A tunnel primary support deformation detection method and system based on three-dimensional point cloud

Abstract

The application belongs to the technical field of tunnel construction monitoring, and particularly relates to a tunnel primary support deformation detection method and system based on three-dimensional point clouds. The method comprises the following steps: performing adaptive feature enhancement processing on original tunnel primary support point clouds to output enhanced point clouds; constructing an edge vector field with double-plane geometric constraints based on the enhanced point clouds; based on the edge vector field, synchronously optimizing and solving point cloud registration parameters and an initial screening deformation point set by using an alternating direction multiplier method through a joint objective function containing a registration item and a deformation detection item weight; applying the optimal registration parameters and the initial screening point set to calculate displacement consistency confidence through orthogonal projection decomposition, cross-verify and output a final deformation point set; and establishing a self-regression model with curvature constraints based on the final deformation point set to perform primary support deformation spatio-temporal evolution prediction and early warning. The application significantly improves the precision and reliability of primary support deformation detection in complex tunnel environments, effectively suppresses registration error transmission, and enhances long-term deformation prediction capability.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of tunnel construction monitoring, and particularly relates to a tunnel primary support deformation detection method and system based on three-dimensional point clouds. BACKGROUND

[0002] Tunnel engineering is a key component of modern transportation networks and infrastructure, and its structural safety is of great importance. The deformation of the tunnel primary support (such as the bending and misalignment of steel arches, the cracking and peeling of sprayed concrete, settlement, convergence, etc.) is a core indicator for assessing its health condition. Timely and accurate detection of primary support deformation is of great significance for preventing safety accidents, guiding maintenance decisions, and ensuring operational safety. In recent years, three-dimensional laser scanning technology has become the main means for obtaining massive three-dimensional point cloud data of tunnel primary support surfaces due to its non-contact, high efficiency, and high precision, providing a data basis for automatic deformation detection methods based on point clouds.

[0003] The core challenge of tunnel primary support deformation detection based on three-dimensional point clouds lies in handling noise interference of point cloud data, environmental heterogeneity (such as sparse point clouds in water infiltration areas, dense and noisy point clouds caused by structural attachments such as steel arches and sprayed concrete layers), point cloud registration accuracy, and effective differentiation between real deformation and measurement error. The current mainstream technical route mainly includes:

[0004] 1. Fixed threshold filtering combined with Iterative Closest Point (ICP) registration scheme: First, a pre-set fixed spatial distance threshold or curvature threshold is used to filter and denoise the original point cloud, then ICP algorithm is used for registration (coordinate alignment) of the benchmark period and monitoring period point clouds, and finally the distance difference (such as the nearest point distance) or surface model difference between point clouds is calculated to identify potential deformation points. This scheme has relatively simple and intuitive algorithm principles, clear calculation process, and is easy to implement. However, the fixed threshold cannot effectively cope with the dramatic changes in point cloud density along the longitudinal direction of the tunnel. In areas with sparse point clouds such as sprayed concrete peeling and water infiltration, the fixed threshold is likely to misjudge the real structure points as noise and filter them out, resulting in the loss of key deformation features; while in dense attachment areas such as steel arch connection areas and bolts, the fixed threshold is difficult to effectively remove discrete noise points (such as scaffold points), leading to a large amount of interference in subsequent analysis. In addition, registration (ICP) and deformation detection are carried out in two steps. ICP registration itself has errors, especially when the point cloud quality is poor (with much noise and few features) or the tunnel has large overall displacement / deformation, registration residuals will inevitably be passed to the subsequent deformation detection link. These registration errors will be mistaken for local deformation, leading to false positives; at the same time, they may also mask real small or asymmetric deformations (such as steel arch misalignment and interlayer peeling of sprayed concrete), leading to false negatives.

[0005] 2. Multi-scale feature fusion scheme: By extracting geometric features (such as normal, curvature) and reflectance intensity features of point cloud at different scales, fusion analysis is performed to obtain more robust deformation indicators. This scheme can improve the adaptability to complex environments to some extent.

[0006] 3. Dynamic modeling scheme based on time series: Establish the evolution model of deformation points by using multi-period monitoring data to make trend prediction. This is of great significance for long-term monitoring and early warning.

[0007] For the widely used fixed threshold filtering combined with step-by-step processing method of ICP registration, when processing point cloud data in complex environment of tunnel primary support, the environment adaptability is insufficient, and the registration error is transmitted to the deformation detection. Therefore, a detection method capable of improving the accuracy and reliability of deformation detection is needed. SUMMARY

[0008] The present application overcomes the shortcomings of the prior art and provides a tunnel primary support deformation detection method and system based on three-dimensional point cloud.

[0009] To achieve the above purpose, the technical scheme adopted by the present application is as follows: a tunnel primary support deformation detection method based on three-dimensional point cloud, comprising:

[0010] Step one: perform adaptive feature enhancement processing on the three-dimensional point cloud data obtained by scanning the original tunnel primary support surface, and output enhanced point cloud data that fuses geometric properties and reflection properties;

[0011] Step two: based on the enhanced point cloud data, construct an edge vector field with double-plane geometric constraints; the edge vector field with double-plane geometric constraints is formed by projecting the enhanced point cloud data onto the XOY plane and the YOZ plane, respectively, identifying edge points in the two projection planes, establishing spatial correlation of the edge points, and containing a vector field of spatial position, direction vector and source plane identifier;

[0012] Step three: based on the edge vector field, use the alternating direction multiplier method to simultaneously optimize and solve the point cloud registration parameters and the preliminary screening deformation point set through a joint objective function containing the registration item weight coefficient and the deformation detection item weight, and output the optimal registration parameters and the preliminary screening deformation point set;

[0013] Step four: apply the optimal registration parameters to coordinate correction of the monitoring period point cloud, and based on the preliminary screening deformation point set, calculate the displacement component and the displacement consistency confidence index through orthogonal projection decomposition, and output the final deformation point set after cross-validation;

[0014] Step five: based on the final deformation point set, establish a self-regression model with curvature constraint to perform spatio-temporal evolution prediction and early warning of deformation.

[0015] Further, the adaptive feature enhancement processing in step one comprises:

[0016] calculating the local point cloud density standard deviation within a preset radius sphere of each point in the original tunnel primary support surface point cloud;

[0017] dynamically calculating an adaptive Euclidean clustering radius according to the local point cloud density standard deviation;

[0018] performing a Euclidean clustering operation, and retaining clusters with a point number greater than the product of a retention proportion coefficient and the total amount of point clouds in the current tunnel longitudinal processing block;

[0019] performing joint filtering of curvature threshold and intensity gradient threshold on the points retained by clustering;

[0020] dividing the tunnel longitudinal processing block by a preset interval, and establishing a negative correlation constraint relationship between the retention proportion coefficient and the total amount of point clouds in the current tunnel longitudinal processing block.

[0021] Further, the step two of constructing the edge vector field of the double-plane geometric constraint comprises:

[0022] projecting the enhanced point cloud data onto the XOY plane and the YOZ plane, respectively;

[0023] calculating, in each projection plane, a vector set formed by a near-neighbor point set of each point, and extracting a maximum included angle value between adjacent vectors of the point;

[0024] labeling edge points according to a dynamic edge detection angle threshold, the dynamic edge detection angle threshold being calculated by a slope compensation coefficient and a tunnel longitudinal slope angle;

[0025] grouping the labeled edge points by direction, and constructing a double-plane edge point position correlation matrix;

[0026] screening high-correlation point pairs satisfying a correlation threshold, and forming an edge vector field containing spatial positions, direction vectors, and source plane identifiers.

[0027] Further, the calculation of the dynamic edge detection angle threshold satisfies that the slope compensation coefficient increases with the increase of the tunnel longitudinal slope angle.

[0028] Further, the step three of simultaneously optimizing and solving the point cloud registration parameters and the primary screening deformation point set comprises:

[0029] defining a time-space deformation differential vector of the edge vector field of the reference period and the monitoring period;

[0030] constructing a joint objective function containing a registration term weight coefficient and a deformation detection term weight, the registration term being used for parameterizing a rigid transformation, and the deformation detection term being used for defining a deformation projection direction;

[0031] iteratively solving an optimal rotation matrix, a translation vector, and a primary screening deformation point set by using an alternating direction multiplier method;

[0032] Confirming the preliminary screening deformation point according to the deformation determination threshold, and adjusting the deformation determination threshold in association with the curvature attribute and the intensity gradient attribute in the enhanced point cloud data output in step one.

[0033] Further, step four includes:

[0034] Correcting the monitoring period point cloud coordinates using the optimal rotation matrix and translation vector in the optimal registration parameters, and calculating the residual displacement vector thereof relative to the reference period point cloud;

[0035] Performing local orthogonal projection decomposition on the residual displacement vector in the XOY plane and the YOZ plane respectively to obtain the XOY plane projection displacement component and the YOZ plane projection displacement component;

[0036] Calculating the two-plane displacement consistency confidence index based on the XOY plane projection displacement component and the YOZ plane projection displacement component;

[0037] Verifying the true deformation point according to the preset confidence threshold and displacement threshold, and starting the re-projection mechanism based on the edge vector field for low confidence points;

[0038] Outputting the final deformation point set containing displacement vectors, confidence and source plane identification.

[0039] Further, step five includes:

[0040] Constructing a space-time feature vector for each deformation point in the final deformation point set, the space-time feature vector containing displacement, curvature attribute, intensity gradient and confidence;

[0041] Using an autoregressive model with curvature constraint to predict deformation evolution, the autoregressive model with curvature constraint containing displacement autoregressive coefficients and curvature response coefficients;

[0042] Capturing new deformation points based on local displacement standard deviation and new point confidence threshold;

[0043] Triggering the artificial review mechanism for historical deformation points according to the residual warning threshold.

[0044] Further, capturing new deformation points includes:

[0045] When the intensity gradient of the new point exceeds the preset intensity gradient threshold, the displacement thereof is required to meet the preset displacement threshold;

[0046] When the curvature attribute of the new point exceeds the preset curvature attribute threshold, the local displacement standard deviation constraint is expanded to 1.5 times.

[0047] Further, processing historical deformation points includes:

[0048] For the deformation points with the confidence lower than the preset confidence threshold, the corresponding residual warning threshold is raised to 2.5 mm;

[0049] For the points with the predicted residual exceeding the residual warning threshold for three consecutive periods, local registration recalculation is performed.

[0050] The application provides another technical scheme: a tunnel primary support deformation detection system based on three-dimensional point clouds, based on the above method, comprising:

[0051] The point cloud preprocessing module is configured to segment the three-dimensional point cloud data obtained by the original tunnel primary support surface scanning according to a preset interval and output blockized point cloud data;

[0052] The feature enhancement module is configured to perform dynamic clustering and joint filtering operations based on the blockized point cloud data and output enhanced point cloud data fusing geometric properties and reflection characteristics;

[0053] The edge vector field generation module is configured to project the enhanced point cloud data to XOY and YOZ planes and construct an edge vector field with double-plane geometric constraints;

[0054] The joint registration preliminary screening module is configured to simultaneously optimize point cloud registration parameters and a preliminary screening deformation point set based on the edge vector field;

[0055] The deformation verification module is configured to verify deformation authenticity through orthogonal projection decomposition and output a final deformation point set;

[0056] The evolution modeling module is configured to construct a space-time feature vector based on the final deformation point set and perform deformation evolution prediction;

[0057] The output end of the point cloud preprocessing module is connected to the input end of the feature enhancement module; the output end of the feature enhancement module is connected to the input end of the edge vector field generation module; the output end of the edge vector field generation module is connected to the input end of the joint registration preliminary screening module; the output end of the joint registration preliminary screening module is connected to the input end of the deformation verification module; the output end of the deformation verification module is connected to the input end of the evolution modeling module; and the control signal end of the evolution modeling module is feedback connected to the recalculation trigger end of the joint registration preliminary screening module.

[0058] The application solves the defects in the background art and has the following beneficial effects:

[0059] By calculating the local point cloud density standard deviation and driving the adaptive Euclidean clustering radius adjustment, the search range is expanded in low-density areas such as sprayed concrete surfaces to retain effective points in water permeable areas, and the radius is reduced in high-density areas such as steel arch connection areas to remove device-attached noise points; at the same time, combined with the joint filtering of curvature threshold and intensity gradient threshold, the synchronous mutation characteristics of curvature and reflected intensity caused by real deformation are used to eliminate the interference of uniformly distributed noise such as sprayed concrete surface floating slurry and steel arch attachments. Compared with the existing fixed threshold filtering method, the present application realizes local parameter dynamic optimization for the spatial heterogeneity of the initial support of the tunnel, significantly improves the signal-to-noise ratio of the feature points, and provides a high-fidelity data basis for subsequent double-plane analysis.

[0060] By constructing a target function containing a registration item weight coefficient and a deformation detection item weight, the optimal rotation matrix, translation vector and initial screening deformation point set are solved synchronously using the alternating direction multiplier method, and the coordinate system alignment and deformation feature extraction are completed in a single iteration using the differential characteristics of the edge vector field. Compared with the traditional step-by-step processing flow, the registration residual is avoided from being passed to the deformation detection link, the detection rate of asymmetric deformation such as steel arch dislocation and interlayer peeling of sprayed concrete is significantly improved, and the calculation time is reduced, especially suitable for initial support monitoring in strong interference environments such as water-rich strata tunnels.

[0061] By orthogonal projection decomposition to generate XOY plane projection displacement components and YOZ plane projection displacement components, combined with confidence threshold and displacement threshold to realize deformation cross-validation; the high-confidence deformation point set output is input into the autoregressive model with curvature constraint, and the displacement autoregressive coefficient and curvature response coefficient are dynamically adjusted with displacement rate, so that the model strengthens the contribution of the steel arch and sprayed concrete interface geometry in the slow deformation stage, and suppresses the prediction oscillation in the accelerated deformation stage, significantly reduces the long-term deformation trend prediction error of typical diseases such as steel arch stress deformation and sprayed concrete crack expansion, and greatly improves the prediction accuracy compared with existing single models. BRIEF DESCRIPTION OF DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, a brief introduction to the drawings needed to be used in the embodiments or prior art description will be given below. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art without creative labor;

[0063] Figure 1 It is a flowchart of a tunnel initial support deformation detection method based on three-dimensional point cloud;

[0064] Figure 2 It is a flowchart of step one of a tunnel initial support deformation detection method based on three-dimensional point cloud;

[0065] Figure 3 This is a flow chart of step 2 of a method for detecting deformation of a tunnel primary branch based on a three-dimensional point cloud;

[0066] Figure 4 This is a flow chart of step three of a method for detecting deformation of a tunnel primary branch based on a three-dimensional point cloud;

[0067] Figure 5 This is a flow chart of step 4 of a method for detecting deformation of a tunnel primary branch based on a three-dimensional point cloud;

[0068] Figure 6 This is a flowchart of step five of a method for detecting deformation of a tunnel primary branch based on a three-dimensional point cloud;

[0069] Figure 7 This is an architectural diagram of a tunnel primary branch deformation detection system based on three-dimensional point cloud. DETAILED DESCRIPTION

[0070] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0071] Exemplary methods:

[0072] like Figure 1 As shown, a method for detecting deformation of a tunnel primary branch based on a three-dimensional point cloud comprises the following steps:

[0073] Step 1: Adaptively enhance the 3D point cloud data obtained by scanning the original tunnel primary branch surface, and output enhanced point cloud data that integrates geometric attributes and reflection characteristics;

[0074] Step 2: Construct a dual-plane geometrically constrained edge vector field based on the enhanced point cloud data. The dual-plane geometrically constrained edge vector field is formed by projecting the enhanced point cloud data onto the XOY plane and the YOZ plane, identifying edge points in the two projection planes, and establishing spatial associations between the edge points. The field includes spatial positions, direction vectors, and source plane identifiers.

[0075] Step 3: Based on the edge vector field, the alternating direction multiplier method is used to simultaneously optimize the point cloud registration parameters and the initial screening deformation point set through a joint objective function that includes the registration item weight coefficient and the deformation detection item weight. The optimal registration parameters and the initial screening deformation point set are output.

[0076] Step 4: Apply the optimal registration parameters to perform coordinate correction on the point cloud during the monitoring period, and based on the initial screening deformation point set, calculate the displacement components and displacement consistency confidence index through orthogonal projection decomposition, and output the final deformation point set after cross-validation;

[0077] Step 5: Based on the final deformation point set, an autoregressive model with curvature constraints is established to predict and warn the spatiotemporal evolution of deformation.

[0078] Below, each step will be described in detail.

[0079] Step 1 dynamically adapts to the spatial heterogeneity of the tunnel environment. While preserving the true primary support geometry, it simultaneously filters out noise and enhances deformation-sensitive features. This addresses feature distortion issues caused by uneven point cloud density and interference from structural attachments such as steel arches, providing a high-quality data foundation for subsequent dual-plane analysis.

[0080] like Figure 2 As shown, step one includes:

[0081] Calculate the standard deviation of the local point cloud density within the preset radius sphere of each point in the original tunnel primary branch surface point cloud;

[0082] Dynamically calculate the adaptive Euclidean clustering radius based on the standard deviation of the local point cloud density;

[0083] Perform Euclidean clustering operation and retain clusters whose number of points is greater than the product of the retention ratio coefficient and the total point cloud volume of the current tunnel longitudinal processing block;

[0084] Perform joint filtering of curvature threshold and intensity gradient threshold on the points retained by the cluster;

[0085] The tunnel longitudinal processing blocks are divided according to the preset spacing, and a negative correlation constraint relationship between the retention ratio coefficient and the total amount of point cloud in the current tunnel longitudinal processing block is established.

[0086] Specifically, based on the 3D point cloud data obtained by scanning the surface of the original tunnel primary branch, the radius of each point is calculated. (Value range 0.3-0.8m) Standard deviation of point cloud density within the sphere (Unit: points / m³), this value represents the unevenness of the local point cloud distribution; using Drives adaptive Euclidean cluster radius calculation: , is the basic radius constant (range 0.03-0.07m), is the density response coefficient (range 0.08-0.15); when (High density threshold, value range 800-1200 points / m 3 ) when decreasing To reduce the risk of over-clustering, (low density threshold, value range 200-500 points / m³) Expand the search range; retain the number of points after clustering Clusters ( To retain the proportionality coefficient, the value range is 0.08-0.15; is the total amount of block point cloud), and curvature-intensity joint filtering is performed: (Curvature threshold, value range 0.03-0.07mm -1 )and (Intensity gradient threshold, value range 12%-25%) points are marked as feature points, and the final output is an enhanced point cloud that integrates geometric attributes and reflection characteristics. .in is the standard deviation of the local point cloud density (points / m³), reflecting the dispersion of the point cloud distribution; is the adaptive Euclidean clustering radius (m), the lower the density, the larger the value; is the local curvature of the point cloud (mm -1 ), characterizing the degree of surface curvature; is the reflection intensity gradient (%), reflecting the material mutation characteristics of the interface between the steel arch and the concrete.

[0087] This step is done by using the density standard deviation Dynamically control cluster radius , so that the search range is expanded to retain valid points in relatively low-density areas such as the shotcrete surface, and the radius is reduced to remove noise in high-density areas such as the steel arch connection area. In low-density areas such as the shotcrete surface where the point cloud density is less than 500 points / m³, the density response coefficient is increased To a value around 0.15, the adaptive Euclidean clustering radius Expand to the range of 0.08-0.12m, effectively avoiding the mistaken deletion of the real initial support point; in high-density areas such as steel arch connection areas with a point cloud density of more than 1000 points / m³, reduce To take a value around 0.08, Narrowed to the range of 0.04-0.06m, the discrete noise points such as steel arch bolts were eliminated, and the density standard deviation triggered by environmental heterogeneity was The mechanism of dynamically changing the cluster radius is used to ensure the optimal denoising effect under different regional characteristics.

[0088] By setting 0.03-0.07mm -1 Curvature threshold The geometric mutation features such as deformation of steel arch and cracks of concrete are captured, and a 12%-25% intensity gradient threshold is used to identify the steel arch and shotcrete interface, material variation area such as concrete spalling, and the double-threshold joint screening strategy uses the curvature and reflection intensity mutation characteristics caused by the true deformation to effectively exclude the interference of uniformly distributed noise such as shotcrete surface floating slurry, and significantly improves the signal-to-noise ratio of feature points.

[0089] Step one is aimed at the gradual change characteristics of the initial support environment in the longitudinal direction of the tunnel, and the processing block is divided at an interval of 20-30 m to ensure that the local density standard deviation is less than 0.05 , and the spatial effectiveness of the calculation is established at the same time , and the reserved proportion coefficient is established , and the negative correlation constraint relationship between the block size and the reserved proportion coefficient is established , and when the total amount of block point cloud exceeds

[0090] , the reserved proportion coefficient is reduced to about 0.08; while maintaining the accuracy of feature extraction, the efficiency of resource allocation is optimized.

[0091] As shown in Figure 3 , step two includes:

[0092] Projecting the enhanced point cloud data onto the XOY plane and the YOZ plane respectively;

[0093] In each projection plane, calculate the vector set composed of the near neighbor point set of each point, and extract the maximum included angle value between the adjacent vectors of the point;

[0094] Marking the edge points according to the dynamic edge detection angle threshold, which is calculated by the slope compensation coefficient and the longitudinal slope angle of the tunnel;

[0095] Grouping the marked edge points by direction, and constructing a double-plane edge point position correlation matrix;

[0096] Screening high-correlation point pairs that meet the correlation threshold to form an edge vector field containing spatial position, direction vector and source plane identifier.

[0097] Specifically, based on the feature-enhanced point cloud output by step one, first project it onto the XOY plane and the YOZ plane to generate two two-dimensional point cloud sets; for any point in each projection plane, calculate the near neighbor point set A vector set consisting of values ​​ranging from 6 to 10) , solve the angle between adjacent vectors And extract the maximum angle value ;when When it is marked as an edge point, the dynamic angle threshold By tunnel slope angle (Unit: °) Control: , slope compensation coefficient The value range is 8-12; the local main direction vector is fitted to the marked edge points using the RANSAC algorithm (unit vector), press The symbol value of is an axial unit vector) is divided into two groups of positive and negative edge points; a dual-plane position correlation matrix is ​​constructed simultaneously , filter to meet (Correlation threshold The high correlation point pairs (value range 0.65-0.75) eventually form an edge vector field that integrates the geometric constraints of the two planes. .in, is the longitudinal slope angle of the tunnel (°), which represents the degree of inclination of the tunnel axis; is the dynamic edge detection angle threshold (°), the greater the slope, the higher the threshold; is the correlation coefficient of the edge point position of the two planes (dimensionless), reflecting the geometric consistency across the planes; through the slope compensation term To offset edge angle distortion caused by tilted projection, dual-plane correlation screening ensures the matching reliability of the spatial position of edge points under orthogonal views. It is particularly suitable for edge identification of typical primary support structures such as steel arch splicing and shotcrete interlayer interfaces.

[0098] In this step, the slope angle Increase the slope compensation coefficient for inclined tunnel sections To set the value around 12, the dynamic angle threshold Increase to range, effectively suppressing missed inspection of key edge points such as steel arch deformation edge and shotcrete cracking boundary caused by projection deformation; and in the straight tunnel section ( )maintain , avoiding false detection caused by excessive thresholds and improving the edge detection accuracy in various linear tunnels.

[0099] By calculating the position correlation coefficient of the edge point set of the XOY plane and the YOZ plane , to satisfy The high correlation point pairs are jointly marked to eliminate the false edge response in the single-view projection; at the same time, the curvature properties in the point cloud are enhanced by features (Output from step 1) Verify the authenticity of the edge point geometric mutation, when The weight of the point in the vector field is strengthened at the same time, and the false alarm rate of edge points caused by non-deformation factors such as steel arch bolt protrusions and temporary support residues is reduced.

[0100] Step 1: Enhanced reflection intensity gradient Participate in the direction vector grouping process, The material variation points are given priority in direction calculation to ensure that the deformation characteristics of special areas such as shotcrete spalling and steel arch rust edges are not ignored; the edge vector field finally constructed Not only spatial location With direction vector , and also records the source plane identification, forming an asymmetric deformation representation base with both geometric invariance and material response characteristics.

[0101] Step three uses the differential characteristics of the edge vector field to simultaneously achieve high-precision alignment and initial deformation screening. By establishing a joint optimization objective function for the alignment term and the deformation detection term, it breaks the error transmission barrier of the traditional step-by-step processing flow and completes the coordinated solution of coordinate system alignment and initial asymmetric deformation feature extraction in a single step.

[0102] like Figure 4 As shown, step three includes:

[0103] Define the spatiotemporal deformation differential vectors of the edge vector field during the baseline period and the monitoring period;

[0104] Construct a joint objective function that includes the weight coefficient of the registration term and the weight of the deformation detection term. The registration term is used to parameterize the rigid transformation, and the deformation detection term is used to define the deformation projection direction.

[0105] The alternating direction multiplier method is used to iteratively solve the optimal rotation matrix, translation vector and initial screening deformation point set;

[0106] The initial screening deformation points are confirmed according to the deformation judgment threshold, and the deformation judgment threshold is adjusted by associating the curvature attribute and the intensity gradient attribute in the enhanced point cloud data output in step 1.

[0107] In this step, based on the edge vector field output in step 2 , define the spatiotemporal deformation differential vector for the baseline period and monitoring period data ( is the direction vector of the monitoring period, is the reference period direction vector); construct the joint objective function of fusion registration parameter optimization and deformation screening: , registration term ( ) Weight coefficient The value range is 0.8-1.2, and the weight range of deformation detection item is 0.25-0.35; is the rotation matrix around the Z axis (the tunnel axis) ( is the rotation angle, unit: radian), is the two-dimensional translation vector (unit: mm), is the local normal vector (given by and The optimal parameters and initial deformation point set are solved by alternating direction multiplier method ADMM. (Deformation judgment threshold The value range is 0.15-0.25mm). Among them, is the space-time deformation differential vector (dimensionless), representing the change of the direction vector; is the rotation angle around the tunnel axis (rad), reflecting the annular torsional deformation of the primary support; is the local normal vector (dimensionless), defining the deformation projection direction; The deformation threshold (mm) is used to initially screen deformation and control the sensitivity of deformation detection. The rigid motion component of the direction vector change is minimized by the registration item, and the deformation detection item extracts the projection mutation of the direction vector in the normal direction. The two are synergistically optimized through the weight coefficient. This method is particularly suitable for identifying typical deformations of primary supports such as steel arch misalignment and interlayer delamination of shotcrete.

[0108] In strong interference environments such as tunnels in water-rich strata, the weight of deformation detection items is increased Set the value to around 0.35 to enhance the response to material variation caused by water seepage; at the same time, reduce the weight of the registration item to 0.8, to prevent geological displacement from masking the real deformation, reduce the registration residual and improve the deformation detection rate of the steel arch.

[0109] Through the alternating optimization strategy of the ADMM algorithm, the deformation set is first fixed in each iteration optimization parameter: ; Then fix the parameters and update the deformation set: , where the dynamic threshold ( is the total number of iterations) to avoid local optimality; converge within 12 iterations, improve computational efficiency, and ensure the processing speed of dense feature points on the initial support surface.

[0110] Edge Vector Field Intensity gradient properties in (From step 1) Participate in deformation verification: When When the corresponding point The deformation judgment threshold Relaxed to 0.3mm to avoid false alarms at the water-stained area at the interface between the steel arch and the concrete; at the same time, the curvature attribute The points with higher weights are assigned to ensure that structural deformations such as concrete crack propagation and steel arch bending are not missed. The final output parameters for point cloud coordinate correction, and the preliminary screening of the deformation point set Drive step four of the fine quantization.

[0111] Step four separates the asymmetric deformation component and the residual error component by orthogonal projection, uses the double-plane geometric constraint to realize cross-validation of the initial support real deformation, and simultaneously solves the asymmetric deformation quantization and registration error verification problem, providing high-confidence input for the evolution modeling of the initial support deformation.

[0112] As shown in Figure 5 , step four includes:

[0113] Apply the optimal rotation matrix and translation vector in the optimal registration parameters to correct the monitoring period point cloud coordinates, and calculate the residual displacement vector relative to the reference period point cloud;

[0114] Perform local orthogonal projection decomposition on the residual displacement vector in the XOY plane and the YOZ plane respectively to obtain the XOY plane projection displacement component and the YOZ plane projection displacement component;

[0115] Calculate the double-plane displacement consistency confidence index based on the XOY plane projection displacement component and the YOZ plane projection displacement component;

[0116] Verify the real deformation points according to the preset confidence threshold and displacement threshold, and start the re-projection mechanism based on the edge vector field for low-confidence points;

[0117] Output the final deformation point set containing displacement vectors, confidence and source plane identification.

[0118] In this step, based on the preliminary screening of the deformation point set output in step three and the optimal registration parameters , first apply the rigid transformation to correct the monitoring period point cloud position: , calculate the residual displacement vector ( for the reference period coordinates); perform local projection decomposition in the XOY plane and the YOZ plane respectively: , where the normal vector and are extracted from the edge vector field of step two; construct the double-plane displacement consistency confidence index: ; when (confidence threshold value range 0.75-0.85) and (displacement threshold Confirm the point as the true deformation point and mark the final deformation point set when the value range is 1.0-1.5mm . wherein, is the XOY plane projection displacement component (mm), representing the initial support ring deformation; is the YOZ plane projection displacement component (mm), representing the initial support radial deformation; is the two-plane displacement consistency confidence (dimensionless), reflecting the deformation authenticity; is the displacement difference tolerance coefficient (mm), with a value range of 0.8-1.2; the orthogonal projection separates the deformation direction component, and the confidence index quantifies the spatial consistency of the two-plane displacement vector, and excludes the registration residual error through geometric constraints, which is especially suitable for deformation verification of complex initial support structures such as steel arch splicing and interlayer interfaces of sprayed concrete.

[0119] In typical asymmetric deformation areas such as arch crown initial support settlement, the confidence threshold is tightened to a value near 0.85, while the displacement threshold is lowered to 1.0mm, enhancing the ability to capture small deformations; while in the side wall initial support convergence area, it is relaxed to 0.75, adapting to the symmetric deformation characteristics, and improving the detection accuracy of concrete cracks and steel arch misplacement.

[0120] Use the curvature attribute output in step one to verify the reliability of the projected displacement: when , it is forced to mm and mm, ensuring that the deformation of geometric mutation areas such as steel arch bending and concrete cracking is not underestimated; at the same time, points with attribute are given a 1.5 times tolerance coefficient to avoid misjudgment in areas such as sprayed concrete peeling and steel arch corrosion interfaces, and reduce the false alarm rate.

[0121] Start the re-projection mechanism for low-confidence points : based on the edge vector field in step two , recalculate the local normal vector and iteratively update the projection component until or the maximum number of iterations (3 times) is reached, and the non-convergent points are handed over to manual review; the final output set not only contains displacement vectors , but also records the confidence and the source plane identifier, forming a complete deformation decoupling description.

[0122] Step five establishes an autoregressive model of the initial support deformation state, realizes the capture of new initial support deformation and the prediction and early warning of deformation trend by fusing historical deformation characteristics and real-time monitoring data, forms a complete closed loop from initial support deformation detection to state evaluation, and solves the dynamic tracking and data continuous utilization problems in long-term monitoring.

[0123] As shown in Figure 6 , step five includes:

[0124] For each deformation point in the final deformation point set, a spatiotemporal feature vector is constructed, which includes displacement, curvature attribute, intensity gradient and confidence;

[0125] An autoregressive model with curvature constraint is used to predict deformation evolution, which includes displacement autoregressive coefficient and curvature response coefficient;

[0126] New deformation points are captured based on local displacement standard deviation and new point confidence threshold;

[0127] The artificial review mechanism of historical deformation points is triggered according to residual warning threshold.

[0128] In this step, based on the final deformation point set output in step four, a spatiotemporal feature vector is constructed for each initial support deformation point: , is the displacement (unit: mm), is the curvature attribute (unit: mm -1 ), is the intensity gradient (unit: %), is the confidence; an autoregressive model with curvature constraint is used to predict initial support deformation evolution: , the model order takes the value range of 2-4, the curvature response coefficient takes the value range of 0.15-0.25; when the new point meets ( is the local displacement standard deviation) and (new point confidence threshold takes the value range of 0.65-0.75), it is added to the deformation field and its feature vector is initialized; for historical points, when the predicted residual (prediction residual threshold takes the value range of 1.8-2.2 mm), the artificial review mechanism is triggered. Among them, is the spatiotemporal feature vector (mixed unit), which integrates displacement, curvature, intensity and confidence; is the displacement autoregressive coefficient (dimensionless), which reflects the inertia of historical deformation; is the curvature response coefficient (mm²), which quantifies the contribution of geometric mutation to evolution; is the residual warning threshold (mm), which controls the sensitivity of primary support trend anomalies. The autoregressive term is used to capture the time inertia of primary support deformation, and the curvature term is used to associate geometric mutations with evolution acceleration, thus achieving short-term prediction and long-term trend assessment. It is particularly suitable for the evolution analysis of typical primary support defects such as stress deformation of steel arches and interlayer delamination of shotcrete.

[0129] In the slow deformation stage of the initial support (displacement rate mm / day), reduce the model order to 2nd order and increase the curvature coefficient to 0.25, strengthening the contribution of geometric properties to the trend; and in the accelerated deformation stage ( mm / day) increases To the fourth order, the prediction oscillation is suppressed by long-range historical data, thus reducing the prediction error of sudden deformation of the primary support structure.

[0130] New point capture fusion step 2 vector field verification: when When, requirements To eliminate the interference of temporary water stains on the surface of the steel arch; mm -1 When Relax it to 1.5 times to quickly absorb new cracks, increased deformation of steel arch frames and other features, and reduce the missed reporting rate.

[0131] Confidence update of step 4 of historical point residual analysis linkage: , improve its The threshold is set to 2.5mm to reduce false alarms. When the residual exceeds the limit for three consecutive periods, the local recalculation in step 3 is triggered to verify whether it is caused by the drift of the primary branch benchmark. Finally, the primary branch deformation evolution map is output. , including three-dimensional data of displacement field, trend field and confidence field.

[0132] Example systems:

[0133] like Figure 7 As shown, a tunnel primary branch deformation detection system based on three-dimensional point cloud includes a point cloud preprocessing module, a feature enhancement module, an edge vector field generation module, a joint registration screening module, a deformation verification module and an evolutionary modeling module connected in sequence.

[0134] First, the point cloud preprocessing module outputs to the feature enhancement module; the feature enhancement module outputs to the edge vector field generation module; the edge vector field generation module outputs to the joint registration screening module; the joint registration screening module outputs to the deformation verification module; the deformation verification module outputs to the evolutionary modeling module.

[0135] In a second aspect, the block size monitoring unit of the feature enhancement module feeds back a block segmentation instruction to the point cloud preprocessing module; the residual overrun signal triggers the local recalculation of the joint registration preliminary screening module; the joint registration preliminary screening module calls the curvature attribute and the intensity gradient attribute stored in the feature enhancement module; the re-projection unit of the deformation verification module calls the local normal vector stored in the edge vector field generation module.

[0136] The point cloud preprocessing module is configured to receive original tunnel primary support scanning point cloud data and perform spatial block division, divide the tunnel longitudinal primary support point cloud at a preset interval of 20-30 meters, and output the block point cloud data to the feature enhancement module; at the same time, the module is provided with a block size monitoring unit, which triggers a warning signal when the total amount of points in a single block exceeds a threshold.

[0137] The feature enhancement module is configured to perform noise filtering and feature enhancement operations based on the block point cloud data, including:

[0138] The density analysis unit calculates the local point cloud density standard deviation in a spherical domain with a radius in the range of 0.3-0.8 meters and each point as the center;

[0139] The dynamic clustering unit dynamically calculates an adaptive Euclidean clustering radius based on the local point cloud density standard deviation, wherein the basic radius constant is in the range of 0.03-0.07 meters, the density response coefficient is in the range of 0.08-0.15 and is controlled by the high density threshold in the range of 800-1200 points per cubic meter and the low density threshold in the range of 200-500 points per cubic meter, and different density areas such as steel arch connection area and sprayed concrete surface are processed;

[0140] The joint filtering unit performs synchronous screening of the curvature threshold in the range of 0.03-0.07 per millimeter and the intensity gradient threshold in the range of 12%-25%, and retains feature points such as steel arch deformation and concrete cracks, and outputs the enhanced point cloud data with geometric properties and reflection characteristics to the edge vector field generation module;

[0141] The module also implements a negative correlation constraint mechanism of the reservation ratio coefficient and the total amount of block point clouds, which automatically reduces the reservation ratio coefficient to a value near 0.08 when the total amount of block point clouds exceeds a threshold, thereby optimizing the efficiency of primary support point cloud processing.

[0142] The edge vector field generation module is configured to construct an edge vector field with double-plane geometric constraints based on the enhanced point cloud data, including:

[0143] The double-plane projection unit projects the primary support point cloud onto the XOY plane and the YOZ plane to generate two-dimensional point cloud sets;

[0144] The edge detection unit calculates a set of vectors composed of a set of neighboring points with a value range of 6-10 of each point, extracts the maximum included angle value between adjacent vectors, and marks the edge points by a dynamic edge detection angle threshold, wherein the angle threshold is determined by the longitudinal slope angle of the tunnel and a slope compensation coefficient value range of 8-12, which is suitable for the primary support structure of the inclined tunnel;

[0145] The direction grouping unit highlights the direction features of the steel arch splicing and the interface between the sprayed concrete layers by fitting the main direction vector of the edge points and performing positive and negative direction grouping through the RANSAC algorithm;

[0146] The correlation matrix unit constructs a two-plane position correlation matrix, filters high-correlation point pairs that meet a correlation threshold value range of 0.65-0.75, and generates an edge vector field containing spatial positions, direction vectors, and source plane identifiers, which is transmitted to the joint registration preliminary screening module.

[0147] The joint registration preliminary screening module is configured to synchronously perform point cloud registration and primary support deformation preliminary screening, including:

[0148] The differential vector unit defines the spatiotemporal deformation differential vector of the edge vector field of the reference period and the monitoring period;

[0149] The joint optimization unit constructs a target function containing a registration term weight coefficient value range of 0.8-1.2 and a deformation detection term weight value range of 0.25-0.35, parameterizes the rotation matrix and translation vector of the rigid transformation, and focuses on adapting the deformation such as the ring-wise twisting and radial displacement of the primary support;

[0150] The iterative solution unit solves the optimal registration parameters and the preliminary screening deformation point set within 12 iterations by using the alternating direction multiplier method, wherein the deformation judgment threshold value range is 0.15-0.25 millimeters;

[0151] The module enhances the curvature attribute and the intensity gradient attribute of the joint feature enhancement module, and when the intensity gradient exceeds the threshold, the deformation judgment threshold is relaxed to 0.3 millimeters to avoid the interference of water stains at the steel arch and concrete interface.

[0152] The deformation verification module is configured to realize the cross-verification of the primary support deformation by orthogonal projection decomposition, including:

[0153] The coordinate correction unit adjusts the monitoring period primary support point cloud coordinates using the optimal registration parameters and calculates the residual displacement vector;

[0154] The projection decomposition unit performs local projection in the XOY plane and the YOZ plane respectively to obtain the XOY plane projection displacement component (representing the ring-wise deformation of the primary support) and the YOZ plane projection displacement component (representing the radial deformation of the primary support);

[0155] A confidence evaluation unit calculates a two-plane displacement consistency confidence index, verifies true deformation points in combination with a confidence threshold value range of 0.75-0.85 and a displacement threshold value range of 1.0-1.5 mm, and focuses on verifying features such as initial support settlement of a vault and initial support convergence of a side wall.

[0156] A re-projection unit starts an edge vector field-based re-projection mechanism for points with a confidence lower than a threshold value, performs at most 3 iterations, and outputs a final deformation point set containing displacement vectors, confidence and source plane identification to an evolution modeling module.

[0157] The module automatically tightens the confidence threshold value to 0.85 and reduces the displacement threshold value to 1.0 mm in the initial support settlement area of the vault.

[0158] The evolution modeling module is configured to establish an initial support deformation autoregressive model, including:

[0159] A feature construction unit generates a spatiotemporal feature vector containing displacement, curvature attribute, strength gradient and confidence for each initial support deformation point.

[0160] An autoregressive prediction unit predicts initial support deformation evolution using an autoregressive model with curvature constraint, with model order value range of 2-4 and curvature response coefficient value range of 0.15-0.25 dynamically adjusted according to displacement rate, adapting to different evolution modes such as stress deformation of a steel arch and interlayer spalling of sprayed concrete.

[0161] A dynamic monitoring unit captures new initial support deformation points based on local displacement standard deviation and a new point confidence threshold value range of 0.65-0.75, and expands the local displacement standard deviation constraint by 1.5 times when the curvature attribute exceeds the threshold value, quickly identifying features such as new cracks and intensified deformation of a steel arch.

[0162] The module sets a residual warning threshold value range of 1.8-2.2 mm for historical initial support deformation points, and triggers local recalculation of the joint registration preliminary screening module when the prediction residual exceeds the limit for 3 consecutive periods.

[0163] If the above functions are realized in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or partially contribute to the prior art, or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the embodiments of the present application. The foregoing storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0164] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present application can be implemented in various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0165] The present application is described with reference to flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1 an apparatus that performs the functions specified in one block or multiple blocks.

[0166] These computer program instructions can also be stored in a computer readable storage medium that can direct the computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable storage medium produce a manufactured product including instruction apparatus, which implements the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1 an apparatus that performs the functions specified in one block or multiple blocks.

[0167] These computer program instructions can also be loaded into computer or other programmable data processing devices, so that a series of operations steps are performed on the computer or other programmable data processing devices to generate computer-implemented processes, thus the instructions executed on the computer or other programmable data processing devices provide the function of realizing the processes specified in the flowchart Figure 1 one flowchart or multiple flowcharts and / or blocks Figure 1 one block or multiple blocks.

[0168] Although the preferred embodiments of the application have been described, those skilled in the art will be able to make additional modifications and variations to these embodiments without departing from the spirit and scope of the application. Accordingly, it is intended that the appended claims be construed to include all such modifications and variations as fall within the scope of the application.

[0169] Obviously, various modifications and changes are possible in the present application without deviating from the spirit and scope of the application. It is therefore intended that the present application encompass all such modifications and changes as fall within the scope of the claims and their equivalents.

Claims

1. A method for detecting deformation of tunnel primary branch based on three-dimensional point cloud, characterized in that: include: Step 1: Adaptively enhance the 3D point cloud data obtained by scanning the original tunnel primary branch surface, and output enhanced point cloud data that integrates geometric attributes and reflection characteristics; Step 2: Constructing a dual-plane geometrically constrained edge vector field based on the enhanced point cloud data; the dual-plane geometrically constrained edge vector field is formed by projecting the enhanced point cloud data onto the XOY plane and the YOZ plane, identifying edge points in the two projection planes, and establishing spatial associations between the edge points. The vector field includes spatial positions, direction vectors, and source plane identifiers. Step 3: Based on the edge vector field, the point cloud registration parameters and the initial screening deformation point set are simultaneously optimized using the alternating direction multiplier method through a joint objective function that includes the registration item weight coefficient and the deformation detection item weight, and the optimal registration parameters and the initial screening deformation point set are output; Step 4: Apply the optimal registration parameters to perform coordinate correction on the monitoring period point cloud, and based on the initial screening deformation point set, calculate the displacement components and displacement consistency confidence index through orthogonal projection decomposition, and output the final deformation point set after cross-validation; Step 5: Based on the final deformation point set, an autoregressive model with curvature constraint is established to predict and warn the spatiotemporal evolution of deformation.

2. The method according to claim 1, characterized in that The adaptive feature enhancement process in step 1 includes: Calculate the standard deviation of the local point cloud density within the preset radius sphere of each point in the original tunnel primary branch surface point cloud; Dynamically calculating an adaptive Euclidean clustering radius according to the standard deviation of the local point cloud density; Perform Euclidean clustering operation and retain clusters whose number of points is greater than the product of the retention ratio coefficient and the total point cloud volume of the current tunnel longitudinal processing block; Perform joint filtering of curvature threshold and intensity gradient threshold on the points retained by the cluster; The tunnel longitudinal processing blocks are divided according to a preset interval, and a negative correlation constraint relationship between the retention ratio coefficient and the total amount of point cloud of the current tunnel longitudinal processing block is established.

3. The method according to claim 1, characterized in that The step 2 of constructing the edge vector field of the dual-plane geometric constraint includes: Projecting the enhanced point cloud data onto the XOY plane and the YOZ plane respectively; In each projection plane, calculate the vector set composed of the neighboring point set of each point, and extract the maximum angle value between the adjacent vectors of the point; Marking edge points according to a dynamic edge detection angle threshold, wherein the dynamic edge detection angle threshold is calculated by a slope compensation coefficient and a longitudinal slope angle of the tunnel; Directionally group the marked edge points and construct a bi-plane edge point position correlation matrix; Highly correlated point pairs that meet the correlation threshold are screened to form an edge vector field containing spatial position, direction vector and source plane identification.

4. The method according to claim 3, characterized in that The calculation of the dynamic edge detection angle threshold satisfies that the slope compensation coefficient increases as the longitudinal slope angle of the tunnel increases.

5. The method according to claim 1, wherein The synchronous optimization of point cloud registration parameters and initial screening of deformation point sets in step 3 include: Define the spatiotemporal deformation differential vectors of the edge vector field during the baseline period and the monitoring period; Constructing a joint objective function including a registration term weight coefficient and a deformation detection term weight, wherein the registration term is used to parameterize the rigid transformation and the deformation detection term is used to define the deformation projection direction; The alternating direction multiplier method is used to iteratively solve the optimal rotation matrix, translation vector and initial screening deformation point set; The deformation points are initially screened according to the deformation determination threshold, and the deformation determination threshold is adjusted by associating the curvature attribute and the intensity gradient attribute in the enhanced point cloud data output in step 1.

6. The method according to claim 1, characterized in that The fourth step includes: Applying the optimal rotation matrix and translation vector in the optimal registration parameters to correct the coordinates of the monitoring period point cloud, and calculating the residual displacement vector relative to the reference period point cloud; Perform local orthogonal projection decomposition on the residual displacement vector in the XOY plane and the YOZ plane respectively to obtain an XOY plane projection displacement component and a YOZ plane projection displacement component; Calculating a bi-plane displacement consistency confidence index based on the XOY plane projection displacement component and the YOZ plane projection displacement component; Verifying the true deformation point according to a preset confidence threshold and displacement threshold, and starting a reprojection mechanism based on the edge vector field for low-confidence points; The output is the final deformed point set containing the displacement vector, confidence score, and source plane identification.

7. The method according to claim 1, characterized in that The step five includes: Constructing a spatiotemporal feature vector for each deformation point in the final deformation point set, wherein the spatiotemporal feature vector includes a combined displacement, a curvature attribute, an intensity gradient, and a confidence level; The deformation evolution is predicted using an autoregressive model with curvature constraint, wherein the autoregressive model with curvature constraint includes a displacement autoregressive coefficient and a curvature response coefficient; Capture new deformation points based on local displacement standard deviation and new point confidence threshold; The manual review mechanism of historical deformation points is triggered according to the residual warning threshold.

8. The method according to claim 7, characterized in that New deformation points captured include: When the intensity gradient of the newly added point exceeds the preset intensity gradient threshold, its displacement is required to meet the preset displacement threshold; When the curvature attribute of the newly added point exceeds a preset curvature attribute threshold, the local displacement standard deviation constraint is expanded to 1.5 times.

9. The method according to claim 7, characterized in that Processing historical deformation points includes: For deformation points with confidence levels lower than the preset confidence threshold, the corresponding residual warning threshold is raised to 2.5mm; For points where the prediction residuals for three consecutive periods exceed the residual warning threshold, local registration recalculation is performed.

10. A tunnel primary branch deformation detection system based on three-dimensional point cloud, based on the method according to any one of claims 1 to 9, characterized in that: include: a point cloud pre-processing module configured to segment the three-dimensional point cloud data obtained by scanning the original tunnel primary branch surface according to a preset interval and output the segmented point cloud data; a feature enhancement module configured to perform dynamic clustering and joint filtering operations based on the block-based point cloud data, and output enhanced point cloud data that integrates geometric attributes and reflectance characteristics; an edge vector field generation module, configured to project the enhanced point cloud data onto an XOY plane and a YOZ plane to construct an edge vector field with dual-plane geometric constraints; a joint registration preliminary screening module, configured to simultaneously optimize point cloud registration parameters and preliminary screening deformation point sets based on the edge vector field; A deformation verification module is configured to verify the authenticity of the deformation through orthogonal projection decomposition and output a final deformation point set; an evolutionary modeling module configured to construct a spatiotemporal feature vector based on the final deformation point set and perform deformation evolution prediction; Among them, the output end of the point cloud preprocessing module is connected to the input end of the feature enhancement module; the output end of the feature enhancement module is connected to the input end of the edge vector field generation module; the output end of the edge vector field generation module is connected to the input end of the joint registration preliminary screening module; the output end of the joint registration preliminary screening module is connected to the input end of the deformation verification module; the output end of the deformation verification module is connected to the input end of the evolutionary modeling module; the control signal end of the evolutionary modeling module is feedback connected to the recalculation trigger end of the joint registration preliminary screening module.

Images