Method for dynamically, finely and quickly sensing multi-source data of tunnel surrounding rock

Abstract

The invention discloses a tunnel surrounding rock multi-source data dynamic fine rapid sensing method, which comprises the following steps: based on a mobile terminal multi-view image sequence, arranging shooting positions according to a preset space interval and an orthogonal angle, generating a sparse point cloud through a motion recovery structure algorithm, and generating a dense point cloud model in combination with multi-view stereo matching optimization; and for the dense point cloud model, delimiting a local neighborhood based on k-nearest neighbor search, resolving a neighborhood point covariance matrix through principal component analysis, extracting a feature vector corresponding to a minimum feature value as a normal vector, and constructing a rock mass surface microscopic geometric feature field. The invention provides a dynamic fine rapid sensing method based on multi-source data, and aims to improve the efficiency, precision and timeliness of tunnel surrounding rock information acquisition and provide accurate surrounding rock information support for tunnel construction and support design.

Description

Technical Field

[0001] The present application relates to the technical field of surrounding rock information collection, and in particular to a method for dynamic, precise and rapid perception of multi-source data of tunnel surrounding rock. Background Art

[0002] Tunnel surrounding rock information is an important basis for tunnel construction and subsequent engineering design. Its accuracy and timeliness directly affect the construction safety of tunnel engineering and the formulation of support plans. Traditional surrounding rock information collection technology has many shortcomings, such as low collection accuracy, inability to achieve real-time dynamic feedback, and high reliance on manual labor, which make the collection process of surrounding rock information face great challenges.

[0003] At present, most tunnel surrounding rock information collection relies on manual observation or conventional instrument detection, which cannot carry out detailed real-time monitoring of the detailed characteristics of the tunnel face. With the continuous development of tunnel construction technology, traditional technology has been unable to meet the high requirements of modern tunnel construction for surrounding rock information. Therefore, how to achieve fast, accurate, real-time and efficient tunnel surrounding rock information collection has become a key issue that needs to be solved in current tunnel construction technology. Summary of the invention

[0004] In order to solve the above problems, an embodiment of the present invention provides a method for dynamic, precise and rapid perception of tunnel surrounding rock multi-source data, the method comprising:

[0005] Based on the multi-view image sequence of the mobile terminal, the shooting positions are arranged according to the preset spatial intervals and orthogonal angles, and the sparse point cloud is generated by the motion recovery structure algorithm. The dense point cloud model is generated by combining the multi-view stereo matching optimization.

[0006] For the dense point cloud model, the local neighborhood is delineated based on the k-nearest neighbor search, the neighborhood point covariance matrix is ​​solved by principal component analysis, the eigenvector corresponding to the minimum eigenvalue is extracted as the normal vector, and the microscopic geometric characteristic field of the rock surface is constructed;

[0007] Based on the microscopic geometric characteristic field of the rock mass surface, a hierarchical clustering algorithm is used to aggregate point clouds according to the normal vector direction consistency criterion to generate structural surface point clouds;

[0008] For the structural surface point cloud, the point cloud clusters with staggered spatial distribution are separated based on the density clustering algorithm, and the independent structural surface units are extracted by combining noise filtering and boundary constraints.

[0009] For the boundary point cloud of the independent structural surface in the independent structural surface unit, the plane equation is iteratively optimized through the random sampling consistency plane fitting algorithm, and the dip and inclination parameters are analyzed by combining the normal vector projection and space geometry, and the mapping rules are used to obtain the occurrence parameters.

[0010] Based on the boundary point cloud of the independent structural plane, the weighted direction vector is optimized by principal component analysis. Through the trace direction constraint and displacement orthogonal correction, the boundary points are iteratively shrunk to the central axis of the trace to obtain the trace distribution.

[0011] Based on the attitude parameters and trace distribution, the joint spacing is statistically calculated by the virtual survey line method. Combining the calculation of the surface gradient modulus of the structural plane and the roughness coefficient mapping model, the surrounding rock classification index is output.

[0012] Further, the calculation method of the normal vector includes:

[0013] ;

[0014] In the formula, is the th neighboring point of is the default value. Let > > be the covariance matrix 's eigenvalue, , be the corresponding eigenvectors. Then the normal vector corresponding to the point is the eigenvector corresponding to the minimum eigenvalue .

[0015] Further, the hierarchical clustering algorithm is the BIRCH clustering algorithm improved based on cosine similarity; based on the BIRCH clustering algorithm improved by cosine similarity, when inserting new data points, cosine similarity is used to determine the closest cluster.

[0016] Further, the extraction of the independent structural plane unit includes using the DBSCAN density clustering algorithm to realize the segmentation of the independent structural plane.

[0017] Further, the calculation method of the attitude parameters includes:

[0018] For the point cloud that constitutes a structural plane, randomly select three points to calculate their plane fitting equation AX + BY + CZ + D = 0;

[0019] Calculate the vertical distance from each point in the point cloud of each structural plane to the fitting plane of the structural plane;

[0020] ;

[0021] Set the distance threshold trsc, and regard the points with > trsc as outliers, and regard the points with < trsc as inliers, and count the number of inliers Nin;

[0022] Iterate Nrsc times and select the fitting plane with the largest number of inlier points;

[0023] Perform plane fitting on the fitting plane with the largest number of inlier points to obtain the final plane fitting equation.

[0024] Furthermore, the calculation method of the plane fitting equation includes:

[0025] In the plane fitting equation The angle with the normal vector of the XOY plane Is the dip angle of the structural plane , and the numerical range is 0° to 90°, which is defined as:

[0026] ;

[0027] Let Be the projection of the normal vector On the XOY plane, Be the angle between the positive X-axis and the projection , then it is expressed as:

[0028] ;

[0029] The trend of the structural plane is the angle between the positive X-axis in the clockwise direction and the trend line , and the numerical range is 0° to 360°; let ∈[0°, 90°].

[0030] Furthermore, the identification of the trace distribution uses a PCA-weighted iterative algorithm for trace orientation contraction, and the identification method includes:

[0031] Let the set of boundary point coordinates be , and re-search for the Points within Of the Neighboring points, and calculate the direction vector of each point;

[0032] For each point Among them, the calculated eigenvalues are > > , then its contraction weight Is defined as including:

[0033] ;

[0034] In the formula, Is the first eigenvalue obtained by PCA calculation for the Th boundary point; Is the Th boundary point obtained by PCA calculation for the Eigenvalues.

[0035] Let the point 's The indices of the nearest neighbor points in are Then the point The shrunk point ' is defined as including:

[0036] ;

[0037] For 's displacement direction is corrected so that the shrunk points are evenly distributed along the trace length direction, canceling the parallel displacement relative to the trace center axis, so that Finally, there is a vertical movement relative to the trace center axis instead of a parallel movement.

[0038] Furthermore, the spacing identification in the surrounding rock classification index adopts a method based on the two-dimensional virtual survey line method for spacing identification.

[0039] Furthermore, the roughness calculation in the surrounding rock classification index uses the joint roughness coefficient value of the closest standard contour line as the appropriate estimated value; the method for estimating the two-dimensional curve JRC based on the root mean square includes:

[0040] ;

[0041] In the formula, represents the average gradient modulus of the structural plane surface, defined as:

[0042] ;

[0043] In the formula, is the number of points on the two-dimensional section intercepted along one direction, and , respectively, are the vertical coordinates of the points in the three-dimensional data of the structural plane; is the distance between points on each profile line.

[0044] The technical effects and advantages of the dynamic fine, fast and accurate perception method for multi-source data of tunnel surrounding rock provided by the present invention:

[0045] Through the mobile device photography and 3D reconstruction technology, the present invention realizes the rapid acquisition and accurate reconstruction of the detailed information of the tunnel face; based on point cloud processing and intelligent algorithms, it can automatically identify and classify the surrounding rock structural planes, reduce manual intervention, and ensure the accuracy of data analysis; through the integration of data models and dynamic feedback, construction personnel can obtain the surrounding rock information in real time, identify potential problems in the construction process in advance, which helps to optimize the construction strategy; with the support of accurate surrounding rock data, the tunnel support design is optimized, potential risks during the construction process are reduced, and the construction efficiency is improved. Description of the Drawings

[0046] Figure 1 It is a flowchart of the dynamic fine and rapid perception method for multi-source data of tunnel surrounding rock in the embodiment of the present application;

[0047] Figure 2 It is a schematic diagram of the technical solution of the dynamic fine and rapid perception of multi-source data of tunnel surrounding rock in the embodiment of the present application;

[0048] Figure 3 It is a flowchart of the fully automatic extraction method for the information of rock mass discontinuity planes in the embodiment of the present application;

[0049] Figure 4 It is a schematic diagram of the face photography of the mobile terminal in the embodiment of the present application;

[0050] Figure 5 It is a schematic diagram of the face photography at the construction site in the embodiment of the present application;

[0051] Figure 6 It is a schematic diagram of the extraction of face point cloud data in the embodiment of the present application;

[0052] Figure 7 It is the calculation result of the spacing of rock mass structural planes in the embodiment of the present application;

[0053] Figure 8 It is a schematic diagram of the trace recognition and contraction in the embodiment of the present application;

[0054] Figure 9 It is a schematic diagram of the attitude information of the structural plane in the embodiment of the present application. Detailed Embodiments

[0055] 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 creative efforts shall fall within the protection scope of the present invention.

[0056] Please refer to Figure 1As shown in the figure, an embodiment of the present invention provides a method for dynamically, finely and rapidly sensing multi-source data of tunnel surrounding rock, and the method includes:

[0057] S1: Based on the multi-view image sequences of the mobile terminal, set up shooting positions at preset spatial intervals and orthogonal angles, generate sparse point clouds through the Structure from Motion (SFM) algorithm, and optimize and generate a dense point cloud model by combining Multi-View Stereo (MVS) matching;

[0058] S2: For the dense point cloud model, delimit a local neighborhood based on k-nearest neighbor search, solve the covariance matrix of neighborhood points through Principal Component Analysis (PCA), extract the eigenvector corresponding to the minimum eigenvalue as the normal vector, and construct the microscopic geometric feature field of the rock mass surface;

[0059] S3: Based on the microscopic geometric feature field (normal vector field) of the rock mass surface, adopt the hierarchical clustering algorithm, aggregate the point clouds according to the consistency criterion of the normal vector direction, and generate the structural plane point clouds;

[0060] S4: For the structural plane point clouds, separate the point cloud clusters with interleaved spatial distributions based on the density clustering algorithm, and extract independent structural plane units by combining noise filtering and boundary constraints;

[0061] S5: For the independent structural plane boundary point clouds in the independent structural plane units, iteratively optimize the plane equation through the Random Sample Consensus (RANSAC) plane fitting algorithm, and map and analyze the dip and dip angle parameters by combining normal vector projection and spatial geometry to obtain the attitude parameters;

[0062] S6: For the independent structural plane boundary point clouds, optimize the weighted direction vectors by using Principal Component Analysis (PCA), and iteratively shrink the boundary points to the trace center axis through trace direction constraint and displacement orthogonal correction to obtain the trace distribution;

[0063] S7: Based on the attitude parameters and trace distribution, statistically calculate the joint spacing through the virtual survey line method, and output the surrounding rock classification index by combining the calculation of the surface gradient modulus of the structural plane and the roughness coefficient mapping model.

[0064] It should be specifically noted that in this solution, the letters and are used as two general index variables. and When they appear in different formulas or contexts, they only represent an integer serial number (for example, the th element, the th nearest neighbor point and the th sample point, etc.). Their specific meanings and value ranges will be clearly defined each time they are used. The letters and themselves do not carry fixed technical meanings.

[0065] The calculation method of the normal vector includes:

[0066] ;

[0067] In the formula, is the th nearest neighbor point of ( <15) will cause a large amount of noise in the generated normal vector, while a large ( >30) will make the local curvature too smooth. Therefore, = 20 is selected as the default value. Let > > be the eigenvalues of the covariance matrix , , and be the corresponding eigenvectors. Then the normal vector corresponding to the point is the eigenvector corresponding to the minimum eigenvalue .

[0068] After constructing the microscopic geometric feature field of the rock mass surface, it is necessary to cluster the point cloud according to the consistency of the normal vector direction to initially identify potential discontinuity surface areas; to achieve this goal, this method uses an improved hierarchical clustering algorithm, namely the BIRCH algorithm optimized based on cosine similarity; based on the cosine similarity to improve the BIRCH clustering algorithm, when inserting a new data point, the cosine similarity is used to determine the closest cluster.

[0069] The core improvement lies in the similarity measurement method:

[0070] When constructing its core data structure, the CF tree, the traditional BIRCH algorithm usually uses the Euclidean distance as the measurement standard to determine which cluster feature (CF) entry or sub-cluster a new data point should be inserted into; this is effective for spatial position clustering, but not ideal for clustering of normal vector directions (essentially unit direction vectors in three-dimensional space) because the Euclidean distance is sensitive to the absolute position of the vector rather than the consistency of its direction.

[0071] This method improves the key steps of the BIRCH algorithm. When a new normal vector data point needs to be inserted into the CF tree, instead of using the Euclidean distance, the cosine similarity between this new point and the clustering feature vector of the existing CF entry (representing the sub-cluster) (usually the mean of the normal vectors of the points within the sub-cluster) is calculated.

[0072] Cosine similarity measures the degree of proximity between two vectors in terms of direction, with a value range of [-1, 1]. The closer the value is to 1, the more consistent the directions of the two vectors are (the angle is close to 0°); the closer the value is to -1, the more opposite the directions are (the angle is close to 180°); a value of 0 indicates orthogonality, which precisely meets the core requirement of "aggregating point clouds according to the normal vector direction consistency criterion" because for the points on the structural plane, the normal vector directions should be highly consistent (i.e., the cosine similarity is close to 1).

[0073] When the improved algorithm searches for the insertion path in the CF tree, it calculates the cosine similarity between the new point and the feature vectors of each CF entry under the current node; the algorithm selects the CF entry with the largest cosine similarity value (i.e., the closest direction) as the target path; if the cosine similarity between the new point and the closest CF entry is lower than the set tree construction threshold (this threshold controls the clustering granularity), new CF entries may be created or the tree may be rebuilt.

[0074] Example illustration:

[0075] Suppose a sub-cluster currently represented by a CF entry has a mean normal vector of (0.707, 0.0, 0.707) (indicating a roughly 45° tilt direction); there is a newly acquired normal vector point (0.695, 0.01, 0.719).

[0076] ≈0.018 (calculate the Euclidean distance between the two points); this distance needs to be compared with the threshold.

[0077] After improvement (cosine similarity): Calculate the dot product of the two vectors divided by the product of the norms:

[0078] ) ≈ (0.491 + 0 + 0.508) / (1 * 1) ≈ 0.999; this value is very close to 1, clearly indicating that the new point is highly consistent with the normal vector direction of the existing sub-cluster and should be classified into this sub-cluster.

[0079] Through this improvement based on cosine similarity, the BIRCH algorithm can efficiently and accurately preliminarily aggregate point clouds into different potential structural plane regions (i.e., the prototype of "structural plane point clouds") according to the normal vector direction. These preliminarily aggregated point cloud clusters with high direction consistency.

[0080] The extraction of independent structural plane units includes using the DBSCAN density clustering algorithm to achieve the segmentation of independent structural planes.

[0081] After initially aggregating the structural point cloud regions with similar normal vector directions through an improved hierarchical clustering algorithm (BIRCH algorithm based on cosine similarity), these regions may contain multiple potential structural point cloud clusters that are spatially intertwined and have unclear boundaries, and are mixed with measurement noise or non-structural point clouds (such as boulders and local irregularities); to accurately segment out the point cloud units representing a single independent structural plane (i.e., "independent structural plane units") and perform boundary extraction, this method uses the DBSCAN density clustering algorithm as the core segmentation tool.

[0082] DBSCAN performs clustering based on "density reachability"; its core assumptions include: a cluster (representing an independent structural plane) consists of density-connected points and is separated by low-density regions (representing the rock mass or voids between structural planes); this highly conforms to the characteristics that structural planes in tunnel surrounding rock are usually distributed as relatively continuous planes or curved surfaces, and there are spatial intervals between different structural planes.

[0083] Core radius (ε): Defines the neighborhood range of a point; points within this radius sphere are considered the "neighbors" of this point.

[0084] Minimum number of points (MinPts): Defines the minimum number of neighbors required for a "core point"; if the ε-neighborhood of a point contains at least MinPts points (including itself), then this point is considered a core point.

[0085] The DBSCAN running method includes:

[0086] For each point in the structural point cloud output by hierarchical clustering, calculate the number of neighbors within its ε radius.

[0087] The core point identification method includes: If the number of neighbors of a point ≥ MinPts, then mark it as a core point.

[0088] Starting from any core point, include all points within its ε neighborhood (including other core points and border points) into the same cluster; recursively include the points within the neighborhoods of these newly added points (if they are also core points) into this cluster. This process aggregates all density-reachable points into one cluster.

[0089] Points that are neither core points nor density-reachable from any core point are marked as noise points.

[0090] When all core points have been visited and the cluster expansion is completed, the algorithm ends, and several independent point cloud clusters (each cluster represents a potential independent structural plane unit) and a set of noise points are output.

[0091] DBSCAN can effectively separate point cloud clusters of different structural surfaces that are spatially adjacent or even partially overlapping but have distinguishable densities. Even if two structural surfaces are very close in space (such as parallel joints with small spacing), as long as there is a sufficient point density gap between them (i.e., the distance between points is greater than ε), DBSCAN can identify them as independent clusters.

[0092] The DBSCAN algorithm automatically identifies and removes isolated points or small point groups (i.e., noise points) that do not meet density requirements; this directly implements "noise filtering" and removes interference points introduced by non-structural surface features or measurement errors.

[0093] In each independent cluster output by DBSCAN, points are divided into core points (cluster internal points) and boundary points (cluster edge points); boundary points naturally constitute the outline of the independent structure surface point cloud cluster; combined with subsequent processing, these boundary point information can be further utilized; at the same time, the algorithm itself implicitly constrains the boundaries of the cluster (i.e., where the density decreases) through the ε and MinPts parameters.

[0094] DBSCAN does not require pre-specification of the number of structural surfaces to be segmented, and can automatically discover independent structural surface units of arbitrary shapes based on the actual spatial density distribution of the point cloud, which is especially important for rock structural surfaces with varied shapes.

[0095] Exemplary:

[0096] Assume that in a certain rock wall area, a set of point clouds is obtained after normal vector clustering, which includes point clouds of two actually parallel but very close (small spacing) joint surfaces G and H, as well as some randomly scattered noise points (such as fallen debris, sensor noise).

[0097] Parameter settings (example values, actual adjustments are required): Set ε=0.05 meters (representing a reasonable threshold for the distance between points on a structural surface), MinPts=10 (representing the minimum number of points that a local area of ​​a valid structural surface should have).

[0098] The operation process includes:

[0099] For a point located in the core area of ​​face G, there are usually far more than 10 points from face G within 0.05 meters around it. These points are identified as core points and expanded to form cluster G.

[0100] Points located in the core area of ​​face H are also identified as core points and expanded to form cluster H. Although faces G and H are close, DBSCAN can distinguish them as long as the minimum distance between points in their boundary area is greater than ε (or the density of points at the boundary is not enough to satisfy the MinPts of both faces at the same time). Points located at the edge of face G or face H (the number of neighbors may be less than MinPts but are located in the neighborhood of the core point) are identified as boundary points and are included in the corresponding cluster.

[0101] Points that are isolated and have far fewer than 10 points within 0.05 meters around them (noise points) will be filtered out; two clear and independent point cloud clusters (cluster G and cluster H) are output, representing the point cloud units of joint planes G and H respectively, and the noise points are removed.

[0102] By applying the DBSCAN density clustering algorithm, this method has successfully achieved the precise segmentation, noise filtering, and boundary point recognition of the preliminarily aggregated structural plane point cloud, and finally outputs clear and independent structural plane units.

[0103] The calculation of attitude parameters is carried out in the following steps:

[0104] S101: For the point cloud that constitutes a structural plane, randomly select three points to calculate their plane fitting equation AX + BY + CZ + D = 0;

[0105] S102: Calculate the perpendicular distance from each point in each structural plane point cloud to the fitting plane of the structural plane;

[0106] ;

[0107] S103: Set the distance threshold trsc, and consider the points >trsc as outlier points, and consider the points <trsc as inlier points, and count the number of inlier points Nin;

[0108] S104: Iterate steps S101 to S103 Nrsc times, and select the fitting plane with the largest number of inlier points;

[0109] S105: Perform plane fitting on the fitting plane with the largest number of inlier points to obtain the final plane fitting equation.

[0110] The calculation method of the plane fitting equation includes:

[0111] In the plane fitting equation The angle with the normal vector of the XOY plane is the dip angle (DA) of the structural plane, and the numerical range is 0° to 90°, which is defined as:

[0112] ;

[0113] Let be the projection of the normal vector on the XOY plane, be the angle between the positive direction of the X-axis and the projection , then it is expressed as:

[0114] ;

[0115] The dip direction (DD) of the structural plane is the included angle between the positive X-axis in the clockwise direction and the dip line, and the value range is 0° to 360°; let ∈ [0°, 90°], then the conversion relationship with DD is shown in Table 1:

[0116] Table 1

[0117]

[0118] The identification of the trace line distribution uses an iterative algorithm with PCA weighting for trace line orientation contraction. The identification method includes:

[0119] S201: Set the boundary point coordinate set , and re-search for the points within to find their neighboring points, and calculate the direction vector of each point;

[0120] The generation method of the boundary point coordinate set includes:

[0121] The sharp points in the cloud map are defined as the edge points and corner points with obvious geometric curvature in the three-dimensional point cloud model; due to the large geometric shape differences in the neighborhoods of sharp points, sharp points are identified based on the change in the included angle of the neighborhood normal vectors. Let the point cloud be P = { , ,..., }, and the normal vector be Vec = { , ,..., }, then the change value of the included angle of the neighborhood of the i-th normal vector, then is defined as:

[0122] ;

[0123] In the formula, the function Ang is used to calculate the acute included angle between two vectors, represents the -th neighboring point of the neighborhood of the point ; the sharp points are defined as the points where the change value of the neighborhood included angle is greater than the average value of the change values of the neighborhood included angles of all points in the point cloud. Substitute into the following formula to calculate the sharp point set (boundary point coordinate set) , that is:

[0124] ;

[0125] ​S202: For each of these points , the calculated eigenvalue is > > , then its contraction weight is defined as follows:

[0126] ;

[0127] In the formula, is the first eigenvalue obtained by PCA calculation for the -th boundary point; is the -th eigenvalue obtained by PCA calculation for the -th boundary point;

[0128] S203: Let the index of the neighboring points of point in be , then the contracted point ' of point is defined as follows:

[0129] ;

[0130] S204: Correct the displacement direction of ' so that the contracted points are evenly distributed along the trace length direction, cancel the parallel displacement relative to the trace center axis, and make finally undergo a vertical movement relative to the trace center axis rather than a parallel movement.

[0131] The spacing identification in the surrounding rock classification index adopts a method based on the two-dimensional virtual measurement line method for spacing identification.

[0132] After obtaining the attitude parameters (dip direction and dip angle) and trace distributions (the intersection lines of the structural planes with the observation plane or the fitting plane) of each independent structural plane in the tunnel surrounding rock, joint spacing identification is required, which is one of the core input indexes of the surrounding rock classification indexes (such as RQD and RMR, etc.); traditional methods directly measure the spacing on three-dimensional point clouds or complex surfaces, with large computational amounts and being easily affected by missing point clouds or noise interference; this method adopts the two-dimensional virtual measurement line method for efficient and robust spacing identification.

[0133] The principle and steps of the two-dimensional virtual measurement line method are as follows:

[0134] According to the average attitude parameters (dip direction, dip angle) of the dominant structural plane group in the target area (such as the tunnel face or the local side wall), calculate the average normal vector of the dominant structural plane group.

[0135] Construct a virtual two-dimensional projection plane with the average normal vector of the dominant structural plane group as the axis; this plane should be as parallel as possible to the target structural plane group in space. For example, if the average attitude of this group of structural planes is dip 120° and dip angle 75°, then construct a plane with a normal direction equivalent to (120° + 90°, 90° - 75°); the core of this step is to reduce the three-dimensional space problem to a two-dimensional plane for processing, that is:

[0136] Orthogonally project the trace point clouds of all independent structural planes that are extracted and belong to the same group (i.e., similar attitudes) (representing the intersection lines of the structural planes and the original scanning plane) onto the constructed virtual two-dimensional projection plane.

[0137] The traces that may be curved, undulating or interlaced in three-dimensional space are "straightened" and "flattened" on a plane parallel to their own average direction, approximated as a set of parallel straight line segments or broken line segments; the projection preserves the relative distance relationship between the traces.

[0138] In the constructed two-dimensional projection plane, lay out one or more virtual measurement lines (survey lines); the directions of these survey lines should be strictly perpendicular to the dominant extension direction of the projected traces (i.e., perpendicular to the average strike of this structural plane group).

[0139] The survey lines should cover the representative range of the target evaluation area; a single long survey line can pass through the area, or multiple parallel short survey lines can be evenly distributed to increase the sampling statistics.

[0140] For each virtual survey line, calculate its intersection points with all the projected traces.

[0141] Sort all the intersection points on this survey line in ascending or descending order according to their positions (coordinate values) along the survey line.

[0142] Calculate the distance along the survey line between two adjacent intersection points; this distance represents the apparent spacing between one structural plane trace and the next structural plane trace that the virtual survey line passes through.

[0143] Collect all the adjacent intersection point spacing values calculated on the virtual survey lines. Conduct statistical analysis on these spacing values (such as calculating the average value, maximum and minimum values, and frequency distribution), and finally obtain the representative joint spacing of this structural plane group (usually the average spacing).

[0144] Exemplary:

[0145] Consider a group of steeply dipping joints (average dip 180°, dip angle 80°) with similar attitudes on the tunnel face.

[0146] Construct a virtual projection plane, and its normal vector direction is approximately (90°, 10°) (to make it parallel to this group of joints).

[0147] Extract the trace point clouds of all joints in this group from the original scanned surface and orthogonally project them onto this virtual plane. After projection, the traces are approximately multiple straight line segments parallel to the strike (such as 270°).

[0148] Layout a virtual survey line L1 in the projection plane in the horizontal direction (i.e., perpendicular to the 270° strike, with the direction of 0° / 180°), and its length covers the width of the tunnel face (such as the survey line is 10 meters long).

[0149] The survey line L1 intersects with the 5 traces after projection, obtaining 5 intersection points; after sorting by the X coordinate, the positions are: P1(1.2m), P2(3.5m), P3(5.8m), P4(7.1m), P5(9.4m).

[0150] Spacing calculation: Spacing D1 = P2 - P1 = 2.3m, D2 = P3 - P2 = 2.3m, D3 = P4 - P3 = 1.3m, D4 = P5 - P4 = 2.3m.

[0151] Statistics: Average spacing = (2.3 + 2.3 + 1.3 + 2.3) / 4 ≈ 2.05 meters. This value can be used as the average spacing of this group of joints.

[0152] Converting the distance measurement in complex three-dimensional space into a simple distance calculation along a straight line in a two-dimensional plane greatly improves the calculation efficiency and meets the requirement of "fast" perception; at the same time, it has a certain tolerance for local missing of point clouds or incomplete traces (relying on statistical average).

[0153] The virtual survey line method simulates the standard method of a geological engineer using a surveying ruler to manually measure the joint spacing on an outcrop or tunnel wall surface, and the results are comparable and have engineering significance.

[0154] The calculated joint spacing is one of the key input parameters for surrounding rock classification indexes (such as RQD and RMR); combined with other indexes such as the roughness coefficient calculated from structural point clouds (through gradient modulus mapping), the comprehensive surrounding rock classification results can be finally output to provide a basis for tunnel support design.

[0155] The roughness calculation in the surrounding rock classification index determines the appropriate estimated value with the joint roughness coefficient value of the closest standard contour line; the method for estimating the two-dimensional curve JRC based on the root mean square includes:

[0156] ;

[0157] In the formula, represents the average gradient modulus of the structural surface, defined as:

[0158] ;

[0159] In the formula, is the number of points on the two-dimensional section intercepted along one direction. and are respectively the vertical coordinates of points in the three-dimensional data of the structural plane. is the distance between points on each profile line.

[0160] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalent technologies, the present invention is also intended to include these changes and modifications.

[0161] The above are only the preferred specific embodiments of the embodiments of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present application, according to the technical solution and its concept of the present application, makes equivalent substitutions or changes, and should be covered by the protection scope of the present application.

Claims

1. A dynamic, precise, rapid perception method for multi-source data of tunnel surrounding rock, characterized in that, The method includes: Based on the multi-view image sequence of the mobile terminal, shooting positions are arranged at preset spatial intervals and orthogonal angles, a sparse point cloud is generated by the Structure from Motion (SfM) algorithm, and a dense point cloud model is generated by combining multi-view stereo matching optimization; For the dense point cloud model, a local neighborhood is delimited based on k-nearest neighbor search, the covariance matrix of neighborhood points is solved by principal component analysis, the eigenvector corresponding to the minimum eigenvalue is extracted as the normal vector, and a microscopic geometric feature field of the rock mass surface is constructed; Based on the microscopic geometric feature field of the rock mass surface, the hierarchical clustering algorithm is used to aggregate the point cloud according to the consistency criterion of the normal vector direction to generate the structural plane point cloud; For the structural plane point cloud, the density clustering algorithm is used to separate the point cloud clusters with interlaced spatial distributions, and independent structural plane units are extracted by combining noise filtering and boundary constraints; For the boundary point cloud of the independent structural plane in the independent structural plane unit, the plane equation is iteratively optimized by the Random Sample Consensus (RANSAC) plane fitting algorithm, and the dip and dip angle parameters are parsed by combining normal vector projection and spatial geometry to obtain the attitude parameters; Based on the boundary point cloud of the independent structural plane, the weighted direction vector of the principal component analysis is optimized, and through the trace direction constraint and displacement orthogonal correction, the boundary points are iteratively shrunk to the central axis of the trace to obtain the trace distribution; Based on the attitude parameters and the trace distribution, the joint spacing is statistically calculated by the virtual survey line method, and the surrounding rock classification index is output by combining the calculation of the surface gradient modulus of the structural plane and the roughness coefficient mapping model; 2. The dynamic fine and rapid perception method for multi-source data of tunnel surrounding rock according to claim 1, characterized in that The calculation method of the normal vector includes: ； In the formula, is the th nearest neighbor point, is the default value. Let > > be the eigenvalues of the covariance matrix , , be the corresponding eigenvectors. Then the normal vector corresponding to the point is the eigenvector corresponding to the minimum eigenvalue .

3. The dynamic fine and rapid perception method for multi-source data of tunnel surrounding rock according to claim 1, characterized in that, The hierarchical clustering algorithm is the BIRCH clustering algorithm improved based on cosine similarity; for the BIRCH clustering algorithm improved based on cosine similarity, when inserting a new data point, cosine similarity is used to determine the closest cluster.

4. The dynamic fine and rapid perception method for multi-source data of tunnel surrounding rock according to claim 1, characterized in that The extraction of the independent structural plane unit includes using the DBSCAN density clustering algorithm to realize the segmentation of the independent structural plane.

5. The dynamic fine and rapid perception method for multi-source data of tunnel surrounding rock according to claim 1, characterized in that The calculation method of the attitude parameters includes: For the point cloud that constitutes a structural plane, three points are randomly selected to calculate its plane fitting equation AX + BY + CZ + D = 0; Calculate the perpendicular distance from each point in the point cloud of each structural plane to the fitting plane of the structural plane; ； Set a distance threshold trsc, and consider the points >trsc as outlier points, and <trsc as inlier points, and count the number Nin of inlier points; Iterate Nrsc times and select the fitting plane with the largest number of inliers; The fitting plane with the largest number of inliers is further subjected to plane fitting to obtain the final plane fitting equation.

6. The dynamic fine and rapid perception method for multi-source data of tunnel surrounding rock according to claim 5, characterized in that The calculation method of the plane fitting equation includes: In the plane fitting equation The included angle with the normal vector of the XOY plane is the dip angle of the structural plane , and the numerical range is , defined as: ； Let be the normal vector projected onto the XOY plane, and be the angle between the positive X-axis and the projection ； The dip direction of the structural plane is the clockwise angle between the positive X-axis and the dip line with a value range of 0° to 360°; let .

7. The dynamic fine and rapid perception method for multi-source data of tunnel surrounding rock according to claim 1, characterized in that The identification of the trace distribution uses the PCA weighted iterative algorithm for trace orientation contraction, and the identification method includes: Set the set of boundary point coordinates , for points, re-search for the nearest neighbors within and calculate the direction vector of each point; For each of these points , the calculated eigenvalue is > > , then its shrinkage weight is defined as: ； In the formula, is the first eigenvalue obtained by PCA calculation for the th boundary point; is the th eigenvalue obtained by PCA calculation for the th boundary point; Let the point 's nearest neighbor points in have indices of . Then the point after contraction, the point ' is defined as including: ； For correct the displacement direction of ', so that the shrunk points can be evenly distributed along the length direction of the trace, and cancel the parallel displacement relative to the central axis of the trace, so that finally, a vertical movement relative to the central axis of the trace occurs instead of a parallel movement.

8. The dynamic fine and rapid perception method for multi-source data of tunnel surrounding rock according to claim 1, characterized in that The spacing identification in the surrounding rock classification index uses the method based on the two-dimensional virtual survey line method for spacing identification.

9. The dynamic fine and rapid perception method for multi-source data of tunnel surrounding rock according to claim 1, wherein The roughness calculation in the surrounding rock classification index determines the appropriate estimated value by the joint roughness coefficient value of the standard contour line closest to it; The method for estimating the two-dimensional curve JRC according to the root mean square includes: ； In the formula, represents the average gradient modulus of the structural plane surface, which is defined as: ； In the formula, is the number of points on the two-dimensional section intercepted along one direction, and are the vertical coordinates of the points in the three-dimensional data of the structural plane, respectively; is the distance between points on each profile line.

Images