Deformation detection method based on three-dimensional point cloud data of river sluice

Abstract

The invention discloses a deformation detection method based on three-dimensional point cloud data of a river sluice, and the method comprises the steps: 1), carrying out the observation station arrangement of the river sluice, scanning the surface of the river sluice through a ground three-dimensional laser scanner, and obtaining the point cloud data of the river sluice; 2) extracting contour points of the river sluice based on a river sluice stable area control point extraction method; 3) registering two-stage river sluice point cloud data based on a registration algorithm optimized by a symmetric objective function; and 4) based on a multi-scale river sluice surface deformation algorithm of a point cloud roughness confidence interval, calculating a deformation amount between two registered river sluice point clouds. According to the method, accurate control points are extracted as feature input based on a non-deformation stable region in an overall environment, a registration correction algorithm objective function is optimized, the spatial position of the river sluice is corrected, and a confidence interval is set by using each part of the river sluice and an overall point cloud roughness distribution rule; and accurate detection of actual deformation of the river sluice is completed based on multi-period prototype observation data.

Description

Technical Field

[0001] This invention relates to the field of deformation detection technology for river dams, specifically a deformation detection method based on three-dimensional point cloud data of river dams. Background Technology

[0002] A river dam is an important hydraulic engineering facility used to control river flow and regulate water levels and volumes. Its structural design (gates, piers, and opening / closing equipment) must comprehensively consider factors such as topography, hydrological conditions, and intended use. During its service life, a river dam is affected by water levels and environmental factors, resulting in varying degrees of deformation. When deformation reaches its limit, it can lead to structural instability. Therefore, deformation monitoring of river dams is particularly important.

[0003] Traditional detection methods primarily rely on point-based inspections, lacking comprehensive control over the overall deformation of river dams. Terrestrial Laser Scanning (TLS) technology can acquire massive amounts of high-density, high-resolution 3D point clouds of the river dam surface in a comprehensive, large-area, and highly efficient manner. As a non-contact, proactive surface measurement technology, it provides a new method for measurement work in the construction and operation management of water conservancy projects. It is of paramount importance for developing intelligent systems platforms for river dam safety inspection, early warning and forecasting, and comprehensive safety status evaluation. However, the following problems still exist when using TLS technology for river dam deformation detection:

[0004] 1) TLS technology can obtain multi-phase point cloud surface models of river dams. Complete river dam data is different from single-station data. It has a larger size and data volume, and there are a lot of redundant points in the huge data, which will waste a lot of computing power of the detection system.

[0005] 2) The point cloud data obtained by TLS technology is spatially random, and it is necessary to register the point cloud data of multiple complete dams to the same spatial coordinates. The registration of large-scale data is difficult.

[0006] 3) TLS technology can effectively detect the deformation of river dams, but directly calculating the distance between unstructured point clouds can lead to significant deviations in deformation analysis, making it difficult to determine the specific extent of the deformation of the river dam. Summary of the Invention

[0007] To overcome the shortcomings of existing technologies, this invention provides a deformation detection method based on three-dimensional point cloud data of a river dam. Using point cloud data obtained from terrestrial three-dimensional laser scanning technology as a foundation, the method first extracts accurate control points from the non-deformable stable regions within the overall environment. Secondly, based on the feature input of the control points, the objective function of the registration and correction algorithm is optimized to correct the spatial position of the river dam. Finally, utilizing the roughness distribution patterns of various components of the river dam and the overall point cloud, a 95% confidence interval for point cloud roughness is established, and accurate detection of the actual deformation of the river dam is completed based on multi-period prototype observation data.

[0008] To achieve the above objectives, the present invention adopts the following technical solution.

[0009] A deformation detection method based on 3D point cloud data of a river dam includes the following steps:

[0010] Step S1: Set up monitoring stations for the dam and use a ground-based 3D laser scanner to scan the surface of the dam and obtain point cloud data of the dam.

[0011] Step S2: Extract the outline points of the dam based on the control point extraction method for the stable zone of the dam.

[0012] In the point cloud data of the control point of the stable area, the point group with characteristic representation is selected by the method of filling the normal vector grid with angle. The threshold points are then selected by the sum vector to further extract the registration points covering the entire dam area, so as to provide accurate feature information for subsequent model registration.

[0013] Step S3: Register the point cloud data of the two phases of the river dam using a registration algorithm based on symmetric objective function optimization;

[0014] The multi-phase point cloud obtained by the point-surface 3D laser scanning is relatively random in spatial location. The contour control points of the dam are usually not in the same coordinate system. In order to achieve deformation detection, the point cloud data of the two phases need to be registered to the same coordinate system. Based on the traditional ICP algorithm, a new symmetric objective function is designed. The zero set of the symmetric objective function is not affected by variables and constraints, and has higher robustness and faster convergence speed.

[0015] Step S4: Calculate the deformation amount between the registered point clouds of the two phases of the river dam surface based on the multi-scale barrage surface deformation algorithm of the point cloud roughness confidence interval.

[0016] Point cloud data acquired by a terrestrial 3D laser scanner is inherently uncertain. Changes in scanning angle and distance can lead to different positional states in the point cloud data. Due to the uncertainty of point cloud position, a simple point-to-point distance calculation model cannot accurately represent the displacement changes of the two phases of the dam. Therefore, a multi-scale dam surface deformation algorithm based on the point cloud roughness confidence interval is designed to eliminate the influence of roughness on dam deformation detection by constructing a reasonable confidence interval.

[0017] Specifically, the details of the method for extracting control points in the stable zone of the dam in step S2 are as follows:

[0018] Step S21: Based on the point cloud data of the dam obtained in step S1, calculate the value of each point in the point cloud. normal vector ;

[0019] Step S22: Calculate each point The normal vector histogram, i.e., neighborhood points normal vector Compared to Histogram, neighborhood points Points within the radius R of the domain are called domain points. The normal vector is , ∈ ; and The angular relationship between them is expressed as:

[0020] (1);

[0021] In the above formula, Represents the cross product; Represents the dot product between two vectors;

[0022] Step S23: Divide the histogram into 18 squares, using values ​​that are always non-negative. Angle-filled histogram has 18 squares, each square being 10° in size. After loading the angles of all neighborhood normals into the squares, the count of neighborhood normals falling into a specific square is obtained. The histogram is then normalized to make it independent of neighborhood density. At this point, points with gentle features will cluster in the first square.

[0023] Step S24: Calculate the kurtosis of the histogram. :

[0024] (2);

[0025] (3);

[0026] In the above formula, Indicates the kurtosis of the histogram; Usually, 18 is chosen, which is the number of squares in the histogram; Indicates the first... Counting of squares; This represents the average value of all squares in the histogram. This represents the standard deviation of the histogram grid.

[0027] Through calculation It can remove flat and less curved areas to perform preliminary screening of the surface contour of the dam.

[0028] Step S25: Calculate each point and within its sampling radius The vector differences between each neighborhood point are weighted using a Gaussian weighting function. The weighted vector differences are then summed and divided by the number of sample points. To obtain the average vector difference, we then calculate the L2 norm of the average vector difference for each point. Assign a sum vector value By setting appropriate It can further filter the contours of the rock mass surface, thereby extracting the desired contours, among which... The calculation formula is:

[0029] (4);

[0030] In the above formula, Represents the Gaussian weighting function; Indicates the sampling method performed within the domain; This indicates the number of neighborhood points within the sampling radius.

[0031] This invention employs two evaluation indicators and To evaluate the performance of feature extraction methods, among which This represents the accuracy of feature extraction. This represents the error rate of feature extraction. and The calculation method is as follows:

[0032] ;

[0033] ;

[0034] In the above formula, This represents the number of correctly extracted feature points out of all extracted feature points. This represents the total number of feature points in the reference point cloud; This indicates the number of feature points that were incorrectly extracted.

[0035] Specifically, the registration algorithm based on symmetric objective function optimization in step S3 is as follows:

[0036] The objective function of the traditional point-to-surface ICP algorithm Represented as:

[0037] (5);

[0038] In the above formula, It is a rotation matrix; It is a translation vector; Source point cloud; For the target point cloud; Let be the normal vector of a point in the target point cloud;

[0039] The optimal registration solution of the traditional point-to-surface ICP algorithm can only achieve theoretical zero residual under specific geometric conditions. That is, the point cloud to be registered satisfies the geometric flatness constraint in the local neighborhood. Since the normal vector field of the flat surface region has a uniform distribution characteristic, the degrees of freedom of the rigid body transformation parameters in the tangent plane direction will not introduce additional registration errors. Only then can the residual reach a strictly zero value when the algorithm iteratively converges to the global optimal solution.

[0040] The zero set of the symmetric objective function defines the unconstrained property of the transformation, that is, when the surface slides along its tangent plane, its geometric features are fixed. Based on this, a new symmetric objective function is designed to optimize the point-to-surface ICP algorithm. The zero set of this symmetric objective function is not affected by variables and constraints, which improves the robustness of the registration method. Therefore, the registration algorithm based on the symmetric objective function optimization has both high robustness and fast convergence speed, and is suitable as a registration and correction algorithm for point clouds of river dams.

[0041] The design process of the symmetric objective function is as follows:

[0042] First, design a symmetric function. The expression of a symmetric function is:

[0043] (6);

[0044] In the above formula, and These are the source point cloud and the target point cloud, respectively. Dot product of vectors; and These are the normal vectors of the source point cloud and the target point cloud, respectively;

[0045] From formula (6), we can obtain that as long as and When the normal vectors of both points coincide with a certain cylinder, the value of the formula becomes zero. Based on this property, and using the designed symmetric function, a dual transformation strategy is employed to transform the point cloud. and To process, that is, to and By applying an inverse rigid body transformation, a symmetric objective function is constructed. Represented as:

[0046] (7);

[0047] (8);

[0048] In the above formula, To compute the optimization function, data that can accelerate the convergence of the objective function; It is the base of the natural logarithm; This represents the number of neighborhood points within the sampling radius.

[0049] This invention uses root mean square error (RMSE) and chamfer distance (CD) as evaluation indicators for registration results.

[0050] Point cloud data acquired by a terrestrial 3D laser scanner is inherently uncertain. Changes in scanning angle and distance can lead to different positional states in the point cloud data. This uncertainty in point cloud position means that simple point-to-point (C2C, Cloud to Cloud) distance calculation models cannot adequately represent the displacement changes of the two phases of the dam. Furthermore, point cloud roughness also has a certain impact on detecting dam deformation. In practical engineering, point cloud roughness manifests as follows: even if the actual object has a smooth convex surface, the scanning results may contain concave areas. Due to the presence of roughness, directly comparing two sets of point cloud data without data processing will result in inaccurate estimations of dam deformation. Therefore, this invention constructs a reasonable confidence interval to eliminate the influence of roughness on dam deformation detection and further designs a multi-scale dam surface deformation algorithm based on the point cloud roughness confidence interval. The specific calculation process is as follows:

[0051] Step S41: Calculation of point cloud roughness;

[0052] Based on statistical analysis of point-to-area distance, for each point in the point cloud... The point cloud roughness is solved by fitting a quadratic surface using points within a local neighborhood, calculating the distance from each point in the neighborhood to the fitted surface, and using the distance from each point in the neighborhood to the surface and the average distance from all points to the surface.

[0053] Step S42: Determining the roughness confidence interval;

[0054] During point cloud data acquisition, due to the influence of the instrument itself and the on-site environment, the point cloud data obtained by the terrestrial 3D laser scanner has a certain degree of surface roughness, which will cause measurement errors in the measured point cloud data. If the measured point cloud data is directly used in deformation calculation, the calculation results will have certain errors and low reliability. To address this, this invention calculates the confidence interval of point cloud roughness, which can effectively avoid the influence of point cloud roughness on deformation calculation and improve the reliability of deformation calculation. Commonly used confidence levels are 90%, 95%, and 99%. This invention adopts a 95% confidence level, which can well meet the accuracy requirements of the method. A 95% confidence level means that the overall mean distribution of point cloud roughness within the confidence interval is 95% reliable.

[0055] A roughness confidence interval construction method based on Gaussian distribution is adopted. For two sets of point cloud data, a KD-tree is constructed to search for local neighborhoods, and a quadratic plane is constructed within the local neighborhoods to calculate the roughness of the two sets of point cloud data. and , and It conforms to two independent Gaussian distributions and has and The roughness confidence interval is obtained from two different variance estimates. The calculation formula is expressed as:

[0056] (9);

[0057] In the above formula, The confidence interval is at a 95% confidence level. and The number of points in the two point clouds; and The roughness of the two sets of point cloud data; and 1.96 represents the variance of the point cloud roughness; 1.96 is the standard score of the Gaussian distribution at the 95% confidence level.

[0058] Based on the roughness confidence interval mentioned above, the deformation within the confidence interval may not be the actual deformation of the dam, but rather a measurement error caused by the instrument itself and environmental noise interference. Therefore, the deformation within this confidence interval should not be included in the subsequent calculation of the deformation.

[0059] Furthermore, the calculation process for the point cloud roughness in step S41 is as follows:

[0060] Step S411: Local neighborhood search;

[0061] For each point in the point cloud KD-tree is used to construct a data topology to find points in their local neighborhood;

[0062] Step S412: Quadratic surface fitting;

[0063] The parametric equations of the quadratic surface are established by using neighborhood points. The general form of the parametric equations of the quadratic surface is:

[0064] (10);

[0065] In the above formula, arrive These are the coefficients that need to be estimated; , , Let these be the coordinates of the point;

[0066] The local quadratic surface is fitted using the least squares method, and the objective function is established as follows:

[0067] (11);

[0068] By combining equations (10) and (11), taking the partial derivatives with respect to each coefficient, and setting the partial derivatives to zero, we obtain the system of linear equations:

[0069] (12);

[0070] Solving the system of linear equations yields the equation of the locally quadratic fitted surface.

[0071] Step S413: Distance calculation;

[0072] For each point within the local domain Based on the local quadratic fitting surface equation obtained in step S412, calculate its distance to the quadratic fitting surface. :

[0073] (13);

[0074] Step S414: Calculate the point cloud roughness;

[0075] The local roughness of a point is obtained by squaring and summing the distance differences between all neighborhood points, and then dividing by the number of neighborhood points. The calculation formula is as follows:

[0076] (14);

[0077] In the above formula, It is the first The distance from each point to the quadratic surface; N is the average distance from all points to the surface, and N is the number of points in the local neighborhood.

[0078] Compared with the prior art, the present invention has the following beneficial effects:

[0079] 1. This invention is based on point cloud data obtained through terrestrial 3D laser scanning technology. It first employs a technique to extract accurate control points from non-deformable, stable regions within the overall environment. This achieves the precise identification of key reference information from a large amount of point cloud data, indirectly reducing invalid computational objects in subsequent data processing. Compared to existing methods that indiscriminately process complete point clouds, resulting in high computational consumption, this invention performs preliminary data screening through control point extraction, avoiding redundant points consuming system computing power, improving detection efficiency, and providing a feasible path for the efficient processing of point clouds from large-scale river dams.

[0080] 2. This invention, by optimizing the objective function of the registration correction algorithm based on control point feature input, corrects the spatial position of the dam, thereby reducing the difficulty of registering large-scale point cloud data and improving the consistency of spatial coordinates across multiple point cloud periods. Compared to existing methods with fixed objective functions in their registration algorithms, which suffer from poor adaptability to large-scale dam point clouds and are prone to registration deviations, this invention deeply integrates control point features with the registration algorithm, making the algorithm more suitable for the registration requirements of large-scale dam data, reducing spatial coordinate deviations, and providing a spatial benchmark for the accuracy of subsequent deformation detection.

[0081] 3. This invention utilizes the roughness distribution patterns of the point cloud of various components and the overall structure of the dam to establish a 95% confidence interval for point cloud roughness. Based on multi-period prototype observation data, it employs a technical method to accurately detect the actual deformation of the dam. This achieves the technical effect of reducing the deviation in distance calculation from unstructured point clouds and accurately locating the deformation range of the dam. Compared to existing methods that rely on direct calculation from unstructured point clouds, resulting in large deformation analysis errors and an inability to clearly define the specific deformation range, this invention establishes a scientific deformation judgment standard through point cloud roughness patterns and confidence intervals. By combining multi-period data to eliminate random error interference, it not only improves the accuracy of deformation detection but also clearly defines the deformation area, providing precise data support for the safety assessment of the dam. Attached Figure Description

[0082] Figure 1 This is a flowchart of a deformation detection method based on three-dimensional point cloud data of a river dam according to the present invention;

[0083] Figure 2 This is a flowchart illustrating the process of extracting control points in the stable zone of the dam according to the present invention.

[0084] Figure 3 These are field measurement diagrams of each measuring point in the embodiments of the present invention;

[0085] Figure 4 This is a visual data diagram of different monitoring stations at the dam in an embodiment of the present invention;

[0086] Figure 5 This is a schematic diagram of the contour feature extraction results of the river dam in an embodiment of the present invention;

[0087] Figure 6 This is a comparison diagram of the contour extraction results of the river-blocking gate components in an embodiment of the present invention;

[0088] Figure 7 This is a roughness gradient map of the point cloud of a prototype river dam in an embodiment of the present invention;

[0089] Figure 8 This is a point cloud roughness distribution map of a prototype river dam in an embodiment of the present invention;

[0090] Figure 9 This is a schematic diagram of the data registration results for the two phases of the dam in an embodiment of the present invention;

[0091] Figure 10 This is a deformation cloud diagram of a prototype river-blocking gate in the direction of water flow in an embodiment of the present invention;

[0092] Figure 11 This is a deformation distribution diagram of a prototype river-blocking gate in the direction of water flow in an embodiment of the present invention;

[0093] Figure 12 This is a deformation cloud diagram of a prototype river-blocking gate working bridge in the direction of water flow in an embodiment of the present invention;

[0094] Figure 13 This is a deformation distribution diagram of a prototype river-blocking gate working bridge in the direction of water flow in an embodiment of the present invention;

[0095] Figure 14 This is a deformation cloud diagram of a prototype river-blocking gate traffic bridge and gate pier in the direction of water flow in an embodiment of the present invention;

[0096] Figure 15 This is a distribution diagram of the deformation of a prototype river-blocking gate traffic bridge and gate piers along the direction of water flow in an embodiment of the present invention. Detailed Implementation

[0097] To facilitate understanding and implementation of the present invention by those skilled in the art, the various steps of the method proposed in this invention are described in detail below. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various modifications or alterations to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0098] Example 1

[0099] like Figure 1 As shown, this invention discloses a deformation detection method based on three-dimensional point cloud data of a river dam, comprising the following steps:

[0100] Step S1: Set up monitoring stations for the dam and use a ground-based 3D laser scanner to scan the surface of the dam and obtain point cloud data of the dam.

[0101] Step S2: Extract the outline points of the dam based on the control point extraction method for the stable zone of the dam.

[0102] In the point cloud data of the control point of the stable area, the point group with characteristic representation is selected by the method of filling the normal vector grid with angle. The threshold points are then selected by the sum vector to further extract the registration points covering the entire dam area, so as to provide accurate feature information for subsequent model registration.

[0103] Step S3: Register the point cloud data of the two phases of the river dam using a registration algorithm based on symmetric objective function optimization;

[0104] The multi-phase point cloud obtained by the point-surface 3D laser scanning is relatively random in spatial location. The contour control points of the dam are usually not in the same coordinate system. In order to achieve deformation detection, the point cloud data of the two phases need to be registered to the same coordinate system. Based on the traditional ICP algorithm, a new symmetric objective function is designed. The zero set of the symmetric objective function is not affected by variables and constraints, and has higher robustness and faster convergence speed.

[0105] Step S4: Calculate the deformation amount between the registered point clouds of the two phases of the river dam surface based on the multi-scale barrage surface deformation algorithm of the point cloud roughness confidence interval.

[0106] Point cloud data acquired by a terrestrial 3D laser scanner is inherently uncertain. Changes in scanning angle and distance can lead to different positional states in the point cloud data. Due to the uncertainty of point cloud position, a simple point-to-point distance calculation model cannot accurately represent the displacement changes of the two phases of the dam. Therefore, a multi-scale dam surface deformation algorithm based on the point cloud roughness confidence interval is designed to eliminate the influence of roughness on dam deformation detection by constructing a reasonable confidence interval.

[0107] Specifically, such as Figure 2 As shown, the specific details of the method for extracting control points in the stable zone of the dam in step S2 are as follows:

[0108] Step S21: Based on the point cloud data of the dam obtained in step S1, calculate the value of each point in the point cloud. normal vector ;

[0109] Step S22: Calculate each point The normal vector histogram, i.e., neighborhood points normal vector Compared to Histogram, neighborhood points Points within the radius R of the domain are called domain points. The normal vector is , ∈ ; and The angular relationship between them is expressed as:

[0110] (1);

[0111] In the above formula, Represents the cross product; Represents the dot product between two vectors;

[0112] Step S23: Divide the histogram into 18 squares, using values ​​that are always non-negative. Angle-filled histogram has 18 squares, each square being 10° in size. After loading the angles of all neighborhood normals into the squares, the count of neighborhood normals falling into a specific square is obtained. The histogram is then normalized to make it independent of neighborhood density. At this point, points with gentle features will cluster in the first square.

[0113] Step S24: Calculate the kurtosis of the histogram. :

[0114] (2);

[0115] (3);

[0116] In the above formula, Indicates the kurtosis of the histogram; Usually, 18 is chosen, which is the number of squares in the histogram; Indicates the first... Counting of squares; This represents the average value of all squares in the histogram. This represents the standard deviation of the histogram grid.

[0117] Through calculation It can remove flat and less curved areas to perform preliminary screening of the surface contour of the dam.

[0118] Step S25: Calculate each point and within its sampling radius The vector differences between each neighborhood point are weighted using a Gaussian weighting function. The weighted vector differences are then summed and divided by the number of sample points. To obtain the average vector difference, we then calculate the L2 norm of the average vector difference for each point. Assign a sum vector value By setting appropriate It can further filter the contours of the rock mass surface, thereby extracting the desired contours, among which... The calculation formula is:

[0119] (4);

[0120] In the above formula, Represents the Gaussian weighting function; Indicates the sampling method performed within the domain; This indicates the number of neighborhood points within the sampling radius.

[0121] In this embodiment, two evaluation indicators are used. and To evaluate the performance of feature extraction methods, among which This represents the accuracy of feature extraction. This represents the error rate of feature extraction. and The calculation method is as follows:

[0122] ;

[0123] ;

[0124] In the above formula, This represents the number of correctly extracted feature points out of all extracted feature points. This represents the total number of feature points in the reference point cloud; This indicates the number of feature points that were incorrectly extracted.

[0125] Specifically, the registration algorithm based on symmetric objective function optimization in step S3 is as follows:

[0126] The objective function of the traditional point-to-surface ICP algorithm Represented as:

[0127] (5);

[0128] In the above formula, It is a rotation matrix; It is a translation vector; Source point cloud; For the target point cloud; Let be the normal vector of a point in the target point cloud;

[0129] The optimal registration solution of the traditional point-to-surface ICP algorithm can only achieve theoretical zero residual under specific geometric conditions. That is, the point cloud to be registered satisfies the geometric flatness constraint in the local neighborhood. Since the normal vector field of the flat surface region has a uniform distribution characteristic, the degrees of freedom of the rigid body transformation parameters in the tangent plane direction will not introduce additional registration errors. Only then can the residual reach a strictly zero value when the algorithm iteratively converges to the global optimal solution.

[0130] The zero set of the symmetric objective function defines the unconstrained property of the transformation, that is, when the surface slides along its tangent plane, its geometric features are fixed. Based on this, a new symmetric objective function is designed to optimize the point-to-surface ICP algorithm. The zero set of this symmetric objective function is not affected by variables and constraints, which improves the robustness of the registration method. Therefore, the registration algorithm based on the symmetric objective function optimization has both high robustness and fast convergence speed, and is suitable as a registration and correction algorithm for point clouds of river dams.

[0131] The design process of the symmetric objective function is as follows:

[0132] First, design a symmetric function. The expression of a symmetric function is:

[0133] (6);

[0134] In the above formula, and These are the source point cloud and the target point cloud, respectively. Dot product of vectors; and These are the normal vectors of the source point cloud and the target point cloud, respectively;

[0135] From formula (6), we can obtain that as long as and When the normal vectors of both points coincide with a certain cylinder, the value of the formula becomes zero. Based on this property, and using the designed symmetric function, a dual transformation strategy is employed to transform the point cloud. and To process, that is, to and By applying an inverse rigid body transformation, a symmetric objective function is constructed. Represented as:

[0136] (7);

[0137] (8);

[0138] In the above formula, To compute the optimization function, data that can accelerate the convergence of the objective function; It is the base of the natural logarithm; This represents the number of neighborhood points within the sampling radius.

[0139] In this embodiment, root mean square error (RMSE) and chamfer distance (CD) are used as evaluation metrics for the registration results.

[0140] Point cloud data acquired by a terrestrial 3D laser scanner is inherently uncertain. Changes in scanning angle and distance can lead to different positional states in the point cloud data. This uncertainty in point cloud position means that simple point-to-point (C2C, Cloud to Cloud) distance calculation models cannot adequately represent the displacement changes of the two phases of the dam. Furthermore, point cloud roughness also has a certain impact on detecting dam deformation. In practical engineering, point cloud roughness manifests as follows: even if the actual object has a smooth convex surface, the scanning results may contain concave areas. Due to the presence of roughness, directly comparing two sets of point cloud data without data processing will result in inaccurate estimations of dam deformation. Therefore, this embodiment constructs a reasonable confidence interval to eliminate the influence of roughness on dam deformation detection and further designs a multi-scale dam surface deformation algorithm based on the point cloud roughness confidence interval. The specific calculation process is as follows:

[0141] Step S41: Calculation of point cloud roughness;

[0142] Based on statistical analysis of point-to-area distance, for each point in the point cloud... The point cloud roughness is solved by fitting a quadratic surface using points within a local neighborhood, calculating the distance from each point in the neighborhood to the fitted surface, and using the distance from each point in the neighborhood to the surface and the average distance from all points to the surface.

[0143] Step S42: Determining the roughness confidence interval;

[0144] During point cloud data acquisition, due to the influence of the instrument itself and the on-site environment, the point cloud data obtained by the terrestrial 3D laser scanner has a certain degree of surface roughness, which will cause measurement errors in the measured point cloud data. If the measured point cloud data is directly used in deformation calculation, the calculation results will have certain errors and low reliability. To address this, this invention calculates the confidence interval of point cloud roughness, which can effectively avoid the influence of point cloud roughness on deformation calculation and improve the reliability of deformation calculation. Commonly used confidence levels are 90%, 95%, and 99%. This invention adopts a 95% confidence level, which can well meet the accuracy requirements of the method. A 95% confidence level means that the overall mean distribution of point cloud roughness within the confidence interval is 95% reliable.

[0145] A roughness confidence interval construction method based on Gaussian distribution is adopted. For two sets of point cloud data, a KD-tree is constructed to search for local neighborhoods, and a quadratic plane is constructed within the local neighborhoods to calculate the roughness of the two sets of point cloud data. and , and It conforms to two independent Gaussian distributions and has and The roughness confidence interval is obtained from two different variance estimates. The calculation formula is expressed as:

[0146] (9);

[0147] In the above formula, The confidence interval is at a 95% confidence level. and The number of points in the two point clouds; and The roughness of the two sets of point cloud data; and 1.96 represents the variance of the point cloud roughness; 1.96 is the standard score of the Gaussian distribution at the 95% confidence level.

[0148] Based on the roughness confidence interval mentioned above, the deformation within the confidence interval may not be the actual deformation of the dam, but rather a measurement error caused by the instrument itself and environmental noise interference. Therefore, the deformation within this confidence interval should not be included in the subsequent calculation of the deformation.

[0149] Furthermore, the calculation process for the point cloud roughness in step S41 is as follows:

[0150] Step S411: Local neighborhood search;

[0151] For each point in the point cloud KD-tree is used to construct a data topology to find points in their local neighborhood;

[0152] Step S412: Quadratic surface fitting;

[0153] The parametric equations of the quadratic surface are established by using neighborhood points. The general form of the parametric equations of the quadratic surface is:

[0154] (10);

[0155] In the above formula, arrive These are the coefficients that need to be estimated; , , Let these be the coordinates of the point;

[0156] The local quadratic surface is fitted using the least squares method, and the objective function is established as follows:

[0157] (11);

[0158] By combining equations (10) and (11), taking the partial derivatives with respect to each coefficient, and setting the partial derivatives to zero, we obtain the system of linear equations:

[0159] (12);

[0160] Solving the system of linear equations yields the equation of the locally quadratic fitted surface.

[0161] Step S413: Distance calculation;

[0162] For each point within the local domain Based on the local quadratic fitting surface equation obtained in step S412, calculate its distance to the quadratic fitting surface. :

[0163] (13);

[0164] Step S414: Calculate the point cloud roughness;

[0165] The local roughness of a point is obtained by squaring and summing the distance differences between all neighborhood points, and then dividing by the number of neighborhood points. The calculation formula is as follows:

[0166] (14);

[0167] In the above formula, It is the first The distance from each point to the quadratic surface; N is the average distance from all points to the surface, and N is the number of points in the local neighborhood.

[0168] The technical effects of the method of the present invention will be further explained below through a deformation detection example of a prototype river dam project.

[0169] Brief introduction to the prototype project of the river dam;

[0170] This example uses a Leica P50 terrestrial 3D laser scanner to test a prototype river-blocking gate in Jiangxi Province. The gate has seven spans, each with a net width of 12 meters. Figure 3 As shown in (a)-(g), this survey deployed 7 monitoring stations on the dam crest road and the river-blocking gate traffic bridge, as follows: Figure 3As shown in (h)-(i), two measuring stations are arranged near the right bank of the dam. The location of each measuring point is shown in Table 1 below, and the visualization data is as follows. Figure 4 As shown.

[0171] Table 1. Specific locations of each monitoring station at the dam.

[0172] ;

[0173] I. Feature contour extraction results;

[0174] In this example, the overall structure of the dam is divided into six parts: dam piers, gate, bank slope, working bridge, traffic bridge, and power plant. The dam piers and gate only appear at monitoring stations 1, 8, and 9. The power plant is located on the left bank slope, and this part only appears at a few monitoring stations. While the traffic bridge exists at most monitoring stations, its longitudinal cross-section is concave, resulting in limited scanning of the coplanar portion of the target object's inner and outer sides at both external and internal monitoring stations. Ultimately, this example uses point cloud contour extraction to select the working bridge and bank slope as the dam's feature components when applying the dam's point cloud contour extraction method. Figure 5 The image shows the result of extracting the feature contour of the river dam (gray points are non-feature points, and other points are contour feature points). Figure 5 (a) represents the working bridge point cloud data. Figure 5 (b) shows the point cloud data of the bank slope.

[0175] In this example, the method proposed in this invention is compared with point cloud contour feature extraction methods based on surface curvature, normal vectors, and density. The comparison results are as follows: Figure 6 As shown. By Figure 6 It can be seen that the method proposed in this invention removes more non-contour information and preserves more complete contour information; furthermore, this example uses two evaluation indicators. and The accuracy of feature extraction was evaluated, and the calculation results are shown in Table 2 below.

[0176] Table 2. Accuracy of feature line extraction using different methods from the point clouds of the two river dams.

[0177] ;

[0178] II. Confidence interval results based on roughness;

[0179] 1. Roughness calculation;

[0180] This invention performs point cloud roughness calculations on four main components of a river dam (working bridge, traffic bridge, dam piers, and power plant), such as... Figure 7As shown, the roughness points of the working bridge and traffic bridge are mainly concentrated on the guardrails and auxiliary structures at the edges of buildings; the overall roughness of the power plant building is relatively small, with larger roughness values ​​concentrated on the roof and the monitoring equipment on the sides; the roughness distribution of the gates and gate piers is relatively uniform; when calculating the overall roughness, some small components are ignored, and the roughness gradients of each component are as follows: Figure 7 As shown in (a)-(d) (representing the working bridge, traffic bridge, power plant, and gate pier in that order), the roughness distribution histogram is as follows: Figure 8 As shown in Table 3, the average roughness is as follows.

[0181] Table 3. Average roughness of point cloud for a prototype dam.

[0182] ;

[0183] Depend on Figure 8 It can be seen that the roughness variation of the traffic bridge is relatively small, ranging from 0.003mm to 0.580mm, with the larger roughness point clouds mainly concentrated in the guardrails. The roughness variation of the gate piers is larger, ranging from 0.007mm to 1.000mm, mainly because the gate piers are in contact with water flow for a long time, and are often interfered with by the water flow during measurement. The high roughness values ​​of the power plant are concentrated on the monitoring equipment on the roof, while the high roughness values ​​of the working bridge are concentrated on the carvings of the gatehouse and the edges of the doors and windows.

[0184] 2. Establish confidence intervals;

[0185] like Figure 8 The figure shows the point cloud roughness distribution of the prototype dam. The roughness distribution patterns of the working bridge, traffic bridge, and dam piers are almost identical, with 95% of the points concentrated near the middle of the roughness range. This indicates that the overall roughness is relatively concentrated, with only some small components affecting the roughness distribution. Table 3 shows the average roughness of each component of the dam. The traffic bridge has the lowest average roughness at 0.243 mm, while the average roughness of other components is at the same level. Based on the roughness of each component of the dam, the roughness confidence interval for each component was calculated, and the results are shown in Table 4 below.

[0186] Table 4. Roughness confidence intervals for various components of the river dam.

[0187] ;

[0188] III. Data registration results for the two phases of the river dam;

[0189] The registration algorithm based on symmetric objective function optimization designed in this invention is used to register two periods of river dam data. The root mean square error and chamfer distance are then used to evaluate the registration results of the two periods of river dam data. Figure 9 As shown in Table 5, Figure 9(a) is the upstream view of the dam, and (b) is the downstream view.

[0190] Table 5. Evaluation Table of Data Registration Accuracy for Two Phases of the Dam

[0191] ;

[0192] IV. Results of Deformation Testing of the Barrier Gate;

[0193] The prototype dam data used in this example were collected in May 2024 and December 2024, respectively. The upstream water level difference reached 1.24m, and the downstream water level difference reached 6.34m. This example uses the May 2024 data as a reference and compares it with the December 2024 data to detect the longitudinal deformation of the dam. The overall deformation cloud map and distribution map of the dam are shown below. Figure 10 and Figure 11 As shown in the figure. Furthermore, to provide a more comprehensive analysis and demonstration of the deformation of the dam, this example also includes deformation detection and analysis of the main components (the working bridge and the traffic bridge). The deformation cloud maps and deformation distribution diagrams of the main components are shown below. Figures 11-14 As shown.

[0194] from Figure 10 The spatial distribution characteristics of the river-direction deformation of the dam can be observed: the largest deformation is 2.52 mm, occurring at the top of the working bridge. The deformation shows a trend of increasing deformation with increasing elevation, and the deformation of the superstructure is greater than that of the lower structure. This is because the dam piers are fixedly connected to the bedrock of the riverbed, thus constraining the deformation at the bottom of the piers. Figure 11 It can be seen that the overall average deformation of the dam is 1.35 mm. This average deformation data is consistent with the micro-deformation magnitude of the dam under actual service conditions, verifying that the method of the present invention can capture the subtle deformation characteristics of the dam that conform to the laws of structural mechanics.

[0195] Depend on Figure 11 and Figure 13 It can be seen that the deformation range of the working bridge is larger than that of the traffic bridge and the gate pier, with deformation ranges of 1.11mm-2.52mm and 0mm-1.15mm respectively. The deformation trends of the working bridge, traffic bridge, and gate pier are quite similar, all showing a trend of greater deformation with increasing elevation. These deformation characteristics are consistent with... Figure 10 The overall deformation characteristics of the dam are consistent with those of the main dam. It is worth noting that the maximum deformation of the traffic bridge and dam piers is 1.15 mm, slightly larger than the minimum deformation of the working bridge (1.11 mm). This is because the railings on both sides of the traffic bridge are slightly higher than the bottom of the working bridge. From Figure 12 and Figure 14It can be seen that the average deformation of the working bridge (1.77 mm) is greater than that of the traffic bridge and the gate pier (0.72 mm). This indicates that the method of the present invention can not only accurately measure the deformation of a single structure, but also accurately quantify the deformation differences between different structures, further verifying the accuracy and reliability of the method.

[0196] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A deformation detection method based on three-dimensional point cloud data of a river dam, characterized in that, Includes the following steps: Step S1: Set up monitoring stations for the dam and use a ground-based 3D laser scanner to scan the surface of the dam and obtain point cloud data of the dam. Step S2: Extract the outline points of the dam based on the control point extraction method for the stable zone of the dam. In the point cloud data of the control point of the stable area, the point group with characteristic representation is selected by the method of filling the normal vector grid with angle. The threshold points are then selected by the sum vector to further extract the registration points covering the entire dam area, so as to provide accurate feature information for subsequent model registration. Step S3: Register the point cloud data of the two phases of the river dam using a registration algorithm based on symmetric objective function optimization; The multi-phase point cloud obtained by the point-surface 3D laser scanning is relatively random in spatial location. The contour control points of the dam are usually not in the same coordinate system. In order to achieve deformation detection, the point cloud data of the two phases need to be registered to the same coordinate system. Based on the traditional ICP algorithm, a new symmetric objective function is designed. The zero set of the symmetric objective function is not affected by variables and constraints, and has higher robustness and faster convergence speed. Step S4: Calculate the deformation amount between the registered point clouds of the two phases of the river dam surface based on the multi-scale barrage surface deformation algorithm of the point cloud roughness confidence interval. Point cloud data acquired by a terrestrial 3D laser scanner is inherently uncertain. Changes in scanning angle and distance can lead to different positional states in the point cloud data. Due to the uncertainty of point cloud position, a simple point-to-point distance calculation model cannot accurately represent the displacement changes of the two phases of the dam. Therefore, a multi-scale dam surface deformation algorithm based on the point cloud roughness confidence interval is designed to eliminate the influence of roughness on dam deformation detection by constructing a reasonable confidence interval.

2. The deformation detection method based on three-dimensional point cloud data of a river dam according to claim 1, characterized in that, The specific details of the method for extracting control points in the stable zone of the dam in step S2 are as follows: Step S21: Based on the point cloud data of the dam obtained in step S1, calculate the value of each point in the point cloud. normal vector ; Step S22: Calculate each point The normal vector histogram, i.e., neighborhood points normal vector Compared to Histogram, neighborhood points Points within the radius R of the domain are called domain points. The normal vector is , ∈ ; and The angular relationship between them is expressed as: （1）； In the above formula, Represents the cross product; Represents the dot product between two vectors; Step S23: Divide the histogram into 18 squares, using values ​​that are always non-negative. Angle-filled histogram has 18 squares, each square being 10° in size. After loading the angles of all neighborhood normals into the squares, the count of neighborhood normals falling into a specific square is obtained. The histogram is then normalized to make it independent of neighborhood density. At this point, points with gentle features will cluster in the first square. Step S24: Calculate the kurtosis of the histogram. : （2）； （3）； In the above formula, Indicates the kurtosis of the histogram; Usually, 18 is chosen, which is the number of squares in the histogram; Indicates the first... Counting of squares; This represents the average value of all squares in the histogram. This represents the standard deviation of the histogram grid. Through calculation It can remove flat and less curved areas to perform preliminary screening of the surface contour of the dam. Step S25: Calculate each point and within its sampling radius The vector differences between each neighborhood point are weighted using a Gaussian weighting function. The weighted vector differences are then summed and divided by the number of sample points. To obtain the average vector difference, we then calculate the L2 norm of the average vector difference for each point. Assign a sum vector value By setting appropriate It can further filter the contours of the rock mass surface, thereby extracting the desired contours, among which... The calculation formula is: （4）； In the above formula, Represents the Gaussian weighting function; This indicates the sampling method performed within the domain; This indicates the number of neighborhood points within the sampling radius.

3. The deformation detection method based on three-dimensional point cloud data of a river dam according to claim 2, characterized in that, The specific details of the registration algorithm based on symmetric objective function optimization in step S3 are as follows: The objective function of the traditional point-to-surface ICP algorithm Represented as: （5）； In the above formula, It is a rotation matrix; It is a translation vector; Source point cloud; For the target point cloud; Let be the normal vector of a point in the target point cloud; Since the null set of the symmetric objective function defines the unconstrained property of the transformation, that is, when the surface slides along its tangent plane, its geometric characteristics are fixed, a new symmetric objective function is designed to optimize the point-to-surface ICP algorithm. The null set of this symmetric objective function is not affected by variables and constraints. The design process of the symmetric objective function is as follows: First, design a symmetric function. The expression of a symmetric function is: （6）； In the above formula, and These are the source point cloud and the target point cloud, respectively. Dot product of vectors; and These are the normal vectors of the source point cloud and the target point cloud, respectively; From formula (6), we can obtain that as long as and When the normal vectors of both points coincide with a certain cylinder, the value of the formula becomes zero. Based on this property, and using the designed symmetric function, a dual transformation strategy is employed to transform the point cloud. and To process, that is, to and By applying an inverse rigid body transformation, a symmetric objective function is constructed. Represented as: （7）； （8）； In the above formula, To compute the optimization function, data that can accelerate the convergence of the objective function; It is the base of the natural logarithm; This represents the number of neighborhood points within the sampling radius.

4. The deformation detection method based on three-dimensional point cloud data of a river dam according to claim 3, characterized in that, The multi-scale surface deformation algorithm for the river dam based on the point cloud roughness confidence interval in step S4 is calculated as follows: Step S41: Calculation of point cloud roughness; Based on statistical analysis of point-to-area distance, for each point in the point cloud... The point cloud roughness is solved by fitting a quadratic surface using points within a local neighborhood, calculating the distance from each point in the neighborhood to the fitted surface, and using the distance from each point in the neighborhood to the surface and the average distance from all points to the surface. Step S42: Determining the roughness confidence interval; A roughness confidence interval construction method based on Gaussian distribution is adopted. For two sets of point cloud data, a KD-tree is constructed to search for local neighborhoods, and a quadratic plane is constructed within the local neighborhoods to calculate the roughness of the two sets of point cloud data. and , and It conforms to two independent Gaussian distributions and has and The roughness confidence interval is obtained from two different variance estimates. The calculation formula is expressed as: （9）； In the above formula, The confidence interval is at a 95% confidence level. and The number of points in the two point clouds; and The roughness of the two sets of point cloud data; and 1.96 represents the variance of the point cloud roughness; 1.96 is the standard score of the Gaussian distribution at the 95% confidence level.

5. The deformation detection method based on three-dimensional point cloud data of a river dam according to claim 4, characterized in that, The calculation process for point cloud roughness in step S41 is as follows: Step S411: Local neighborhood search; For each point in the point cloud KD-tree is used to construct a data topology to find points in their local neighborhood; Step S412: Quadratic surface fitting; The parametric equations of the quadratic surface are established by using neighborhood points. The general form of the parametric equations of the quadratic surface is: （10）； In the above formula, arrive These are the coefficients that need to be estimated; , , Let these be the coordinates of the point; The local quadratic surface is fitted using the least squares method, and the objective function is established as follows: （11）； By combining equations (10) and (11), taking the partial derivatives with respect to each coefficient, and setting the partial derivatives to zero, we obtain the system of linear equations: （12）； Solving the system of linear equations yields the equation of the locally quadratic fitted surface. Step S413: Distance calculation; For each point within the local domain Based on the local quadratic fitting surface equation obtained in step S412, calculate its distance to the quadratic fitting surface. : （13）； Step S414: Calculate the point cloud roughness; The local roughness of a point is obtained by squaring and summing the distance differences between all neighborhood points, and then dividing by the number of neighborhood points. The calculation formula is as follows: （14）； In the above formula, It is the first The distance from each point to the quadratic surface; N is the average distance from all points to the surface, and N is the number of points in the local neighborhood.

Images