the triangular mesh.

[0146] In some embodiments of this application, the connection of the centroids of adjacent voxels within the same slice includes two parts: first, connecting the centroids of adjacent line segment numbers on the same side; second, connecting the centroids of the first and last voxels on adjacent sides, ensuring that the resulting roadway cross-section outline is a closed outline. The segmentation of the spatial quadrilateral follows the short diagonal segmentation rule, that is, connecting the centroids of a pair of shorter diagonal voxels in the spatial quadrilateral divides the spatial quadrilateral into two triangular faces. This rule can ensure that the generated triangular faces are closer to equilateral triangles, avoid the appearance of long and narrow triangles, and ensure the manifoldness (continuous, foldless, and distortion-free) of the triangular mesh model. The summarized vertex coordinates and vertex connection relationships of the triangular faces can be saved as standard triangular mesh model files such as .ply or .obj. The resulting target model can completely characterize the geometric features of the surrounding rock surface of the target coal mine roadway section.

[0147] This application proposes a method for reconstructing the surface of deep coal mine roadways based on triangular prism voxels and spatial positioning. Through the innovative application of triangular prism voxels, combined with point cloud completion rules that fit the geometric features of the roadway and a simple triangular mesh construction method, the method solves the problems of low accuracy, low efficiency and poor point cloud completion effect of traditional coal mine roadway surface reconstruction methods. Triangular prism voxels are used instead of traditional cubic voxels for downsampling. These prism voxels closely match the geometry of long, cylindrical, and arched roofs in coal mine roadways and can adaptively adjust the segmentation point positions according to the roadway cross-sectional shape, significantly improving the fitting accuracy of the roadway cross-section and reducing outliers. Simultaneously, efficient compression of point cloud data is achieved, balancing modeling accuracy and computational efficiency. Based on the geometric characteristics of the roadway surrounding rock cross-section, straight and curved feature regions are divided, and differentiated target non-empty voxel selection rules are designed for different regions. Combined with three completion methods—same-edge completion, slice-direction completion, and symmetric completion—and corresponding coordinate calculation formulas, efficient completion of large-area point cloud gaps is achieved without relying on large amounts of labeled training data or strict symmetry assumptions. The completed point cloud... The absence of voids ensures the integrity of the model. Based on the globally unique numbering of voxels, adjacent voxel cores are directly connected to construct the roadway cross-section outline and spatial quadrilaterals. Then, triangular mesh models are formed by short diagonal segmentation, skipping the traditional complex Poisson reconstruction algorithm and eliminating the need for additional filtering algorithms, thus simplifying the modeling process. The segmentation rules ensure the manifold nature of the mesh model, resulting in a smooth surface and high geometric feature reproduction of the generated target model. The entire modeling process is structured and parameterized. Parameters such as slice length, number of equal parts, and preset completion quantity can be adjusted according to actual modeling needs. It can adapt to various coal mine roadway cross-section shapes such as rectangles, arches, and semicircles, and has strong versatility and adaptability. It is suitable for various scenarios such as digital modeling and deformation monitoring of deep coal mine roadways.

[0148] Based on the above embodiments, in another embodiment of this application, exemplarily, key parameters need to be preset before specific implementation. Taking a rectangular roadway with a cross-section of 6m (width) × 4.2m (height) in a coal mine as an example, the parameters are set as follows: Slice distance: L = 0.2m (along the tunneling direction X), this value is set based on the balance between roadway modeling accuracy requirements (centimeter level) and computational efficiency; Number of equal divisions of bounding box edges: the horizontal edge (Y direction) is divided into NY = 40 parts, and the vertical edge (Z direction) is divided into NZ = 24 parts; Segmentation point position: the default setting is H / 2 from the bottom (H is the height of the bounding box), which is suitable for rectangular cross-section roadways and semi-circular roof roadways; it can also be adjusted to H / 3 from the bottom according to the shape of the roof, such as the center of the arched or semi-circular roof being H / 3 from the bottom.

[0149] For example, the steps of the rapid and accurate modeling method for coal mine roadways based on triangular prism elements in this application include:

[0150] Step 1: Point cloud data acquisition and segmentation.

[0151] This step aims to obtain complete and continuous tunnel point cloud data, and then extract the target section for fine processing.

[0152] Step 1.1: Point cloud data stitching.

[0153] Using a lidar mounted on the tail of the tunneling machine, point clouds are stitched together in real time using SLAM algorithms (such as LOAM or Fast-LIO). IMU data and lidar point clouds are synchronized with timestamps to eliminate motion distortion and generate an original point cloud model covering the entire length of the tunnel. The stitching accuracy must be controlled to ensure that the relative error of the point clouds is less than 2 cm.

[0154] Step 1.2: Fragment extraction.

[0155] Based on the modeling requirements, a tunnel segment of a specific length (10m) is extracted from the original model. When extracting the segment, it is necessary to ensure that the segment contains the complete tunnel cross-section, and the coordinates of the segment's starting point in the global coordinate system are recorded to facilitate subsequent model alignment.

[0156] Step 2: Extraction of surrounding rock point cloud data.

[0157] This step uses deep learning and filtering techniques to accurately extract point clouds of the surrounding rock in complex scenes.

[0158] Step 2.1: Model training.

[0159] Using a self-built dataset (containing labeled point clouds of roadway surrounding rock, pipelines, equipment, personnel, etc.), we performed transfer learning on the PointNet++ network, and the accuracy on the validation set stabilized at over 95%.

[0160] Step 2.2: Semantic segmentation.

[0161] The extracted point cloud fragments are input into the trained PointNet++ model, which outputs the semantic label for each point. The output results are then visualized, as follows: Figure 2 The image shows the visualization result of the point cloud after semantic segmentation. Points labeled "surrounding rock" are retained, while all non-surrounding rock points (including belt conveyors, workers, pipelines, etc.) are filtered out, resulting in the surrounding rock point cloud, as shown below. Figure 3 As shown.

[0162] Step 3: Triangular prism simplification downsampling.

[0163] This step achieves efficient compression and structured management of point cloud data through triangular prism pixel meshes.

[0164] Step 3.1: Slice the point cloud data.

[0165] Along the tunnel excavation direction (X-axis), the surrounding rock point cloud is divided into multiple slices of equal thickness at a preset slicing distance (0.2m). Each slice is processed independently.

[0166] Step 3.2: Construction of triangular prism elements.

[0167] Perform the following sub-steps for each slice:

[0168] Step 3.2.1: Bounding box and edge numbering.

[0169] Calculate slice point cloud in In-plane axial bounding box, reading the bounding box in The four sides on the plane are numbered clockwise as side 1 (upper horizontal side), side 2 (right vertical side), side 3 (lower horizontal side), and side 4 (left vertical side).

[0170] Step 3.2.2: Parameter settings.

[0171] Set the coordinates of the split point O. By default, O is the bounding box within... The center point on the plane. It can also be set manually according to the shape of the top plate, for example, setting point O to (…). , , ), where H is the height of the bounding box. The horizontal sides are divided into NY=40 equal parts, and the vertical sides are divided into NZ=24 equal parts.

[0172] Step 3.2.3: Divide the sides equally and number the line segments.

[0173] Divide each edge into equal parts to obtain several line segments. Number the line segments on each edge sequentially from one end to the other along the positive direction of the coordinate axis (k=1, 2...).

[0174] Step 3.2.4: Triangle generation.

[0175] Connect the two endpoints of each line segment to the dividing point O, in Triangular regions are formed within the plane. Each triangle corresponds to a primitive base of a triangular prism.

[0176] Step 3.2.5: Generation of triangular prism elements.

[0177] All triangles generated in step 3.2.4 are aligned along the negative X-axis (from... arrive The material is stretched to a thickness equal to the slice distance (0.2m) to form a triangular prism voxel space. Each triangular prism voxel V is uniquely determined by its spatial position.

[0178] Step 3.3: Voxel numbering.

[0179] Each triangular prism voxel is assigned a three-level number (i, j, k). Here, i is the slice number (along the X direction), j is the edge number (1-4), and k is the line segment number on the edge. This numbering system ensures that each voxel has a unique identifier globally. A schematic diagram of the triangular prism voxels and their numbering is shown below. Figure 4 As shown, L is the slice thickness.

[0180] Step 3.4: Point cloud allocation.

[0181] Traverse all points within the slice and assign them to the corresponding triangular prism element based on their YOZ plane coordinates.

[0182] Step 3.5: Centroid calculation.

[0183] For each non-empty voxel, calculate the centroid coordinates of all points within it. The formula for calculating the centroid coordinates is: , , .

[0184] Step 3.6: Save the data.

[0185] Save a list containing voxel numbers (i, j, k) and their corresponding centroid coordinates. This completes the downsampling. The downsampled point cloud data is as follows: Figure 5 As shown. This processing can significantly reduce the amount of point cloud data; the specific reduction depends on the set slice distance and the number of equal divisions of the bounding box edges.

[0186] Step 4: Point cloud completion.

[0187] This step completes the data for empty voxels caused by occlusion, ensuring the integrity of the model.

[0188] Step 4.1: Execute the completion method according to the completion rules.

[0189] Iterate through all empty voxels multiple times. Let the coordinates of the completed centroid be (X, Y, Z) and the coordinates of the target non-empty voxel be (xi, yi, zi). Perform centroid coordinate completion for the empty voxels according to the following procedure:

[0190] First, determine the slice containing the empty voxel, and then calculate the average X-axis coordinate of the centroids of all non-empty voxels in that slice. Calculate the average coordinates of the centroids of all non-empty volumes on sides 1 and 3 of the axial bounding box on the Z-axis, and denote them as follows: and Calculate the average Y-axis coordinates of the centroids of all non-empty volumes on edges 2 and 4, and denote them as follows: and Calculate the unit line segment lengths DY and DZ along the Y-axis and Z-axis, respectively. The calculation methods for DY and DZ can be expressed as follows: and Where NY is the number of equal parts of the horizontal side and NZ is the number of equal parts of the vertical side.

[0191] When completing edge-to-edge fill, for empty voxels, select the non-empty voxels among the four voxels with the smallest difference between the segment index k' and k on edge j, and use the average of their centroid coordinates for fill-in. When j=1 (edge ​​1: upper horizontal edge) or j=3 (edge ​​3: lower horizontal edge), , , ,in, The line segment number is the empty voxel. For the target number of non-empty voxels, , , Let j be the centroid coordinates of the target non-empty voxel. When j=2 (edge ​​2: right perpendicular edge) or j=4 (edge ​​4: left perpendicular edge), , , .

[0192] When performing slice direction completion, for empty voxels, select the non-empty voxels among the four voxels whose edge j and line segment k are the same, and whose slice number i' has the smallest difference from i. Take the average of their centroid coordinates for completion. The completion formula is: , , .

[0193] During symmetric completion, for empty voxels, completion is based on symmetric voxel coordinate transformation, and the completion formula is as follows: , , .

[0194] Step 4.2: Update the list.

[0195] The completed centroid coordinates are updated in the voxel-centroid mapping list to obtain a complete, hole-free tunnel point cloud model. The completion result is as follows: Figure 6 As shown.

[0196] Step 5: Constructing the triangular mesh model.

[0197] This step connects the discrete centroids of the volume into a continuous triangular mesh surface.

[0198] Step 5.1: Join with slices.

[0199] Within the same slice i, connect the centroids of adjacent line segment voxels on the same side, such as V(i,j,k) and V(i,j,k+1), and connect the centroids of the first and last voxels on adjacent sides (V(i,1,1) and V(i,4,NZ), V(i,1,NY) and V(i,2,NZ), V(i,2,1) and V(i,3,NY), V(i,3,1) and V(i,4,1)) to form the outline within the slice.

[0200] Step 5.2: Constructing the spatial quadrilateral.

[0201] Connect the centroids of adjacent slices (i and i+1) that share the same edge j and the same line segment k, i.e., connect the two points V(i, j, k) and V(i+1, j, k). A spatial quadrilateral can be formed by connecting these points in the order V(i, j, k), V(i+1, j, k), V(i+1, j, k+1), and V(i, j, k+1).

[0202] Step 5.3: Triangulation.

[0203] like Figure 7 As shown, for each spatial quadrilateral, connecting the shorter pair of opposite corner points, such as C(n, j, k+1) and C(n+1, j+1, k), divides the quadrilateral into two triangular faces. This triangulation scheme ensures the manifold nature of the mesh, and the resulting triangular faces are shown below. Figure 8 As shown.

[0204] Step 5.4: Model output.

[0205] Save the vertex information (centroid coordinates) and connectivity of all triangular faces, and output a standard triangular mesh model file (e.g., .ply or .obj format). The output result is as follows: Figure 9 As shown in the figure, this model fully represents the surface geometry of the surrounding rock in the tunnel.

[0206] The method for reconstructing the surface of deep coal mine roadways based on triangular prism voxels and spatial positioning in this application mainly focuses on the application of triangular prism voxels, point cloud completion rules, and mesh construction methods.

[0207] In some embodiments of this application, triangular prism voxel downsampling mainly includes: slicing the surrounding rock point cloud equidistantly along the tunnel excavation direction, constructing a triangular prism voxel mesh based on the bounding box of each slice, assigning a globally unique number to each voxel, and calculating the centroid of the point cloud within each voxel; wherein, the construction of the triangular prism voxel mainly includes: reading the bounding box in... Define four sides on the plane and number them; set the coordinates of the dividing points and the number of equal parts for the horizontal and vertical sides; divide the horizontal and vertical sides into equal parts to obtain the set number of line segments, and number the line segments on each side in sequence; connect the two ends of each line segment to the dividing points to obtain the corresponding number of triangles; connect all the triangles to the bounding box. The plane is used to form a triangular prism space, and triangular prism voxels are constructed. Each triangular prism voxel is numbered in three levels: the first level is the slice number, the second level is the bounding box edge number of the voxel line segment, and the third level is the line segment number on the edge to which the voxel belongs. According to the coordinates of each point in the slice, they are assigned to the corresponding triangular prism voxels. The centroid coordinates of all points in each triangular prism voxel are calculated. According to the three-level numbering, the corresponding list of triangular prism voxels and their centroid coordinates is saved to obtain the tunnel point cloud model after filtering and downsampling of triangular prism voxels.

[0208] Furthermore, point cloud completion can be performed. For empty voxels that do not contain point cloud data, non-empty voxels at adjacent or symmetrical positions are queried based on their numbers, and the centroid coordinates of the current empty voxel are completed based on the centroid coordinates of the non-empty voxels. Three completion methods can be defined: edge completion, where for an empty voxel V(i, j, k), the difference between the line segment number k' of the non-empty voxel and the line segment number k of the empty voxel is found on its slice i and edge j. For the smallest n voxels, if there are at least two non-empty voxels V(i, j, k') that meet the conditions, interpolate and complete the slice by taking the average of their centroid coordinates; Slice direction completion: For an empty voxel V(i, j, k), find the difference between its edge j and its line segment k corresponding to the slice number i'. For the smallest n elements, if there are at least two non-empty voxels V(i', j, k) that meet the conditions, interpolate and complete the data using the average of their centroid coordinates. Symmetrical completion: For an empty voxel V(i, j, k), find the voxel V(i, j ± 2, k) on the same line segment k along the symmetrical edge (j ± 2), and complete it using coordinate symmetry transformation. Alternatively, find the voxel V(i, j, k') on the same edge j with line segment index k' = NY + 1 - k, and complete it using coordinate symmetry transformation, where NY is the number of equal divisions of the horizontal edge. The completion method can be selected based on the geometric characteristics of the roadway surrounding rock cross-section: For straight characteristic segments in the cross-section (such as the sidewalls and floor of a rectangular roadway), same-edge completion is preferred; for curved characteristic segments in the cross-section (such as an arched roof), if the empty voxel has enough adjacent non-empty voxels in the slicing direction, slicing direction completion is preferred; otherwise, symmetric completion is used.

[0209] In some embodiments of this application, all empty voxels can be traversed multiple times, a suitable completion method can be selected, and the corresponding coordinate calculation formula can be applied to determine the centroid coordinates of the empty voxels; the correspondence list between the triangular prism voxels and their centroid coordinates can be updated to obtain a complete, void-free tunnel point cloud model.

[0210] In some embodiments of this application, the construction of the triangular mesh model is as follows: based on the adjacency relationship represented by the voxel number, the centroids of adjacent voxels are connected to construct a triangular mesh model representing the surface of the roadway; the centroid coordinates of all adjacent voxels within the same slice are connected; the centroid coordinates of adjacent voxels with the same edge number and line segment number in all adjacent slices are connected to form several spatial quadrilaterals; for each spatial quadrilateral formed, the centroid coordinates of the pair of diagonal voxels with shorter length are connected to divide the spatial quadrilateral into two triangular faces, and all the established triangular faces are saved to obtain the triangular network model of the coal mine roadway.

[0211] In summary, this application uses triangular prism voxels instead of traditional cubic voxels. By equidistantly slicing along the tunnel excavation direction and constructing a voxel mesh based on the slice bounding boxes, the voxel structure better fits the long, cylindrical, and predominantly planar geometry of the tunnel, significantly reducing outliers and improving fitting accuracy. Compared to traditional voxel methods, the triangular prism voxels in this application, through adaptive adjustment of the segmentation points, can optimize the surface representation, resulting in a smoother point cloud. Simultaneously, it significantly reduces redundant points after downsampling, demonstrating its adaptability to the tunnel contour. By defining multiple completion rules and querying adjacent non-empty voxels based on the voxel numbering system, the centroid coordinates are directly interpolated, achieving efficient completion of large-area point cloud missing data without relying on large amounts of training data or strict symmetry assumptions. Compared to traditional completion methods and DL completion methods, the completion method in this application has higher computational efficiency, higher closeness to tunnel geometric features, and a significantly increased model area that can be completed. The triangular mesh is constructed by directly connecting the centroids based on the adjacency relationship of voxel numbers, which simplifies the surface reconstruction process. It eliminates the need for additional algorithms such as Poisson reconstruction, while ensuring mesh manifold, resulting in high modeling efficiency and a smoother model.

[0212] Based on the above embodiments, another embodiment of this application provides an electronic device, such as... Figure 10 As shown, the electronic device 1 proposed in this application embodiment may include a processor 11 and a memory 12 storing instructions executable by the processor 11; further, the electronic device 1 may also include a communication interface 13 and a bus 14 for connecting the processor 11, the memory 12 and the communication interface 13.

[0213] In the embodiments of this application, the processor 11 can be at least one of the following: Application-Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), controller, microcontroller, and microprocessor. It is understood that for different devices, the electronic device used to implement the above-mentioned processor function can also be other, and this application embodiment does not specifically limit it. The memory 12 can be connected to the processor 11, wherein the memory 12 is used to store executable program code, which includes computer operation instructions. The memory 12 may include high-speed RAM memory, and may also include non-volatile memory, such as at least two disk drives.

[0214] In embodiments of this application, bus 14 is used to connect communication interface 13, processor 11 and memory 12 to enable communication between these devices.

[0215] In embodiments of this application, memory 12 is used to store instructions and data.

[0216] In practical applications, the aforementioned memory 12 can be volatile memory, such as random-access memory (RAM), or non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state drive (SSD); or a combination of the above types of memory, and provide instructions and data to the processor 11.

[0217] Furthermore, in this embodiment, the functional modules can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional module.

[0218] If the integrated unit is implemented as a software functional module and is not sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the method of this embodiment.

[0219] Specifically, the program instructions corresponding to a method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning in this embodiment can be stored on storage media such as optical discs and hard disks. When the program instructions corresponding to the method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning in the storage media are read or executed by an electronic device, the following steps are included:

[0220] The surrounding rock point cloud of the target coal mine roadway section is obtained and divided into multiple slices; the surrounding rock point cloud represents the point cloud corresponding to the surrounding rock area within the target coal mine roadway section.

[0221] For each slice, triangular prism prismatic downsampling is performed to obtain triangular prism pixels; the triangular prism pixels conform to the cross-sectional morphology of the target coal mine roadway section;

[0222] The surrounding rock points in the surrounding rock point cloud are assigned to triangular prism voxels to determine non-empty voxels and empty voxels; non-empty voxels represent voxels containing surrounding rock points, and empty voxels represent voxels without surrounding rock points.

[0223] Construct a first list of centroids for non-empty voxels; the first list of centroids is used to indicate the mapping relationship between the numbers of non-empty voxels and the coordinates of their centroids.

[0224] The centroids of empty voxels are filled in based on the first centroid list to determine the second centroid list; the second centroid list includes the mapping relationship between the number of non-empty voxels and the centroid coordinates of non-empty voxels, as well as the mapping relationship between the number of empty voxels and the centroid coordinates of empty voxels.

[0225] Based on the second centroid list, a triangular mesh model is constructed to obtain the target model of the target coal mine roadway segment.

[0226] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of hardware embodiments, software embodiments, or embodiments combining software and hardware aspects. Furthermore, this application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage and optical storage) containing computer-usable program code.

[0227] This application is described with reference to schematic and / or block diagrams of implementations of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each step and / or block in the schematic and / or block diagrams, as well as combinations thereof, 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, create means for implementing the functions specified in one or more steps of the schematic and / or one or more blocks of the block diagrams.

[0228] 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 the implementation flow diagram. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0229] 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.

[0230] The above embodiments are merely preferred embodiments provided to fully illustrate this application, and the scope of protection of this application is not limited thereto. Equivalent substitutions or modifications made by those skilled in the art based on this application are all within the scope of protection of this application.

Claims

1. A method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning, characterized in that, The method includes: The surrounding rock point cloud of the target coal mine roadway section is obtained and the surrounding rock point cloud is divided into multiple slices; the surrounding rock point cloud represents the point cloud corresponding to the surrounding rock area within the target coal mine roadway section. For each slice, triangular prism prismatic downsampling is performed to obtain triangular prism prismatic elements; the triangular prism prismatic elements conform to the cross-sectional morphology of the target coal mine roadway segment; The surrounding rock points in the surrounding rock point cloud are assigned to the triangular prism voxels to determine non-empty voxels and empty voxels; the non-empty voxels represent voxels containing surrounding rock points, and the empty voxels represent voxels without surrounding rock points. Construct a first centroid list for the non-empty voxels; the first centroid list is used to indicate the mapping relationship between the numbers of the non-empty voxels and the centroid coordinates of the non-empty voxels; The centroids of the empty voxels are filled in based on the first centroid list to determine the second centroid list. The second centroid list includes the mapping relationship between the numbers of the non-empty voxels and the centroid coordinates of the non-empty voxels, as well as the mapping relationship between the numbers of the empty voxels and the centroid coordinates of the empty voxels. The centroid of the empty voxel is determined based on the target non-empty voxel corresponding to the empty voxel. Determining the target non-empty voxel corresponding to the empty voxel includes: if the geometric feature region to which the empty voxel belongs in the roadway surrounding rock section is a straight feature region, determining whether there exists a first non-empty voxel in the first centroid list that is greater than or equal to a preset number and has the same slice number and edge number as the empty voxel; if there exists the preset number of first non-empty voxels in the first centroid list, then the target non-empty voxel is determined based on the first non-empty voxels; otherwise, determining whether there exists a preset number of first non-empty voxels in the first centroid list that has the same edge number and line segment number as the empty voxel. The second non-empty voxels have the same number; if the first centroid list contains the preset number of the second non-empty voxels, then the target non-empty voxel is determined based on the second non-empty voxels; otherwise, the target non-empty voxel is determined based on the third non-empty voxel in the first centroid list that is symmetrical to the empty voxel; if the geometric feature region is a curved feature region, determine whether the first centroid list contains a number of second non-empty voxels greater than or equal to the preset number; if the first centroid list contains the preset number of the second non-empty voxels, then the target non-empty voxel is determined based on the second non-empty voxels; otherwise, the target non-empty voxel is determined based on the third non-empty voxel in the first centroid list that is symmetrical to the empty voxel; the slice number is the number used to identify the slice to which the voxel belongs; the edge number is the number used to identify the four edges of the axial bounding box corresponding to the voxel; the line segment number is the line segment identifier obtained by dividing the four edges of the axial bounding box into preset equal parts. Based on the second centroid list, a triangular mesh model is constructed to obtain the target model of the target coal mine roadway segment.

2. The method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning according to claim 1, characterized in that, The process of performing triangular prism prismatic downsampling to obtain triangular prism elements includes: Within a cross-section perpendicular to the tunnel excavation direction, an axial bounding box is constructed for the slice; the axial bounding box represents the smallest cube that encloses the slice. The four sides of the axial bounding box are divided into four equal parts according to a preset number of equal parts to obtain the divided line segments, and the division points of the axial bounding box are determined. Connect the dividing point to the two endpoints of the divided line segment on each side to form multiple triangular bases; A stretching operation is performed on the base of the plurality of triangles in the opposite direction of the tunnel excavation to obtain a plurality of triangular prism elements; wherein the stretching thickness of the stretching operation is the same as the thickness of the slice.

3. The method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning according to claim 2, characterized in that, The construction of the first centroid list of the non-empty voxels includes: The non-empty voxel is assigned a number based on the slice, edge, and segmented line segment corresponding to it, thus obtaining the number of the non-empty voxel; wherein the number of the non-empty voxel includes the slice number, edge number, and line segment number. The average value of the three-dimensional coordinates of all surrounding rock points within the non-hollow voxel is determined as the centroid coordinates of the non-hollow voxel. Establish a mapping relationship between the numbers of the non-empty voxels and the corresponding centroid coordinates to obtain the first centroid list.

4. The method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning according to claim 3, characterized in that, The step of filling in the centroids of the empty voxels based on the first centroid list to determine the second centroid list includes: Based on the geometric feature region to which the hollow voxel belongs in the cross-section of the surrounding rock of the roadway, the target non-hollow voxel corresponding to the hollow voxel is determined in the first centroid list; The centroid coordinates of the empty voxel are determined based on the centroid coordinates of the target non-empty voxel. The mapping relationship between the number of the empty voxel and the corresponding centroid coordinates of the empty voxel is added to the first centroid list to obtain the second centroid list.

5. The method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning according to claim 1, characterized in that, The construction of the triangular mesh model based on the second centroid list to obtain the target model of the target coal mine roadway segment includes: Within the same slice, adjacent bulk centroids are connected based on the second centroid list to obtain the tunnel cross-sectional outline. Connect the centroids of voxels with the same edge and line segment numbers in adjacent slices to obtain multiple spatial quadrilaterals; Each of the spatial quadrilaterals is divided into two triangular faces, and the coordinates of all the vertices of the triangular faces and their connection relationships are summarized to obtain the target model.

6. The method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning according to claim 1, characterized in that, The process of dividing the surrounding rock point cloud into multiple slices includes: Along the tunnel excavation direction, the surrounding rock point cloud is divided into multiple slices of equal thickness according to a preset length.

7. The method for reconstructing the surface of deep coal mine roadways based on triangular prism elements and spatial positioning according to claim 1, characterized in that, The acquisition of the surrounding rock point cloud of the target coal mine roadway section includes: Obtain the first point cloud of the target coal mine roadway segment; The trained model is used to perform semantic segmentation on the first point cloud to obtain semantic labels; Points in the first point cloud whose semantic label is not surrounding rock are removed, and the surrounding rock point cloud is determined based on the remaining points whose semantic label is surrounding rock.

8. An electronic device, characterized in that, The electronic device includes a processor, a memory, and a memory storing instructions executable by the processor; when the instructions are executed by the processor, the method for reconstructing the surface of deep coal mine roadways based on triangular prism voxels and spatial positioning as described in any one of claims 1 to 7 is implemented.

Images