A method for reconstructing the morphological characteristics of the bridge scour surface based on 3D sonar point cloud

Abstract

The present invention discloses a method for reconstructing the morphological characteristics of a bridge scour surface based on three-dimensional sonar point cloud, which includes the following steps: collecting the point cloud data of the three-dimensional morphology of the underwater riverbed near the bridge foundation to obtain the original point cloud data of the complete morphology of the underwater riverbed; performing preprocessing of denoising and spherical rotation reconstruction on the collected original point cloud data to complete the preliminary reconstruction of the point cloud; based on the computer vision principle, performing scour morphology recognition of point cloud binarization according to the preliminarily reconstructed point cloud data, calculating the scour pit element classification values of all point clouds, and identifying the scour pit point cloud; according to the identified scour pit point cloud, using the K-mean clustering algorithm for clustering and calculating the maximum depth of each pier scour pit; according to the obtained scour pit element classification values of all point clouds, proposing a surface reconstruction method based on scour recognition to perform high-precision surface reconstruction on the scour pit. The present invention can achieve high-precision reconstruction of the scour pit morphology.

Description

Technical Field

[0001] The present invention belongs to the technical field of bridge underwater scouring morphology recognition, and particularly relates to a method for reconstructing the morphological characteristics of a bridge scouring curved surface based on three-dimensional sonar point cloud. Background Art

[0002] Scouring is the process by which water erodes sediment and other substances around structures such as riverbeds, banks, and bridge foundations. It mainly manifests as natural evolution scouring, general scouring, and local scouring. The bridge pier will block the water flow, causing high-speed water flow to form on both sides of the bridge pier, resulting in a sharp deformation of the riverbed around the bridge pier and forming a local scouring pit. Local scouring will erode the soil around the bridge foundation and hollow it out, causing the bearing capacity of the bridge foundation to decline. In severe cases, it will lead to the collapse of the bridge. Currently, scouring is gradually becoming the dominant cause of bridge collapse worldwide, and more than half of the bridge failures are caused by scouring.

[0003] Monitoring scouring during the operation period of a bridge helps to reasonably formulate the bridge operation plan and emergency rescue engineering measures during the flood period, and is the most effective method to reduce the damage of bridges due to excessive scouring of the foundation. In recent years, with the maturity of sonar technology in underwater applications, although certain achievements have been made in using sonar technology to detect the underwater structure of bridges, currently, for scouring detection based on three-dimensional sonar point cloud, it mainly relies on drawing contour lines and then depends on experienced technicians to separately segment the scouring pit point cloud of each bridge pier one by one to obtain the maximum scouring depth for description and interpretation. Moreover, there is a lack of research on the three-dimensional reconstruction of bridge scouring pits. The above situation results in low automation, visualization degree, and efficiency, and the three-dimensional sonar point cloud information is not fully utilized.

[0004] The underwater riverbed point cloud obtained based on three-dimensional sonar technology usually has a complex terrain, sparse point cloud, and many holes. Specifically, the challenges are as follows: on the one hand, the riverbed terrain has high and low undulations, and the scouring pit point cloud of a bridge spanning different water depths cannot be segmented by contour lines; on the other hand, due to the limitations of the acoustic principle of the instrument and the interference of underwater organisms, the scanned point cloud will still have holes and be sparse even after removing noise. However, currently, due to the limitations of the complex underwater detection environment, there is a lack of research on the recognition of the wide-area bridge scouring morphology, and conventional point cloud reconstruction methods cannot be well applied to the high-precision reconstruction of sparse and a large number of underwater scouring pit sonar point clouds.

[0005] Therefore, it is urgent to solve the above problems. Summary of the Invention

[0006] Objective of the Invention: The objective of the present invention is to provide a method for reconstructing the morphological characteristics of a bridge scour surface based on three-dimensional sonar point clouds. This method can perform automatic identification, segmentation, and reconstruction on sonar point clouds for wide-area surveys, has strong robustness, and can provide accurate data support for subsequent bridge underwater foundation maintenance decisions.

[0007] Technical Solution: To achieve the above objective, the present invention discloses a method for reconstructing the morphological characteristics of a bridge scour surface based on three-dimensional sonar point clouds, including the following steps:

[0008] (1) Collect point cloud data of the three-dimensional morphology of the underwater riverbed near the bridge foundation. Use a three-dimensional real-time sonar system carried on a waterborne survey ship to conduct multi-directional surveys around the underwater riverbed of the bridge and perform real-time stitching, thereby obtaining the original point cloud data of the complete morphology of the underwater riverbed;

[0009] (2) Perform denoising and spherical rotation reconstruction preprocessing on the collected original point cloud data to complete the preliminary reconstruction of the point cloud;

[0010] (3) Based on the point cloud data preliminarily reconstructed in step (2), perform scour morphology recognition of point cloud binarization based on the principle of computer vision, calculate the scour pit element classification values of all point clouds, and identify the scour pit point clouds;

[0011] (4) According to the scour pit point clouds identified in step (3), use the K-mean clustering algorithm for clustering and calculate the maximum depth of each pier scour pit;

[0012] (5) According to the scour pit element classification values of all point clouds obtained in step (3), propose a surface reconstruction method based on scour recognition to perform high-precision surface reconstruction on the scour pits.

[0013] Among them, step (2) specifically includes the following steps:

[0014] (2.1) After performing attitude correction processing and noise interference preprocessing on each survey data, perform survey line merging;

[0015] (2.2) Process the obtained water depth and position, and remove the noise interference in the overlapping coverage range of two adjacent survey lines one by one;

[0016] (2.3) Finally, perform preliminary surface reconstruction on the denoised and merged point clouds using the spherical rotation reconstruction method to form a triangular mesh.

[0017] Preferably, in step (2.3), search is first performed by calculating the density upper limit of the sonar point cloud, then a sphere with a specified radius ρ contacts any three points that do not contain other points to form a seed triangle, and the sphere is rotated around the edge until it touches the third point to form a new triangle. Finally, all reachable edges are traversed to complete the preliminary reconstruction of the point cloud and form a triangular mesh.

[0018] Furthermore, step (3) specifically includes the following steps:

[0019] (3.1) Calculate the distances and semi-variances between all point clouds after preliminary reconstruction, and select the fitting Gaussian model as the theoretical variogram to fit the function of semi-variance and distance;

[0020] (3.2) Set the search radius L and the skip radius R, and based on Kriging interpolation, search for interpolation points within the search radius and outside the skip radius in the eight main directions of each point cloud to obtain the interpolation point clouds on the eight elevation profiles;

[0021] (3.3) According to the interpolation point clouds on the eight elevation profiles obtained in step (3.2), on the main direction profile of each point cloud, according to the maximum elevation angle and the minimum elevation angle calculate the upper opening angle and the lower opening angle

[0022]

[0023] (3.4) Set the flatness t that changes with the height of the point cloud, and calculate the elevation identification symbol of each profile according to the upper and lower opening angles obtained in step (3.3)

[0024]

[0025] (3.5) According to the principle of local ternary value pattern, calculate the (+1) symbols in the elevation identification symbols on the 8 elevation profiles of each point cloud, and the (-1) symbols to establish the scour identification search coordinates

[0026] (3.6) According to the scour identification search coordinates obtained in step (3.5) calculate the scour pit element classification value θ of each point cloud;

[0027] (3.7) Calculate the average value θ p of the scour pit element classification values θ of all point clouds as the threshold for segmenting the scour pit point cloud;

[0028] (3.8) For the parameter search radius, skip radius, and flatness within the value range, change them step by step one by one and perform recognition calculation of the recognition accuracy. When the recognition accuracy meets the extraction requirements, perform scour recognition on the underwater riverbed point cloud data of the bridge according to the optimized parameters, and obtain the scour pit point cloud.

[0029] Further, in step (3.1), calculate the distances between all sample points and sort them in ascending order, and then use the Gaussian model to perform function fitting on the relationship between the semivariance value and the distance. The general formula of the Gaussian model is:

[0030]

[0031] where γ represents the semivariance, h represents the distance, C0 is the nugget constant, C is the sill, and a is the range, all of which are parameters that need to be preset when using Kriging interpolation.

[0032] Preferably, in step (3.4), the flatness t varying with height is set to a single straight line model:

[0033]

[0034] where t is the flatness, t min is the minimum flatness, set to 0, t max is the maximum flatness, z max is the highest elevation of the point cloud to be recognized, z min is the lowest elevation of the point cloud to be recognized.

[0035] Furthermore, in step (3.6), the calculated value of the scour pit element classification value θ is defined as follows: First, calculate the scour recognition search coordinates represents the recognition symbols of the eight profiles in the local ternary value pattern, is the number of (+1) symbols among them, is the number of (-1) symbols among them; then calculate the mapped point of the distance from the mapped point of the pit element to (8,0) to define the classification value θ of each point cloud scour pit element, and its calculation formula is as follows:

[0036]

[0037] Further, in step (3.8), the spatial geometric features of the sonar point cloud obtained by scanning include: the minimum average spacing and the elevation coefficient of variation; use a pretest point cloud randomly generated by several Gaussian amplitudes with the same spatial geometric features as the sonar point cloud, and mark it as the scour pit point cloud below the horizontal plane. Perform recognition on the pretest point cloud. If the recognition result satisfies the recall rate R e greater than 0.85, the F1 value greater than 0.89, and the precision Pr Greater than 0.98 and accuracy rate A c If it reaches 0.93, the sonar point cloud of the sweep test is identified by flushing according to this optimized parameter, where the evaluation index is the accuracy rate A c , recall rate R e , precision rate P r And the calculation formulas for the F1 value are as follows:

[0038]

[0039] Among them, TP is the true positive, which refers to the sample point where the target point cloud is the point cloud in the flushing pit and is marked as the point cloud in the flushing pit; FP is the false positive, which refers to the sample point where the target point cloud is the point cloud not in the flushing pit but is identified as the point cloud in the flushing pit; TN is the true negative, which refers to the sample point where the target point cloud is the point cloud not in the flushing pit and is marked as the point cloud not in the flushing pit; FN is the false negative, which refers to the sample point where the target point cloud is the point cloud in the flushing pit but is marked as the point cloud not in the flushing pit.

[0040] Furthermore, step (4) specifically includes the following steps:

[0041] (4.1) According to the flushing pit point cloud identified in step (3), calculate the coordinate mean and variance of the flushing pit point cloud in the transverse bridge direction and perform denoising;

[0042] (4.2) Set the number k of bridge piers within the sonar point cloud range and initialize k samples as the centers a of the initial clustering k ;

[0043] (4.3) For each point cloud p in the sonar point cloud, calculate the distances from the point cloud p to the k clustering centers and assign it to the class corresponding to the clustering center with the minimum distance;

[0044] Calculate the distances from the point cloud p to the k clustering centers using the squared Euclidean distance and assign it to the class corresponding to the clustering center with the minimum distance. The calculation formula for the squared Euclidean distance is as follows:

[0045] d(x,c)=(x - c) 2

[0046] Among them, d represents the distance, x represents the spatial coordinate row vector of the point cloud, and c represents the spatial coordinate row vector of the centroid of the clustering;

[0047] (4.5) Repeat steps (4.2) and (4.3) until the number of iterations reaches the set upper limit to obtain the clustering results of the flushing pit point cloud of each bridge pier;

[0048] (4.6) For the flushing pit point cloud of each clustered bridge pier, calculate its minimum elevation to obtain the maximum depth of each flushing pit of the sonar-swept bridge.

[0049] Preferably, step (5) specifically includes the following steps:

[0050] (5.1) Calculate the incenter of each triangle in the triangular mesh preliminarily reconstructed according to the preliminary spherical rotation algorithm in step (2).

[0051] (5.2) Estimate the normal vector of each point cloud by performing covariance analysis on the neighboring points of the point cloud obtained according to the KNN algorithm.

[0052] Construct a KD-Tree search structure for the scanned point cloud data, search for the points in the local neighborhood of each point cloud, and calculate the centroid p c of the neighborhood. Then calculate the covariance matrix of the point cloud p i as follows:

[0053] M = ∑(p i - p c )(p i - p c ) T

[0054] In the formula, M represents the covariance matrix. Then, according to the singular value decomposition method, obtain the eigenvalues of the covariance matrix, where the smallest eigenvalue represents the normal vector of the point.

[0055] (5.3) Calculate the tangent plane of each point in the point cloud.

[0056] (5.4) Calculate the projection points of the incenter of the triangles adjacent to each point in the point cloud onto the tangent plane of the point. The projection points form an adjustable polygon. Adjust the size of the adjustable polygon by moving the position of the projection points on the straight line between the point and the projection points. The adjustment weight is the scouring pit element classification value θ calculated in step (3).

[0057] (5.5) Update the position of the adjustable triangle, move the vertices of the original triangular mesh to the positions of the projection points, and sequentially traverse all triangles to construct new adjustable triangles.

[0058] (5.6) Divide the space quadrilateral between the adjustable polygon and the adjustable triangle according to the principle of the shortest diagonal to obtain triangularized patches.

[0059] (5.7) Integrate the adjustable triangles, adjustable polygons, and triangularized patches to obtain the finally reconstructed scouring pit surface.

[0060] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages: The present invention uses a sonar system to collect point cloud data of the three-dimensional shape of the underwater riverbed of the bridge. Based on the principle of computer vision and local ternary value pattern, by introducing the scouring pit classification and recognition value for point cloud binarization, the purpose of efficiently and accurately recognizing the shape of the underwater wide-area scouring pit of the bridge based on the three-dimensional digital model is achieved. Based on the binarization recognition result, K-means clustering is performed and the maximum scouring depth of each pier of the bridge is calculated, providing technical support for the non-contact detection of underwater scouring of the bridge based on three-dimensional sonar. At the same time, the present invention improves the density of the reconstructed target point cloud by increasing the number of point clouds, solves the problem of the sparsity of underwater sonar point clouds, combines the information of the local elevation of the point cloud, and the reconstructed surface has better fitting accuracy and smaller error due to combining the local elevation information of the point cloud, can achieve high-precision reconstruction of the scouring pit shape, and provides technical support for the high-precision quantitative processing of underwater scouring of the bridge based on three-dimensional sonar. Description of the Drawings

[0061] Figure 1 is the implementation flowchart of the present invention;

[0062] Figure 2 is the schematic diagram of the local ternary value pattern of the point cloud adopted by the present invention;

[0063] Figure 3 is the schematic diagram of the point cloud binarization transformation of the present invention;

[0064] Figure 4 is the result diagram of realizing the recognition of the scouring pit point cloud shape of the present invention;

[0065] Figure 5 is the result diagram of realizing the reconstruction of the scouring pit shape of the bridge of the present invention. Detailed Embodiment

[0066] The technical solution of the present invention will be further described below in conjunction with the drawings.

[0067] The present invention mainly serves the technical field of the recognition and reconstruction of the underwater scouring shape of the bridge, and the applied object is the field measurement data of the underwater structure of the bridge.

[0068] As Figure 1 shown, a method for reconstructing the morphological characteristics of the scouring surface of a bridge based on three-dimensional sonar point cloud includes the following steps:

[0069] (1) Collect the point cloud data of the three-dimensional shape of the underwater riverbed terrain near the bridge foundation, and use a three-dimensional real-time sonar system to carry a waterborne survey ship to conduct multi-directional scanning and real-time stitching around the underwater riverbed terrain of the bridge, so as to obtain the original point cloud data of the complete shape of the underwater riverbed;

[0070] (2) Preprocess the acquired original point cloud data by denoising and sphere rotation reconstruction to complete the preliminary reconstruction of the point cloud;

[0071] Step (2) specifically includes the following steps:

[0072] (2.1) After performing attitude correction processing and noise interference preprocessing on each survey data, merge the survey lines;

[0073] (2.2) Process the obtained water depth and position, and remove the noise interference in the overlapping coverage range of two adjacent survey lines one by one;

[0074] (2.3) Finally, perform preliminary surface reconstruction on the denoised and merged point cloud using the sphere rotation reconstruction method to form a triangular mesh;

[0075] First, search by calculating the density upper limit of the sonar point cloud, then specify a sphere with a radius ρ to contact any three points that do not contain other points to form a seed triangle, and rotate the sphere around the edge until it touches the third point to form a new triangle. Finally, traverse all reachable edges to complete the preliminary reconstruction of the point cloud and form a triangular mesh;

[0076] (3) Based on the point cloud data after preliminary reconstruction in step (2), perform scour morphology recognition of point cloud binarization based on the principle of computer vision, calculate the scour pit element classification values of all point clouds, and identify the scour pit point cloud;

[0077] As Figure 2 and Figure 3 shown, step (3) specifically includes the following steps:

[0078] (3.1) Calculate the distance and semi-variance between all point clouds after preliminary reconstruction, and select the fitting Gaussian model as the theoretical variogram to fit the function of semi-variance and distance;

[0079] Calculate the distance between all sample points and sort them in ascending order, then use the Gaussian model to perform function fitting on the relationship between semi-variance values and distances. The general formula of the Gaussian model is:

[0080]

[0081] where γ represents the semi-variance, h represents the distance, C0 is the nugget constant, C is the sill, and a is the range, all of which are parameters that need to be preset when using Kriging interpolation;

[0082] (3.2) Set the search radius L and skip radius R, and perform interpolation based on Kriging interpolation within the search radius and outside the skip radius in the eight main directions of each point cloud to obtain the interpolated point clouds on the eight elevation profiles;

[0083] (3.3) Based on the interpolated point clouds on the eight elevation profiles obtained in step (3.2), on the main direction profile of each point cloud, according to the maximum elevation angle and the minimum elevation angle calculate the upper opening angle and the lower opening angle

[0084]

[0085] (3.4) Set the flatness t that varies with the point cloud height. Calculate the elevation identification symbol

[0086]

[0087] for each profile according to the upper and lower opening angles obtained in step (3.3). The flatness t that varies with height is set to a single straight-line model:

[0088]

[0089] where t is the flatness, t min is the minimum flatness, generally set to 0, t max is the maximum flatness, which needs to be determined through optimization testing, z max is the highest elevation of the point cloud to be identified, z min the lowest elevation of the point cloud to be identified;

[0090] (3.5) According to the principle of local ternary value pattern, calculate the number of (+1) symbols and the number of (-1) symbols in the elevation identification symbols on the 8 elevation profiles of each point cloud, and establish the scour identification search coordinates

[0091] (3.6) According to the scour identification search coordinates obtained in step (3.5), calculate the scour pit element classification value θ for each point cloud;

[0092] Define the calculation value of the scour pit element classification value θ as follows: First, calculate the scour identification search coordinates which represent the identification symbols of the eight profiles in the local ternary value pattern, is the number of (+1) symbols among them, is the number of (-1) symbols among them; then calculate the mapping point of the distance from the mapping point of the pit element to the mapping point (8, 0) to define the scour pit element classification value θ for each point cloud, and its calculation formula is as follows:

[0093]

[0094] (3.7) Calculate the average value θ of the scour pit element classification value θ of all point clouds p as the threshold for segmenting the scour pit point cloud;

[0095] (3.8) For the parameter search radius, skip radius, and flatness within the value range, change them step by step one by one and perform recognition calculations for the recognition accuracy. When the recognition accuracy meets the extraction requirements, perform scour recognition on the underwater riverbed point cloud data of the bridge according to the optimized parameters, and identify the scour pit point cloud;

[0096] Statistically analyze the spatial geometric features of the sonar point cloud obtained by scanning, including: the minimum average spacing and the elevation coefficient of variation; use randomly generated pre-test point clouds with the same spatial geometric features as the sonar point cloud composed of several Gaussian amplitudes, and mark the point clouds below the horizontal plane as scour pit point clouds. Perform recognition on the pre-test point clouds. If the recognition result meets the recall rate R e greater than 0.85, the F1 value greater than 0.89, and the precision rate P r greater than 0.98 and the accuracy rate A c reach 0.93, then perform scour recognition on the scanned sonar point cloud according to this optimized parameter; the calculation formulas for the above evaluation indicators are:

[0097]

[0098] where TP is the true positive, referring to the sample points where the target point cloud is the point cloud inside the scour pit and is marked as the scour pit point cloud; FP is the false positive, referring to the sample points where the target point cloud is not the scour pit point cloud but is identified as the scour pit point cloud; TN is the true negative, referring to the sample points where the target point cloud is not the scour pit point cloud and is marked as the non-scour pit point cloud; FN is the false negative, referring to the sample points where the target point cloud is the scour pit point cloud but is marked as the non-scour pit point cloud;

[0099] (4) According to the scour pit point cloud identified in step (3), use the K-mean clustering algorithm for clustering and calculate the maximum depth of each pier scour pit;

[0100] As Figure 4 shown, step (4) specifically includes the following steps:

[0101] (4.1) According to the scour pit point cloud identified in step (3), calculate the coordinate mean and variance of the scour pit point cloud in the transverse direction of the bridge and perform denoising;

[0102] (4.2) Set the number k of piers within the sonar point cloud range and initialize k samples as the centers a of the initial clustering k ;

[0103] (4.3) For each point cloud p in the sonar point cloud; calculate the distances from the point cloud p to k clustering centers and assign it to the class corresponding to the clustering center with the minimum distance;

[0104] Calculate the distances from the point cloud p to k clustering centers using the squared Euclidean distance and assign it to the class corresponding to the clustering center with the minimum distance. The calculation formula for the squared Euclidean distance is as follows:

[0105] d(x,c)=(x - c) 2

[0106] where d represents the distance, x represents the row vector of the spatial coordinates of the point cloud, and c represents the row vector of the spatial coordinates of the centroid of the cluster;

[0107] (4.5) Repeat steps (4.2) and (4.3) until the number of iterations reaches the set upper limit to obtain the clustering results of the point cloud of each pier scour hole;

[0108] (4.6) For the point cloud of each pier scour hole in each cluster, calculate the minimum elevation value to obtain the maximum depth of each scour hole in the sonar - surveyed bridge;

[0109] (5) According to the scour hole element classification values of all the point clouds obtained in step (3), propose a surface reconstruction method based on scour recognition to perform high - precision surface reconstruction on the scour holes;

[0110] As Figure 5 shown, step (5) specifically includes the following steps:

[0111] (5.1) Calculate the in - center of each triangle in the mesh according to the triangular mesh preliminarily reconstructed by the preliminary spherical rotation algorithm in step (2);

[0112] (5.2) Obtain the neighboring points of the point cloud according to the KNN algorithm and perform covariance analysis to estimate the normal vector of each point cloud;

[0113] Construct a KD - Tree search structure for the scanned point cloud data, search for the point cloud within the local neighborhood of each point cloud, calculate the centroid p c of the neighborhood, and then calculate the covariance matrix of the point cloud p i as shown below:

[0114] M=∑(p i - p c )(p i - p c ) T

[0115] In the formula, M represents the covariance matrix. Then, according to the singular value decomposition method, obtain the eigenvalues of the covariance matrix, where the smallest eigenvalue represents the normal vector of the point;

[0116] (5.3) Calculate the tangent plane of each point in the point cloud;

[0117] (5.4) Calculate the projection points of the in - center of the triangles adjacent to each point in the point cloud onto the tangent plane of that point. The projection points form an adjustable polygon. Adjust the size of the adjustable polygon by moving the position of the projection points on the straight line between that point and the projection points. The adjustment weight is the scour pit element classification value θ calculated in step (3.6);

[0118] (5.5) Update the position of the adjustable triangles. Move the vertices of the original triangular mesh to the positions of the projection points, and traverse all the triangles in turn to construct new adjustable triangles;

[0119] (5.6) Divide the space quadrilateral between the adjustable polygon and the adjustable triangles according to the principle of the shortest diagonal to obtain triangulated patches;

[0120] (5.7) Integrate the adjustable triangles, the adjustable polygon, and the triangulated patches to obtain the finally reconstructed scour pit surface.

Claims

1. A method for reconstructing the morphological characteristics of a bridge scour surface based on three-dimensional sonar point cloud, characterized in that, It includes the following steps: (1) Collect the point cloud data of the three-dimensional shape of the underwater riverbed near the bridge foundation. Use a three-dimensional real-time sonar system carried on a waterborne survey ship to conduct multi-directional scans around the underwater riverbed of the bridge and perform real-time stitching, so as to obtain the original point cloud data of the complete shape of the underwater riverbed; (2) Perform preprocessing of denoising and spherical rotation reconstruction on the collected original point cloud data to complete the preliminary reconstruction of the point cloud; (3) According to the point cloud data after preliminary reconstruction in step (2), based on the principle of computer vision, perform erosion morphology recognition of point cloud binaryization, calculate the erosion pit element classification values of all point clouds, and identify the erosion pit point cloud; The specific steps of step (3) include the following steps: (3.1) Calculate the distances and semi-variances between all point clouds after preliminary reconstruction, and select the fitting Gaussian model as the theoretical variogram to fit the function of semi-variance and distance; (3.2) Set the search radius L and the skip radius R. Based on Kriging interpolation, within the search radius and outside the skip radius in the eight main directions of each point cloud, perform interpolation to obtain the interpolated point clouds on eight elevation profiles; (3.3) Based on the interpolated point clouds on the eight elevation profiles obtained in step (3.2), on the principal direction profile of each point cloud, according to the maximum elevation angle and the minimum elevation angle calculate the upper opening angle and the lower opening angle (3.4) Set the flatness t that varies with the height of the point cloud, and calculate the elevation identification symbol of each section according to the upper and lower opening angles obtained in step (3.3). (3.5) Calculate the (+1) symbols in the elevation identification symbols on the 8 elevation profiles of each point cloud according to the principle of the local ternary value pattern The (-1) symbols in the symbols Establish the scouring identification search coordinates (3.6) Scouring recognition search coordinates obtained according to step (3.5) Calculate the classification value θ of the scouring pit elements for each point cloud; (3.7) Calculate the average value $\theta$ of the scour hole element classification value $\theta$ of all point clouds p as the threshold for segmenting the scour hole point cloud; (3.8) For the parameters search radius, skip radius, and flatness within the value range, change them step by step one by one and perform recognition calculations to identify the accuracy. When the recognition accuracy meets the extraction requirements, perform erosion recognition on the point cloud data of the underwater riverbed of the bridge according to the optimized parameters, and identify the erosion pit point cloud; (4) According to the erosion pit point cloud identified in step (3), use the K-mean clustering algorithm for clustering and calculate the maximum depth of each pier erosion pit; (5) According to the erosion pit element classification values of all point clouds obtained in step (3), propose a surface reconstruction method based on erosion recognition to perform high-precision surface reconstruction on the erosion pit.

2. A method for reconstructing the morphological characteristics of a bridge scour surface based on three-dimensional sonar point clouds according to claim 1, characterized in that: The specific steps of step (2) include the following steps: (2.1) After performing attitude correction processing and noise interference preprocessing on each survey data, perform survey line merging; (2.2) Process the obtained water depth and position, and remove the noise interference in the overlapping coverage range of two adjacent survey lines one by one; (2.3) Finally, perform preliminary surface reconstruction on the denoised and merged point cloud using the spherical rotation reconstruction method to form a triangular mesh.

3. The method for reconstructing the morphological characteristics of the bridge scour surface based on three-dimensional sonar point cloud according to claim 2, wherein: In step (2.3), first search by calculating the density upper limit of the sonar point cloud, then specify a sphere with a radius ρ to contact any three points that do not contain other points to form a seed triangle, and rotate the sphere around the edge until it touches the third point to form a new triangle. Finally, traverse all reachable edges to complete the preliminary reconstruction of the point cloud and form a triangular mesh.

4. A method for reconstructing the morphological characteristics of a bridge scour surface based on three-dimensional sonar point cloud according to claim 1, characterized in that: In step (3.1), calculate the distances between all sample points and sort them in ascending order, and then use the Gaussian model to perform function fitting on the relationship between the semi-variance value and the distance. The general formula of the Gaussian model is: where γ represents the semi-variance, h represents the distance, C0 is the nugget constant, C is the sill, and a is the range, which are all parameters that need to be preset when using Kriging interpolation.

5. A method for reconstructing the morphological characteristics of a bridge scour surface based on 3D sonar point cloud according to claim 4, characterized in that: In step (3.4), the flatness t varying with height is set to a single straight line model: where t is the flatness, t min is the minimum flatness, set to 0, t max is the maximum flatness, z max is the highest elevation of the point cloud to be recognized, z min The lowest elevation of the point cloud to be recognized.

6. A method for reconstructing the morphological characteristics of a bridge scour surface based on three-dimensional sonar point cloud according to claim 5, characterized in that: In the step (3.6), the calculated value of the scour pit element classification value θ is defined as follows: First, calculate the scour identification search coordinates represents the identification symbols of eight profiles in the local ternary value pattern, is the number of (+1) symbols among them, is the number of (-1) symbols among them; then calculate the mapping point of the distance from the mapping point of the pit element to the mapping point (8, 0) of the pit element to define the scour pit element classification value θ of each point cloud, and its calculation formula is as follows:

7. A method for reconstructing the morphological characteristics of a bridge scour surface based on three-dimensional sonar point cloud according to claim 6, characterized in that: The spatial geometric features of the sonar point cloud obtained by statistical sounding in step (3.8) include: minimum average spacing and elevation coefficient of variation; a pre-test point cloud composed of several Gaussian amplitudes randomly generated and having the same spatial geometric features as the sonar point cloud is used. Points below the horizontal plane are marked as scour pit point clouds. The pre-test point cloud is identified. If the recognition result satisfies that the recall rate R e is greater than 0.85, the F1 value is greater than 0.89, and the precision rate P r is greater than 0.98, and the accuracy rate A c reaches 0.93, then the sonar point cloud obtained by sounding is subjected to scour identification according to this optimized parameter. Among them, the evaluation index accuracy rate A c , recall rate R e , precision rate P r and the calculation formulas of the F1 value are as follows: Among them, TP is the true positive, which refers to the sample points where the target point cloud is the point cloud in the scour pit and is marked as the scour pit point cloud; FP is the false positive, which refers to the sample points where the target point cloud is the non-scour pit point cloud but is recognized as the scour pit point cloud; TN is the true negative, which refers to the sample points where the target point cloud is the non-scour pit point cloud and is marked as the non-scour pit point cloud; FN is the false negative, which refers to the sample points where the target point cloud is the scour pit point cloud but is marked as the non-scour pit point cloud.

8. A method for reconstructing the morphological characteristics of a bridge scour surface based on three-dimensional sonar point cloud according to claim 1, characterized in that: The specific steps of step (4) are as follows: (4.1) According to the scour pit point cloud identified in step (3), calculate the coordinate mean and variance of the scour pit point cloud in the transverse bridge direction and perform denoising. (4.2) Set the number k of bridge piers within the sonar point cloud range, and initialize k samples as the centers a of the initial clustering k ; (4.3) For each point cloud p in the sonar point cloud, calculate the distance from the point cloud p to k cluster centers and assign it to the class corresponding to the cluster center with the minimum distance. The distance from the point cloud p to k cluster centers is calculated using the squared Euclidean distance and assigned to the class corresponding to the cluster center with the minimum distance. The calculation formula of the squared Euclidean distance is as follows: d(x,c) = (x - c) 2 Where d represents the distance, x represents the spatial coordinate row vector of the point cloud, and c represents the spatial coordinate row vector of the centroid of the cluster. (4.5) Repeat steps (4.2) and (4.3) until the number of iterations reaches the set upper limit to obtain the clustering results of each pier scour pit point cloud. (4.6) For the pier scour pit point cloud of each cluster, calculate the minimum elevation value to obtain the maximum depth of each scour pit of the sonar-scanned bridge.

9. A method for reconstructing the morphological characteristics of a bridge scour surface based on three-dimensional sonar point cloud according to claim 1, characterized in that: The specific steps of step (5) are as follows: (5.1) According to the triangular mesh initially reconstructed by the preliminary spherical rotation algorithm in step (2), calculate the incenter of each triangle in the mesh. (5.2) Use the KNN algorithm to obtain the neighboring points of the point cloud and perform covariance analysis to estimate the normal vector of each point cloud. Construct the KD-Tree search structure for the scanned point cloud data, search for the point cloud within its local neighborhood for each point cloud, and calculate the centroid p of the neighborhood c , and then calculate the covariance matrix of the point cloud p i as shown below: M = Σ(p i - p c )(p i - p c ) T In the formula, M represents the covariance matrix, and then the eigenvalues of the covariance matrix are obtained by the singular value decomposition method. The smallest eigenvalue represents the normal vector of the estimated point cloud. (5.3) Calculate the tangent plane of each point in the point cloud. (5.4) Calculate the projection points of the incenter of each adjacent triangle in the point cloud onto the tangent plane of the point. The projection points form an adjustable polygon. The size of the adjustable polygon is adjusted by moving the position of the projection point on the straight line between the point and the projection point. The adjustment weight is the scour pit element classification value θ calculated in step (3). (5.5) Update the position of the adjustable triangle, move the vertices of the original triangular mesh to the position of the projection point, and traverse all triangles in turn to construct new adjustable triangles. (5.6) Divide the spatial quadrilateral between the adjustable polygon and the adjustable triangle according to the principle of the shortest diagonal to obtain triangular meshes. (5.7) Integrate the adjustable triangles, adjustable polygons, and triangular meshes to obtain the finally reconstructed scour pit surface.

Images