Point cloud dynamic three-dimensional reconstruction method based on greedy projection triangle reconstruction

Abstract

The invention discloses a point cloud dynamic three-dimensional reconstruction method based on greedy projection triangle reconstruction. The method comprises the following steps: S1, obtaining three-dimensional laser point cloud data in a tunneling roadway through a laser scanning technology; s2, loading the acquired three-dimensional laser point cloud data, and removing zero three-dimensional laser point cloud data or three-dimensional laser point cloud data with an incorrect format; s3, carrying out preprocessing operation on the three-dimensional laser point cloud data; s4, performing gridding processing, restoring a spatial topological structure between points in the three-dimensional laser point cloud data, and performing greedy projection triangle reconstruction by using optimization parameters to generate triangular mesh data; and S5, displaying and storing the roadway three-dimensional model according to the triangular mesh data. According to the method, efficient and complete three-dimensional modeling is carried out according to the point cloud data obtained after acquisition and processing, modeling trueness of entities such as coal walls without drawings, with various details and with dynamic changes is high, and the development requirements of digital transformation in the coal mine industry for the digital twinning technology are met.

Description

Technical Field

[0001] The present invention relates to the field of mine exploitation, and particularly to a point cloud dynamic three-dimensional reconstruction method based on greedy projection triangle reconstruction. Background Art

[0002] With the advancement of industrial intelligent manufacturing globally, the coal industry is also undergoing digital transformation. As one of the key technologies, digital twin provides a new solution for the intelligentization of coal mine exploitation. The government has also introduced a series of policies to encourage coal mine intelligentization. Therefore, the application of digital twin technology in the coal mine industry not only improves safety and production efficiency but also promotes the intelligentization and sustainable development of the industry, and is expected to play a greater role in the future.

[0003] Three-dimensional laser scanning technology, also known as reality capture technology, uses the principle of laser ranging to quickly obtain information such as three-dimensional coordinates and reflectivity of the object surface. It has characteristics such as high efficiency, high precision, non-contact, real-time dynamic, and automation, and can obtain three-dimensional data of the surface of the measured object with a large area and high resolution. It mainly consists of a three-dimensional laser scanner, a computer, a power supply system, a bracket, and system supporting software. It can quickly and accurately obtain spatial information, support remote control and automated operations, reduce the work intensity and safety risks of technicians, and improve operation efficiency and quality.

[0004] With the improvement of computer performance and the progress of digital image processing technology, the processing ability of point cloud data has been significantly improved. This enables researchers to extract accurate three-dimensional information from complex environments and objects and perform rapid real-time modeling.

[0005] However, in the digital twin technology for the mine field, for equipment with drawings, refined modeling can be carried out to improve authenticity, restore the real working conditions, and meet the needs of simulation, virtual reality, etc. But for models such as coal walls without drawings, with numerous details and dynamic changes, the existing technology usually adopts a simplified processing method, which cannot meet the requirements of real-time modeling and planned cutting use in roadway working faces, and cannot meet the development requirements of digital twin technology for the digital transformation of the coal mine industry. Summary of the Invention

[0006] The purpose of the present invention is to address the above problems and provide a point cloud dynamic three-dimensional reconstruction method based on greedy projection triangle reconstruction that can perform real-time modeling of roadway working faces.

[0007] To achieve the above purpose, the technical solution of the present invention is as follows: A point cloud dynamic three-dimensional reconstruction method based on greedy projection triangle reconstruction, comprising the following steps: S1. Obtain three-dimensional laser point cloud data in the driving roadway through laser scanning technology; S2. Load the acquired 3D laser point cloud data and remove the 3D laser point cloud data with zero quantity or incorrect format; S3. Perform preprocessing operations on the 3D laser point cloud data; S4. Use the greedy projection triangulation reconstruction algorithm to perform meshing on the preprocessed 3D laser point cloud data, restore the spatial topological structure between points in the 3D laser point cloud data, and use optimization parameters for greedy projection triangulation reconstruction to generate triangular mesh data; S5. Implement the display and saving of the 3D roadway model based on the triangular mesh data.

[0008] Furthermore, in step S1, set the left direction facing the coal wall as the positive y-axis direction, the working face advancing direction as the positive x-axis direction, and the upward direction perpendicular to the working face advancing direction as the positive z-axis direction, and establish a 3D coordinate system; let the laser scanner start from the origin of the 3D coordinate system and move along the positive y-axis direction to obtain the 3D laser point cloud data in the driving roadway.

[0009] Furthermore, in step S3, the preprocessing operations on the 3D laser point cloud data include the following steps: S31. Cut off the point cloud data outside the reconstruction range in the 3D laser point cloud data; S32. Filter and remove the duplicate point cloud data in the 3D laser point cloud data, and fill in the missing point cloud data; S33. Calculate the normal vector of each point in the 3D laser point cloud data; S34. Establish a search tree for quickly finding the position information of points; S35. Perform feature extraction on the 3D laser point cloud data; S36. Calculate the matching points of several partial point cloud data sets in the 3D laser point cloud data through ICP, and merge the several partial point cloud data sets in the 3D coordinate system to generate the 3D laser point cloud data of a complete scene; S37. Perform density increase or density decrease processing on the point cloud data in the 3D laser point cloud data to meet the requirements of subsequent operations.

[0010] Furthermore, step S4 specifically includes the following steps: S41. Set optimization parameters and find the nearest neighbor points of each point in the 3D laser point cloud data according to the search radius; S42. Construct a local triangulated surface based on the searched nearest neighbor points; S43. Iteratively add new points and update the triangulated surface; S44. Generate complete triangular mesh data.

[0011] Further, in step S41, the optimization parameters include the search radius of adjacent points, the ratio between the maximum side length and the search radius, the maximum number of adjacent points within the search radius, the maximum surface angle of the triangulated surface, the minimum allowable angle of the formed triangles, the maximum allowable angle of the formed triangles, and the normal vector consistency.

[0012] Compared with the prior art, the advantages and positive effects of the present invention are as follows: The present invention provides a point cloud dynamic three-dimensional reconstruction method based on greedy projection triangle reconstruction for the mining field. It uses three-dimensional laser scanning technology to obtain a three-dimensional coordinate set of a large number of points on the object surface as point cloud data, generates a triangular mesh by iteratively connecting adjacent points in the point cloud, and finally constructs a three-dimensional model of the entire roadway surface. This modeling method is suitable for a variety of model application scenarios and can adjust model parameters according to scene requirements. At the same time, it can dynamically adjust the model according to the update of point cloud data, and greatly improve the modeling efficiency through an optimization algorithm. Moreover, the present invention performs efficient and complete three-dimensional modeling based on the point cloud data obtained after collection and processing, and has a high degree of authenticity for modeling entities such as coal walls without drawings, with numerous details, and dynamic changes, and can fully meet the real-time modeling and planning cutting and other usage requirements of the roadway working face, meeting the development requirements of digital twin technology for the digital transformation of the coal mining industry, and having great market development prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0014] Figure 1 is a schematic diagram of the coordinate for point cloud data acquisition of the present invention; Figure 2 is a flowchart of the modeling method of the present invention; Figures 3A - 3B is a schematic diagram of point cloud data downsampling; wherein, Figure 3A is a schematic diagram before filtering, Figure 3B is a schematic diagram after filtering; Figure 4 is a schematic diagram of removing outliers from point cloud data; Figure 5 is a schematic diagram of the reconstructed model of point cloud data. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0015] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts, any modifications, equivalent replacements, improvements, etc., shall be included in the protection scope of the present invention.

[0016] This embodiment discloses a point cloud dynamic three-dimensional reconstruction method based on greedy projection triangle reconstruction. This method generates triangles by iteratively connecting adjacent points in the point cloud to construct the entire surface. The steps are as follows: Figure 2 shown; Step 1: Obtain three-dimensional laser point cloud data of the driving roadway through laser scanning technology: The laser scanner starts from the starting position (0, 0, 0) and rotates around the Z-axis to obtain the three-dimensional scene point cloud data of the driving roadway. The positive direction of the y-axis is defined as the left direction facing the coal wall, the positive direction of the x-axis is the driving direction of the driving roadway, and the positive direction of the z-axis is upward perpendicular to the driving direction of the working face; as shown in Figure 1 shown, Step 2: Load the laser point cloud data; 1. The system loads the point cloud data of the driving roadway collected by the laser scanner; 2. Check whether the data header contains valid information, and remove the point cloud data with zero quantity or incorrect format.

[0017] Step 3: Process the point cloud data, processing from multiple directions and dimensions such as filtering, normal vector, search tree, feature extraction, matching, etc., to improve data quality, adapt to different application requirements, reduce the burden of post-processing, and improve the overall system efficiency; the processing process is as follows: 1. Preprocessing of point cloud data; ① Filtering: Through voxel downsampling of the point cloud, the volume of the point cloud is effectively reduced, and the subsequent processing speed is accelerated; the given leaf_size value for downsampling can generally be taken as 0.08. The larger the value, the sparser the point cloud; as shown in Figure 3A , Figure 3B shown; ② Removing outliers: Through statistical filtering, calculate the distance between each point and its neighboring points, and identify and remove outliers according to the distribution of points. The average number of points in the point cloud is generally set to 50, and the standard deviation multiple threshold is generally set to 1.0; as shown in Figure 4 shown; ③ Data hole filling: For the missing parts in the point cloud data, adopt appropriate methods to identify holes, such as moving least squares and other methods, and fill them through interpolation to ensure the integrity of the data.

[0018] 2. Estimate the normal vector; provide key information about the shape and direction of the point cloud surface.

[0019] ① Set the search method: Build an octree or kd'Tree for searching; quickly find the position information of points, perform neighbor point search, etc.

[0020] ② Set the K-neighborhood or radius search: The size of the K-neighborhood is generally 50; ③ Perform the estimation of the normal vector.

[0021] 3. Extract the surface shape of the point cloud; ① Create a clustering extraction object and use Euclidean clustering for extraction; obtain key information and reduce the data volume, such as quickly finding the cut coal wall.

[0022] ② Set the cluster tolerance to 0.02 m, the minimum cluster to 100 points, and the maximum cluster to 15000 points; ③ Perform clustering and store the clustering results.

[0023] 4. Calculate the matching points of the point cloud through ICP and merge multiple point clouds.

[0024] ① Set the maximum number of iterations, generally 50; ② Set the tolerance of the transformation matrix, generally 0.00001; ③ Perform matching and store the results.

[0025] Step 4: The point cloud after the preprocessing operation is still an unorganized scattered point cloud, and there is no geometric connection between points. Therefore, to perform surface reconstruction, it is necessary to grid it to restore the spatial topological structure between points in the point cloud, that is, use the greedy projection triangulation reconstruction algorithm to search the local neighborhood of the point cloud and gradually construct a triangular mesh, including steps such as obtaining the adjacent point set of the data points, projecting to a two-dimensional plane, sorting by angle, and deleting data points according to visibility, to ensure the generation of high-quality non-overlapping triangles; For this, for the point cloud of the driving roadway, use optimized parameters to build a greedy projection triangulation reconstruction model, as Figure 5 shown; its steps are: 1. Given a search radius, find the nearest neighbor points of each point; 2. Construct a local triangulated surface based on these nearest neighbor points; 3. Iteratively add new points and update the triangulated surface; 4. Generate a complete triangulated mesh model.

[0026] Step 5: Triangular mesh optimization; In the greedy projection triangulation reconstruction algorithm, there are several key parameters that can be adjusted to optimize the reconstruction effect and calculation efficiency. The following are some main parameters and their impacts on the reconstruction results: 1. Set the search radius, which defines the search radius used when finding the nearest neighbor points. For high-density point clouds, the search radius can be relatively small (e.g., 1 - 2 times the average point spacing of the point cloud); for low-density point clouds, a larger search radius is required (possibly several times the average point spacing of the point cloud), and it is generally set to 0.05.

[0027] 2. The maximum edge length factor (mu), which defines the proportional relationship between the maximum edge length and the search radius. A smaller mu value will result in more small triangles being generated, while a larger mu value will generate larger triangles. Generally, it is between 1 and several times, and commonly it is between 1 and 2 or 3.

[0028] 3. The maximum number of nearest neighbor points (num), which defines the maximum number of nearest neighbor points to be considered within the search radius. Increasing this value can improve the stability of the algorithm but also increase the calculation time. The common range may be from 10 to several hundred.

[0029] 4. The maximum surface angle (surfaceAngle), which defines the maximum allowable angle of the triangulated surface. A smaller angle value will generate a smoother surface, while a larger angle value will generate a rougher surface. Usually, it is between 0 degrees and 90 degrees, but it may also be adjusted as needed.

[0030] 5. The minimum angle (angleMin), which defines the minimum allowable angle of the triangle. A smaller angle value can generate more detailed details but may also result in unstable triangles. Usually, it is greater than 0 degrees, and commonly it is between 10 degrees and 30 degrees.

[0031] 6. The maximum angle (angleMax), which defines the maximum allowable angle of the triangle. A larger angle value can generate a flatter surface but may also cause distortion of the model. Usually, it is less than 180 degrees, and commonly it is between 120 degrees and 150 degrees.

[0032] 7. Normal vector consistency (consistent), which defines whether to consider the consistency of the normal vector when calculating the normal vector. Enabling normal vector consistency can improve the quality of triangulation.

[0033] Taking the display of the model based on digital twin software such as Unity as an example, the specific steps for 3D reconstruction when describing the application of the reconstructed model are as follows: Step 1. Package the developed C++ code into a DLL (Dynamic Link Library), and output the calculated and optimized mesh information (normal vector, triangle patch index); Step 2. Call the above DLL in Unity to implement model display and saving according to the triangular mesh information; Step 3: Process the model according to actual production requirements, such as cutting simulation, volume calculation, and dynamic programming, etc.

[0034] The present invention provides a dynamic three-dimensional reconstruction method for point clouds based on greedy projection triangle reconstruction in the mining field. It uses three-dimensional laser scanning technology to obtain a set of three-dimensional coordinates of a large number of points on the object surface as point cloud data, generates a triangular mesh by iteratively connecting adjacent points in the point cloud, and finally constructs a three-dimensional model of the entire roadway surface. This modeling method is suitable for a variety of model application scenarios and can adjust model parameters according to scene requirements. At the same time, it can dynamically adjust the model according to the update of point cloud data, and greatly improve the modeling efficiency through an optimization algorithm. Moreover, the present invention conducts efficient and complete three-dimensional modeling based on the collected and processed point cloud data, has a high degree of authenticity for modeling entities such as coal walls without drawings, with numerous details and dynamic changes, and can fully meet the real-time modeling and planning cutting and other usage requirements of the roadway working face, meeting the development requirements of digital twin technology for the digital transformation of the coal mining industry, and having great market development prospects.

Claims

1. A dynamic three-dimensional reconstruction method for point clouds based on greedy projection triangle reconstruction, characterized in that: It includes the following steps: S1. Obtain the three-dimensional laser point cloud data in the driving roadway through laser scanning technology; S2. Load the obtained three-dimensional laser point cloud data and remove the three-dimensional laser point cloud data with a quantity of zero or incorrect format; S3. Perform preprocessing operations on the three-dimensional laser point cloud data; S4. Use the greedy projection triangulation reconstruction algorithm to perform meshing on the preprocessed three-dimensional laser point cloud data, restore the spatial topological structure between points in the three-dimensional laser point cloud data, and use optimized parameters for greedy projection triangulation reconstruction to generate triangular mesh data; S5. Implement the display and saving of the roadway three-dimensional model according to the triangular mesh data.

2. The method for dynamic three-dimensional reconstruction of point cloud based on greedy projection triangle reconstruction according to claim 1, characterized in that: In step S1, set the left direction facing the coal wall as the positive direction of the y-axis, the advancing direction of the working face as the positive direction of the x-axis, and the upward direction perpendicular to the advancing direction of the working face as the positive direction of the z-axis, and establish a three-dimensional coordinate system; let the laser scanner start from the origin of the three-dimensional coordinate system and move along the positive direction of the y-axis to obtain the three-dimensional laser point cloud data in the driving roadway.

3. The point cloud dynamic three-dimensional reconstruction method based on greedy projection triangle reconstruction according to claim 2, characterized in that: In step S3, the preprocessing operations on the three-dimensional laser point cloud data include the following steps: S31. Cut off the point cloud data outside the reconstruction range in the three-dimensional laser point cloud data; S32. Filter and remove the duplicate point cloud data in the three-dimensional laser point cloud data and fill in the missing point cloud data; S33. Calculate the normal vector of each point in the three-dimensional laser point cloud data; S34. Establish a search tree for quickly finding the position information of points; S35. Perform feature extraction on the three-dimensional laser point cloud data; S36. Calculate the matching points of several partial point cloud data sets in the three-dimensional laser point cloud data through ICP, and merge the several partial point cloud data sets in the three-dimensional coordinate system to generate the three-dimensional laser point cloud data of a complete scene; S37. Perform density increase or density decrease processing on the point cloud data in the three-dimensional laser point cloud data to meet the requirements of subsequent operations.

4. The point cloud dynamic three-dimensional reconstruction method based on greedy projection triangle reconstruction according to claim 3, characterized in that: Step S4 specifically includes the following steps: S41. Set optimized parameters and find the nearest neighbor points of each point in the three-dimensional laser point cloud data according to the search radius; S42. Construct a local triangulated surface based on the searched nearest neighbor points; S43. Iteratively add new points and update the triangulated surface; S44. Generate complete triangular mesh data.

5. The method for dynamic three-dimensional reconstruction of point cloud based on greedy projection triangle reconstruction according to claim 4, wherein: In step S41, the optimized parameters include the search radius of neighboring points, the ratio between the maximum side length and the search radius, the maximum number of neighboring points within the search radius, the maximum surface angle of the triangulated surface, the minimum allowable angle of the formed triangles, the maximum allowable angle of the formed triangles, and the normal vector consistency.

Images