A method for denoising 3D point cloud data of tunnel surrounding rock

Abstract

The present invention relates to a method for denoising three-dimensional point cloud data of tunnel surrounding rock, comprising the following steps: obtaining an initial point cloud of the tunnel surrounding rock, calculating the average distance between each point in the initial point cloud and its adjacent points, denoising the initial point cloud based on the average distance to obtain a denoised point cloud; clustering the denoised point cloud to obtain the number of points in each cluster, using the cluster with the largest number of points as the main face of the tunnel face, eliminating other clusters, and obtaining a maximum point cloud block based on the cluster with the largest number of points; extracting the edge point set of the largest point cloud block, calculating the normal vector of the plane on which the edge point set lies, obtaining the angle between the normal vector and the normal vector of the XOZ plane, and rotating the largest point cloud block according to the angle; projecting the current largest point cloud block onto the XOZ plane to obtain a convex hull point cloud within the XOZ plane; and reducing the convex hull point cloud by a set ratio to obtain a final point cloud. Compared with existing technologies, the present invention has the advantages of high noise removal efficiency and more complete preservation of the data characteristics of the original point cloud.

Description

Technical Field

[0001] The present invention relates to the field of tunnel face surrounding rock information collection, and in particular to a method for reducing noise in three-dimensional point cloud data of tunnel surrounding rock. Background Art

[0002] In recent years, the technology of constructing 3D point cloud models of surrounding rock based on multi-camera imaging and 3D laser scanning has been widely used in the collection of surrounding rock information at tunnel faces to identify surrounding rock geometric features, assist in determining surrounding rock grade, and adjust construction measures.

[0003] The three-dimensional point cloud model of the tunnel surrounding rock constructed by the above technology includes two parts: the tunnel face area and the adjacent tunnel wall area. When used for surrounding rock information identification and processing, the point cloud of the tunnel wall area needs to be eliminated and only the point cloud of the tunnel face area is retained.

[0004] There are currently three main methods for removing cave wall point clouds:

[0005] Method 1:

[0006] With the help of dedicated point cloud data processing software such as ARCGIS, CloudCompare, etc., the cave wall point cloud is removed through manual processing, such as the technology disclosed in CN115358940A and CN113222416A. This method cannot achieve fast and automatic processing.

[0007] Method 2:

[0008] By limiting the collection area, only the point cloud of the tunnel face is collected, such as the technology disclosed in CN113689394A. This method can only be used to collect point clouds by 3D laser scanning, which will greatly increase the difficulty and time of on-site collection. Due to the influence of the shooting environment and space, it is impossible to collect point clouds only from the tunnel face when using digital photography and multi-lens imaging technology to collect point clouds.

[0009] Method 3:

[0010] With the help of relevant algorithms, the tunnel wall area is automatically identified and the point cloud in this area is eliminated. For example, CN115527041A discloses a tunnel contour extraction method, which includes: obtaining point cloud data and performing ground feature separation on the point cloud data to obtain data to be processed; using the DBSCAN clustering method to identify and filter out the first type of noise points in the data to be processed, where the distance between the first type of noise points and the main tunnel point cloud is greater than or equal to a preset value; and using a method combining least squares fitting and RANSAC-based circle fitting to identify and filter out the second type of noise points in the data to be processed to obtain the tunnel contour, where the distance between the second type of noise points and the main tunnel point cloud is less than the preset value. However, the retrieval and elimination of tunnel wall point clouds based on this solution generally takes more than 20 minutes, and the point cloud elimination effect in the area adjacent to the tunnel face and the tunnel wall is not ideal, especially the point clouds of the arch foot and invert arch are not effectively eliminated.

[0011] Obviously, manually removing point clouds (method one) or limiting the point cloud collection area on site (method two) are inefficient and difficult to operate, and cannot achieve rapid and automatic removal of noise data.

[0012] The automated removal of cave wall point clouds using method 3 faces the following problems:

[0013] (1) This method distinguishes and eliminates the tunnel face area and the tunnel wall area based on the difference in the local normal vector features of the point cloud. It is suitable for eliminating the tunnel wall point cloud in the area far away from the tunnel face, but cannot eliminate the noise points in the adjacent area between the tunnel face and the tunnel wall. Especially when the blasting quality is poor, the tunnel face flatness is poor, or the core soil excavation method is used, the point cloud of the tunnel face area will be mistakenly eliminated on a large scale, affecting the subsequent recognition results.

[0014] (2) This method uses the moving least squares method to perform multiple iterative analyses of the global scope of the point cloud model. The computational complexity is large, resulting in an analysis time of more than 20 minutes, making it difficult to provide timely feedback on the surrounding rock identification analysis results.

[0015] (3) This method specifies the default normal direction of the tunnel face in the point cloud model, and does not consider the rotation and deflection of the point cloud model caused by multi-camera imaging or laser scanning, which reduces the accuracy of subsequent point cloud feature recognition results.

[0016] Therefore, how to quickly and accurately remove noise information from the point cloud model and obtain more accurate and complete tunnel face point cloud data is a technical problem that needs to be solved urgently in this field. Summary of the Invention

[0017] The purpose of the present invention is to overcome the defects of the above-mentioned existing technologies and provide a method for denoising three-dimensional point cloud data of tunnel surrounding rock, so as to solve the technical problem of quickly eliminating noise points on the wall of the three-dimensional point cloud model.

[0018] The purpose of the present invention can be achieved by the following technical solutions:

[0019] A method for reducing noise in three-dimensional point cloud data of tunnel surrounding rock, comprising the following steps:

[0020] 1) Obtaining an initial point cloud of the tunnel surrounding rock, calculating the average distance between each point in the initial point cloud and its neighboring points, and performing denoising on the initial point cloud based on the average distance to obtain a denoised point cloud, wherein the denoising specifically includes removing points with an average distance outside a set range as noise points;

[0021] 2) clustering the denoised point cloud to obtain the number of points in each cluster, taking the cluster with the largest number of points as the main face of the tunnel face, eliminating other clusters, and obtaining the largest point cloud block based on the cluster with the largest number of points;

[0022] 3) extracting the edge point set of the largest point cloud block, calculating the normal vector of the plane where the edge point set is located, obtaining the angle between the normal vector and the normal vector of the XOZ plane, and rotating the largest point cloud block according to the angle;

[0023] 4) Project the current largest point cloud block to the XOZ plane to obtain the convex hull point cloud in the XOZ plane;

[0024] 5) Reduce the convex hull point cloud by a set ratio to obtain the final point cloud.

[0025] Furthermore, the initial point cloud of the tunnel surrounding rock is obtained by using multi-camera photography technology or three-dimensional laser scanning technology.

[0026] Furthermore, the setting range is obtained based on the average distance of all points in the initial point cloud, and the setting range is (E0-xσ0, E0+xσ0), wherein E0 and σ0 are the expectation and variance of all average distances respectively, and x is the offset coefficient.

[0027] Furthermore, the clustering is achieved based on the Euclidean distance growing algorithm.

[0028] Furthermore, the clustering process specifically includes:

[0029] 201) Marking all points in the denoised point cloud as unclassified, recorded as point set α;

[0030] 202) Pick any point P from the point set α i , put into the empty point set Ω and mark it as classified;

[0031] 203) Judgment point P i Is there a neighboring point within the growth distance r? If so, put the neighboring point into the point set γ to be classified and go to step 204), otherwise go to step 205;

[0032] 204) Select any point Q from the set of points to be classified γ j Put it into the point set Ω, mark it as classified, and repeat step 203);

[0033] 205) Treat the points in the point set Ω as a cluster and determine whether the number of points in the cluster is greater than the minimum cluster size S. If so, retain the cluster; otherwise, abandon the cluster.

[0034] 206) Return to step 203) until there are no points in the unclassified state.

[0035] Furthermore, the value of the growth distance r is determined based on the collected tunnel excavation location.

[0036] Furthermore, the edge point set is obtained by the following steps:

[0037] 301) Take any point N in the largest point cloud block i , project all points in its neighborhood R onto its tangent plane;

[0038] 302) N i is an endpoint, and is connected with the projected neighboring points to form vectors. The angles between two adjacent vectors are calculated in counterclockwise or clockwise order to form a set Λ;

[0039] 303) Get the maximum angle Λ among all angles max =max(Λ), add the points whose maximum angle is greater than or equal to the angle threshold ξ to the edge point set.

[0040] Furthermore, the least square method is used to calculate the normal vector of the plane where the edge point set is located.

[0041] Furthermore, the reducing the convex hull point cloud according to the set ratio is specifically: setting a scale factor fSc, and reducing the convex hull point cloud toward the center by the scale factor fSc.

[0042] Furthermore, the proportional coefficient fSc is determined based on the collected tunnel excavation location.

[0043] Compared with the existing technology, the present invention uses the three-dimensional spatial features of the tunnel face and tunnel wall point clouds to perform global recognition of the tunnel face point cloud and automatically remove noise point clouds on the tunnel wall. This greatly improves the efficiency of noise removal and achieves rapid and automated removal of tunnel wall point clouds, with the following beneficial effects:

[0044] (1) The present invention completely eliminates the point cloud of the tunnel wall through point cloud rotation and convex hull contraction, thus achieving fully automated extraction of the tunnel face point cloud;

[0045] (2) The present invention first removes noise points from the initial point cloud and further determines the main face of the tunnel face through clustering, avoiding the least squares fitting iterative calculation of the global point cloud, which greatly reduces the time consumed by automatic noise reduction;

[0046] (3) The present invention uses a method of global identification of the tunnel face to process point cloud data, avoiding the situation where similar technologies will eliminate useful point cloud data, more completely retaining the data features of the original point cloud, and providing more complete materials for subsequent point cloud analysis and calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the process of the present invention;

[0048] Figure 2 is the initial point cloud image in the embodiment of the present invention;

[0049] Figure 3 Schematic diagram of the clustering process in an embodiment of the present invention;

[0050] Figure 4 Schematic diagram of edge point recognition principle in an embodiment of the present invention;

[0051] Figure 5 is a convex hull point cloud image in an embodiment of the present invention;

[0052] Figure 6 Schematic diagram of convex hull contraction in an embodiment of the present invention. DETAILED DESCRIPTION

[0053] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0054] Explanation of terms

[0055] Tunnel face: Tunnel face, also known as tunnel face, is a term used in tunnel construction. It refers to the working face that continuously advances during tunnel excavation (coal mining, mining or tunnel engineering).

[0056] Tunnel wall: refers to the general term for the arch, waist and bottom areas in other directions of the tunnel face area except the face.

[0057] Point cloud: refers to a massive collection of points that represent the surface characteristics of a target. In this patent, it refers to the point cloud collection of the tunnel face and tunnel wall area.

[0058] Denoising: The process of removing noise point clouds caused by the influence of camera and other acquisition equipment, surrounding environment, human disturbance, target characteristics, etc.

[0059] Convex hull: In a real vector space V, for a given set X, the intersection S of all convex sets containing X is called the convex hull of X.

[0060] Neighboring points: The set of all points within a certain Euclidean distance from a given point A.

[0061] This embodiment provides a method for denoising 3D point cloud data of tunnel surrounding rock. Based on the 3D spatial characteristics of the tunnel face and tunnel wall point clouds, a series of processing is performed on the collected 3D initial point cloud of the tunnel surrounding rock to obtain a reliable and useful tunnel face area point cloud, thereby solving the problem of fast and automatic removal of tunnel wall point clouds. Figure 1 As shown, the method includes:

[0062] Step S1: Point cloud denoising. An initial point cloud of the tunnel surrounding rock is obtained. The average distance from each point in the initial point cloud to its neighboring points is calculated. Based on the average distance, the initial point cloud is denoised to obtain a denoised point cloud. Specifically, the denoising step includes removing points with an average distance outside a set range as noise points.

[0063] In a specific embodiment, multi-camera photography technology or three-dimensional laser scanning technology can be used to obtain the initial point cloud of the tunnel surrounding rock.

[0064] In this embodiment, the acquisition Figure 2 The initial point cloud shown is used to calculate point A in the point cloud. i (i=1~n) to the adjacent point The average distance Dist i :

[0065]

[0066] Where n is the number of points in the initial point cloud, m is the i-th point A i The number of corresponding neighboring points, (x i ,y i ,z i ), Point A i 、 's coordinates.

[0067] Then, calculate the Dist of all point clouds i The expectation E0 and variance σ0 of:

[0068]

[0069]

[0070] With (E0-xσ0, E0+xσ0) as the setting range, the average distance Dist iPoints outside the set range are treated as noise points and removed, where x is the offset coefficient. The number of neighboring points and the offset coefficient value need to be determined based on the point cloud removal effect. Generally, 30 and 0.1 are suitable values, respectively.

[0071] Step S2: Extract the largest point cloud. Cluster the denoised point cloud to obtain the number of points in each cluster. The cluster with the largest number of points is used as the main face of the tunnel face. Other clusters are eliminated, and the largest point cloud block is obtained based on the cluster with the largest number of points.

[0072] In a specific embodiment, the clustering can be implemented based on the Euclidean distance growing algorithm. Figure 3 As shown in Figure 2, the clustering process specifically includes:

[0073] (1) Mark all points as unclassified, denoted as point set α;

[0074] (2) Pick any point P from the point set α i , put into the empty point set Ω and mark it as classified;

[0075] (3) Judgment point P i Is there a neighboring point within the growth distance r? If so, put the neighboring point into the set of points to be classified γ and go to step (4). Otherwise, go to step (5).

[0076] (4) Select any point Q from the set of points to be classified γ j Put it into the point set Ω, mark it as classified, based on point Q j Repeat step (3). If there are still unclassified points in the set of points to be classified γ, the remaining points are regarded as noise points and are not considered.

[0077] (5) Treat the points in the point set Ω as a cluster and determine whether the number of points in the cluster is greater than the minimum size S of the cluster. If it is less, the cluster is discarded; if it is greater, the cluster is retained.

[0078] (6) Take another point from the point set α and return to step (3) until there are no unclassified points in the point cloud. The point with the largest number of final point clouds is output as the result point cloud.

[0079] The values ​​of the minimum cluster size S and the growth distance r are determined according to the tunnel excavation location to be collected. Preferably, the minimum cluster size S is 1 / k of the number of point clouds obtained in the previous step (i.e., the number of points in the denoised point cloud). Generally, k can be 10. The value of the growth distance r is related to the tunnel excavation location. In this embodiment, the full-section collection is performed, and the r value is 2E0; when collecting the upper steps of the tunnel, the r value is 1.5E0; when collecting the left and right pilot tunnels, the r value is 1.2E0.

[0080] Step S3: Determine the point cloud orientation. Extract the outermost edge points of the largest point cloud block to form an edge point set. Calculate the normal vector of the plane containing the edge point set, obtain the angle between the normal vector and the XOZ plane normal vector, and determine whether the point cloud needs to be rotated. If so, rotate the largest point cloud block according to the angle.

[0081] The specific process of step S3 includes:

[0082] (1) Take any point N in the largest point cloud block i , project all m points in its neighborhood R onto its tangent plane.

[0083] (2) N i As an endpoint, connect the projected neighboring points to form a vector, calculate the angle between two adjacent vectors in counterclockwise or clockwise order and form a set Λ={β1,β2,…,β n}, n = m-1, where:

[0084]

[0085] (x1, y1) and (x2, y2) are two N i are the adjacent vector coordinates of the endpoint.

[0086] (3) Traverse the largest angle Λ among all included angles max =max(Λ).

[0087] (4) By Figure 4 It can be seen that the maximum angle of internal points and edge points is different. If Λ max The larger the angle is, the more likely the point is an edge point. An angle threshold ξ is set to judge the edge points and obtain the edge point set O. Preferably, the angle threshold ξ is not less than 130°.

[0088] (5) Using the least squares method, calculate the normal vector ν of the plane where the edge point set is located i .

[0089] The plane normal vector ν i The calculation formula is:

[0090]

[0091] Among them, (x i ,y i ,z i ) is the coordinate of point O, (a0, b0, c0) is the normal vector ν i .

[0092] (6) If ν iIf the angle θ with the XOZ plane normal vector (0,1,0) is not 0, the point cloud is rotated according to the angle θ to obtain the final point cloud.

[0093] Step S4: Point cloud projection and convex hull calculation: Project the current largest point cloud block onto the XOZ plane to obtain the convex hull point cloud in the XOZ plane.

[0094] In this embodiment, the convex hull point (x, y, z) of the point cloud block in the XOZ plane and the centroid (x c ,y c ,z c ), the calculated convex hull is as follows Figure 5 shown.

[0095] Step S5: Extract the point cloud of the tunnel face. Reduce the convex hull point cloud by a set ratio, extract the point set in the reduced convex hull as the point cloud to be analyzed, and obtain the final point cloud, such as Figure 6 shown.

[0096] The specific method of reducing the convex hull point cloud by a set ratio is as follows: setting a scale factor fSc, reducing the convex hull point cloud toward the center by the scale factor fSc, and the calculation formula of the convex hull point after reduction is:

[0097]

[0098] The scaling factor fSc is determined based on the tunnel excavation location being sampled. The x, y, and z columns on the left side of the equation represent the scaled-down point, while the columns on the right side represent the original point. Preferably, fSc is 0.15 when sampling the entire cross-section; 0.1 when sampling only the upper step; and 0.05 when sampling the left and right pilot tunnels.

[0099] If the above method is implemented in the form of a software functional unit and 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 the present invention, or the part that contributes to the prior art, or the 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 for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0100] The above describes in detail the preferred embodiments of the present invention. It should be understood that those skilled in the art can make numerous modifications and variations based on the concepts of the present invention without inventive effort. Therefore, any technical solutions that can be derived by those skilled in the art through logical analysis, reasoning, or limited experimentation based on the concepts of the present invention and the prior art should be within the scope of protection defined by the claims.

Claims

1. A method for denoising three-dimensional point cloud data of tunnel surrounding rock, characterized in that: The following steps are involved: 1) Obtaining an initial point cloud of the tunnel surrounding rock, calculating the average distance between each point in the initial point cloud and its neighboring points, and performing denoising on the initial point cloud based on the average distance to obtain a denoised point cloud, wherein the denoising specifically includes removing points with an average distance outside a set range as noise points; 2) clustering the denoised point cloud to obtain the number of points in each cluster, taking the cluster with the largest number of points as the main face of the tunnel face, eliminating other clusters, and obtaining the largest point cloud block based on the cluster with the largest number of points; 3) extracting the edge point set of the largest point cloud block, calculating the normal vector of the plane where the edge point set is located, obtaining the angle between the normal vector and the normal vector of the XOZ plane, and rotating the largest point cloud block according to the angle; 4) Project the current largest point cloud block to the XOZ plane to obtain the convex hull point cloud in the XOZ plane; 5) Reduce the convex hull point cloud by a set ratio to obtain the final point cloud.

2. The method for denoising tunnel surrounding rock 3D point cloud data according to claim 1, characterized in that: The initial point cloud of the tunnel surrounding rock is obtained by using multi-camera photography technology or three-dimensional laser scanning technology.

3. The method for denoising 3D point cloud data of tunnel surrounding rock according to claim 1, characterized in that: The setting range is obtained based on the average distance of all points in the initial point cloud, and the setting range is (E0-xσ0, E0+xσ0), where E0 and σ0 are the expectation and variance of all average distances respectively, and x is the offset coefficient.

4. The method for denoising tunnel surrounding rock 3D point cloud data according to claim 1, characterized in that: The clustering is achieved based on the Euclidean distance growing algorithm.

5. The method for denoising tunnel surrounding rock 3D point cloud data according to claim 4, characterized in that: The clustering process specifically includes: 201) Marking all points in the denoised point cloud as unclassified, recorded as point set α; 202) Pick any point P from the point set α i , put into the empty point set Ω and mark it as classified; 203) Judgment point P i Is there a neighboring point within the growth distance r? If so, put the neighboring point into the point set γ to be classified and go to step 204), otherwise go to step 205; 204) Select any point Q from the set of points to be classified γ j Put it into the point set Ω, mark it as classified, and repeat step 203); 205) Treat the points in the point set Ω as a cluster and determine whether the number of points in the cluster is greater than the minimum cluster size S. If so, retain the cluster; otherwise, abandon the cluster. 206) Return to step 203) until there are no points in the unclassified state.

6. The method for denoising 3D point cloud data of tunnel surrounding rock according to claim 5, characterized in that: The value of the growth distance r is determined based on the collected tunnel excavation location.

7. The method for denoising tunnel surrounding rock 3D point cloud data according to claim 1, characterized in that: The edge point set is obtained by the following steps: 301) Take any point N in the largest point cloud block i , project all points in its neighborhood R onto its tangent plane; 302) N i is an endpoint, and is connected with the projected neighboring points to form vectors. The angles between two adjacent vectors are calculated in counterclockwise or clockwise order to form a set Λ; 303) Get the maximum angle Λ among all angles max =max(Λ), add the points whose maximum angle is greater than or equal to the angle threshold ξ to the edge point set.

8. The method for denoising tunnel surrounding rock 3D point cloud data according to claim 1, characterized in that: The normal vector of the plane where the edge point set is located is calculated using the least squares method.

9. The method for denoising tunnel surrounding rock 3D point cloud data according to claim 1, characterized in that: The step of reducing the convex hull point cloud according to a set ratio specifically includes setting a scale factor fSc and reducing the convex hull point cloud toward the center by the scale factor fSc.

10. The method for denoising tunnel surrounding rock 3D point cloud data according to claim 9, characterized in that: The proportional coefficient fSc is determined based on the collected tunnel excavation location.

Images