A method for identifying the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds

Abstract

The present invention discloses a method for identifying the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds. The method comprises the following steps: collecting point cloud data of the three-dimensional morphology of the bridge's underwater pile foundations; utilizing a three-dimensional real-time sonar system equipped with an underwater survey vessel to perform multi-directional scanning around the bridge's underwater pile foundations and perform real-time splicing to obtain point cloud data of the complete morphology of the underwater pile foundations; extracting the pile foundations based on the point cloud data of the complete morphology of the underwater pile foundations using a point cloud minimum cut method; projecting the pile foundation point cloud onto the XOY plane based on the segmented and extracted individual pile foundations to obtain a planar pile foundation point cloud; identifying the morphological features of the pile foundations based on the planar pile foundation point cloud using an iterative MAD method; and mapping the complete pile foundation morphological feature point cloud back to the original three-dimensional space based on the recognition results according to the elevation information, thereby obtaining a three-dimensional pile foundation point cloud with complete morphological features. The present invention can achieve accurate identification of the morphological features of the complete pile foundation point cloud.

Description

Technical Field

[0001] The present invention belongs to the technical field of underwater bridge structure morphology recognition, and in particular relates to a method for recognizing the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds. Background Art

[0002] As critical load-bearing components of water-borne bridges, underwater pile foundations are constantly exposed to the impact of water currents, erosion from the aquatic environment, and collisions with ships and floating objects. These conditions make them susceptible to underwater damage, including pile damage, exposed reinforcement, necking, and localized scour. The emergence and progression of these underwater damages lead to the continuous degradation of bridge structural performance, directly threatening the overall safety of the bridge structure. According to existing infrastructure statistics, approximately 42% of bridges are over 50 years old, and 7.5% of bridges suffer structural defects, placing them in a precarious condition. As of 2018, China had approximately 70,000 dangerous bridges, representing one-twelfth of its existing bridge population. However, due to the difficulty of testing in aquatic environments, most research on bridge inspection and safety assessment has focused on the superstructure. In recent years, with the maturity of underwater applications of sonar technology, although certain achievements have been made in using sonar technology to detect underwater structures of bridges, these studies mainly focus on the acquisition and analysis of two-dimensional images based on underwater sonar, rather than direct spatial morphological recognition based on three-dimensional digital models. Only a small amount of research focuses on the direct analysis and recognition of three-dimensional digital models, and qualitative identification is only performed through manual observation of point clouds, resulting in low detection efficiency.

[0003] To efficiently and accurately identify the morphology of underwater bridge structures based on 3D digital models, it is essential to extract underwater point cloud data that meets the requirements of structural damage detection. However, underwater pile foundation point clouds acquired using 3D sonar technology often contain numerous interfering features, such as riverbed point clouds and aquatic organism point clouds. Specifically, the riverbed's undulating terrain, with pronounced unevenness and unevenness, is subject to years of erosion, resulting in potholes and siltation. Furthermore, aquatic organisms surrounding the pile foundations can partially obscure them. Consequently, existing technologies are unable to accurately identify the complete morphological features of the pile foundation point cloud. Summary of the Invention

[0004] Purpose of the invention: The purpose of the present invention is to provide a method for identifying the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds, which can remove complex riverbed point clouds and reduce the shielding effect of aquatic organisms on pile foundations, and realize accurate identification of the morphological features of complete pile foundation point clouds.

[0005] Technical solution: To achieve the above objectives, the present invention discloses a method for identifying the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds, comprising the following steps:

[0006] (1) Collecting point cloud data of the three-dimensional shape of the underwater pile foundation of the bridge: Using a three-dimensional real-time sonar system equipped with an underwater survey vessel to conduct multi-directional scanning around the underwater pile foundation of the bridge and splicing it in real time, thereby obtaining point cloud data of the complete shape of the underwater pile foundation;

[0007] (2) Based on the point cloud data of the complete form of the underwater pile foundation obtained in step (1), the pile foundation is extracted based on the point cloud minimum cut method. The specific steps are as follows:

[0008] (2.1) Constructing the graph structure of point cloud based on Kd-Tree;

[0009] (2.2) Assign different weight functions to all types of edges in the point cloud structure and set parameters;

[0010] (2.3) Set the prior points of the pile foundation to be segmented, and then calculate the minimum value of the sum of weights using the point element minimum cut method to achieve the segmentation of the pile foundation;

[0011] (2.4) Optimize the weight parameters. When the segmentation accuracy meets the extraction requirements, the pile groups are completely segmented and extracted one by one according to the optimized parameters.

[0012] (3) Segmenting the extracted single pile foundation according to step (2) and projecting the pile foundation point cloud onto the XOY plane to obtain a planar pile foundation point cloud;

[0013] (4) Based on the plane pile foundation point cloud obtained in step (3), pile foundation morphological features are identified based on the iterative MAD method;

[0014] (5) According to the recognition result of step (4), the complete pile foundation morphological feature point cloud is mapped back to the original three-dimensional space according to the elevation information, thereby obtaining a three-dimensional pile foundation point cloud with complete morphological features.

[0015] The specific steps of constructing the graph structure of the point cloud based on Kd-Tree in step (2.1) are as follows:

[0016] The graph structure of the point cloud is constructed, which contains two types of vertices and edges, one type of vertex is numbered 0 and the other type of vertex is numbered 1; among them, the two types of vertices are each point in the point cloud data, and a source point s and a sink point t are specified outside the point cloud. The two types of vertices, source points and sink points of the point cloud graph structure are all regarded as the main vertices constituting the point cloud graph structure; the edge is each vertex in the point cloud data and its connection with the nearest neighbor vertex, including the connection between point 0 and point 1, that is, 0-1, the connection between each vertex in the point cloud data and the source point s, that is, s-0 and s-1, and the connection between each vertex in the point cloud data and the sink t, that is, 0-t and 1-t.

[0017] Preferably, the specific steps of step (2.2) are: assigning different weight values ​​to each type of edge defined in the minimum cut algorithm of the point cloud, namely, the smoothing cost term SC, the background penalty term BGP, and the distance term DTC to the center point;

[0018]

[0019]

[0020]

[0021] The parameter σ is the average density of all point clouds, the parameter Radius is the distance from the center position of the structure to the non-target points outside it, that is, the horizontal radius of the target point cloud; the parameters centerX and centerY are the center coordinates of the target structure point cloud to be segmented, and dist is the Euclidean distance between point clouds.

[0022] Furthermore, the specific steps of step (2.3) are: the center position of the pile foundation to be divided is used as the prior point. Since the pile foundation is a cylindrical structure, the center position of the pile foundation to be divided is selected at the middle position along the elevation direction of the pile foundation.

[0023] Furthermore, the specific steps of step (2.4) are as follows: after determining the unique prior point position of each pile foundation, the first pile foundation to be segmented is optimized, and then segmented using the point cloud minimum cut method. Finally, the point cloud of the first pile foundation extracted by segmentation is compared and evaluated with its point cloud true value, where the point cloud true value is the point cloud true value of the first pile foundation selected; if the segmentation result meets the evaluation index, the other pile foundations to be segmented are segmented and extracted one by one according to the optimized parameters; the evaluation indexes of segmentation are precision Pr, recall rate Re and F1 value, that is, precision Pr, recall rate Re and F1 value all reach their respective standard values, and their calculation formulas are:

[0024]

[0025]

[0026]

[0027] Among them, TP is the data point that is actually the target and is marked as the target point, FP is the data point that is actually the non-target and is marked as the target point, and FN is the data point that is actually the target and is marked as the non-target point.

[0028] Preferably, the XOY plane in step (3) is a plane projected along the elevation direction of the pile foundation.

[0029] Furthermore, the specific steps of step (4) are as follows:

[0030] (4.1) The Pratt ultra-precise circle fitting method is used to estimate the center and radius parameters (x o ,y o ,r o );

[0031] (4.2) Calculate the distance from all point clouds to the center of the fitting circle (x o ,y o ) the square of the distance;

[0032] (4.3) Use the median absolute deviation to eliminate outliers;

[0033] (4.4) Iterate steps (4.1) to (4.3). When the calculation converges, the final point set is the pile foundation point cloud that completely eliminates outliers, that is, the complete pile foundation morphological feature point cloud.

[0034] Furthermore, any point (x i ,y i ) to the center of the circle is calculated as follows:

[0035] d i 2 =(x i -x o ) 2 +(y i -y o ) 2 .

[0036] Preferably, the formula for eliminating outliers in step (4.3) is:

[0037]

[0038] MAD=median(|d i 2 -median(d i 2 )|)×1.4826

[0039] Where M is d i 2 The median of d; MAD is d i 2 The median of the absolute values ​​of the median deviations; median is the function for calculating the median, and the constant value k is 2.5.

[0040] Furthermore, the specific steps of step (4.4) are as follows: the remaining points obtained after executing step (4.3) are regarded as a new fitting point set, and the new point set is iterated according to steps (4.1) to (4.3); after several iterations, when the number of remaining points no longer changes or the parameters of the fitting circle no longer change, the calculation is considered to have converged, thereby determining the final point set.

[0041] Beneficial effects: Compared with the existing technology, the present invention has the following significant advantages: the present invention uses a sonar system to collect point cloud data of the three-dimensional morphology of the underwater pile foundation of the bridge, and then divides and extracts the pile group one by one through the point cloud minimum cut algorithm, and finally realizes the identification and extraction of the complete morphological features of the underwater pile foundation of the bridge through the iterative MAD-based method, and finally achieves the purpose of providing data support for the safety diagnosis of the underwater structure of the bridge based on the three-dimensional digital model, and provides technical support for the non-contact detection of the underwater structure of the bridge based on three-dimensional sonar; the present invention does not require underwater operations, and does not need to reconstruct the three-dimensional morphology based on two-dimensional images, and can accurately obtain the three-dimensional complete morphological feature data of the underwater structure that meets the requirements of apparent damage identification and quantitative analysis, remove the complex riverbed point cloud and reduce the shielding effect of aquatic organisms on the pile foundation, and realize the accurate identification of the complete pile foundation point cloud morphological features. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the process of the present invention;

[0043] Figure 2 Schematic diagram of the point cloud minimum segmentation method of the present invention;

[0044] Figure 3 This is a single pile result image extracted by segmentation using the point cloud minimum cut method of the present invention;

[0045] Figure 4 The result diagram of the present invention achieving complete extraction of pile group foundations one by one;

[0046] Figure 5 This is a result diagram of the present invention's recognition of single pile morphological features;

[0047] Figure 6 This is the result diagram of the present invention's recognition of pile group morphological features. DETAILED DESCRIPTION

[0048] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0049] like Figure 1 As shown, the present invention provides a method for identifying the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds, comprising the following steps:

[0050] (1) Collecting point cloud data of the three-dimensional shape of the underwater pile foundation of the bridge: Using a three-dimensional real-time sonar system equipped with an underwater survey vessel to conduct multi-directional scanning around the underwater pile foundation of the bridge and splicing it in real time, thereby obtaining point cloud data of the complete shape of the underwater pile foundation;

[0051] (2) Based on the point cloud data of the complete form of the underwater pile foundation obtained in step (1), the pile foundation is extracted based on the point cloud minimum cut method, such as Figure 2 The specific steps are as follows:

[0052] (2.1) Constructing the graph structure of point cloud based on Kd-Tree;

[0053] The specific steps of constructing the graph structure of the point cloud based on Kd-Tree in step (2.1) are:

[0054] Construct a point cloud graph structure, which contains two types of vertices and edges, one type of vertex is numbered 0 and the other type of vertex is numbered 1; the two types of vertices are each point in the point cloud data, and a source point s and a sink point t are specified outside the point cloud. The two types of vertices, the source point and the sink point of the point cloud graph structure are all used as the main vertices of the point cloud graph structure; the edge is each vertex in the point cloud data and its connection with the nearest neighbor vertex, including the connection between point 0 and point 1, that is, 0-1, the connection between each vertex in the point cloud data and the source point s, that is, s-0 and s-1, and the connection between each vertex in the point cloud data and the sink point t, that is, 0-t and 1-t;

[0055] (2.2) Assign different weight functions to all types of edges in the point cloud structure and set parameters;

[0056] The specific steps of step (2.2) are: assign different weight values ​​to each type of edge defined in the minimum cut algorithm of the point cloud, namely the smoothing cost term SC, the background penalty term BGP, and the distance term DTC to the center point;

[0057]

[0058]

[0059]

[0060] The parameter σ is the average density of all point clouds, the parameter Radius is the distance from the center position of the structure to the non-target points outside it, that is, the horizontal radius of the target point cloud; the parameters centerX and centerY are the center coordinates of the target structure point cloud to be segmented, and dist is the Euclidean distance between point clouds;

[0061] (2.3) Set the prior points of the pile foundation to be segmented, and then calculate the minimum value of the weight sum through the point element minimum cut method to achieve the segmentation of the pile foundation, such as Figure 3 As shown;

[0062] The specific steps of step (2.3) are as follows: the center position of the pile foundation to be segmented is used as the prior point. Since the pile foundation is a cylindrical structure, the center position of the pile foundation to be segmented is selected at the middle position along the elevation direction of the pile foundation;

[0063] (2.4) The weight parameters are tuned. When the segmentation accuracy meets the extraction requirements, the pile groups are completely segmented and extracted one by one according to the tuned parameters, such as Figure 4 As shown;

[0064] The specific steps of step (2.4) are as follows: after determining the unique prior point position of each pile foundation, the first pile foundation to be segmented is optimized, and then segmented using the point cloud minimum cut method. Finally, the point cloud of the first pile foundation extracted by segmentation is compared and evaluated with its point cloud true value, where the point cloud true value is the point cloud true value of the first pile foundation selected; if the segmentation result meets the evaluation index, the other pile foundations to be segmented are segmented and extracted one by one according to the optimized parameters; the evaluation indexes of segmentation are precision Pr, recall rate Re and F1 value, that is, precision Pr, recall rate Re and F1 value all reach their respective standard values, that is, precision reaches 0.73, recall rate reaches 0.97 and F1 value reaches 0.84; their calculation formulas are:

[0065]

[0066]

[0067]

[0068] TP is a data point that is actually a target and is marked as a target point, FP is a data point that is actually a non-target and is marked as a target point, and FN is a data point that is actually a target and is marked as a non-target point.

[0069] (3) Segmenting the extracted single pile foundation according to step (2) and projecting the pile foundation point cloud onto the XOY plane to obtain a planar pile foundation point cloud; the XOY plane is a plane projected along the elevation direction of the pile foundation;

[0070] (4) Based on the plane pile foundation point cloud obtained in step (3), the pile foundation morphological features are identified based on the iterative MAD method, as follows: Figure 5 As shown;

[0071] The specific steps of step (4) are as follows:

[0072] (4.1) The Pratt ultra-precise circle fitting method is used to estimate the center and radius parameters (x o ,y o ,r o );

[0073] (4.2) Calculate the distance from all point clouds to the center of the fitting circle (x o ,y o ) the square of the distance;

[0074] Any point (x i ,y i ) to the center of the circle is calculated as follows:

[0075] d i 2 =(x i -x o ) 2 +(y i -y o ) 2

[0076] (4.3) Use the median absolute deviation to eliminate outliers;

[0077] The formula for eliminating outliers in step (4.3) is:

[0078]

[0079] MAD=median(|d i 2 -median(d i 2 )|)×1.4826

[0080] Where M is d i 2 The median of d; MAD is d i 2 The median of the absolute values ​​of the median deviations; median is the function for calculating the median, and the constant value k is 2.5;

[0081] (4.4) Iterate steps (4.1) to (4.3). When the calculation converges, the final point set is the pile foundation point cloud that completely eliminates outliers, that is, the complete pile foundation morphological feature point cloud;

[0082] The specific steps of step (4.4) are as follows: the remaining points obtained after executing step (4.3) are regarded as a new fitting point set, and the new point set is iterated according to steps (4.1) to (4.3); after several iterations, when the number of remaining points does not change or the parameters of the fitting circle do not change, the calculation is considered to have converged, and the final point set is determined;

[0083] (4) Based on the plane pile foundation point cloud obtained in step (3), pile foundation morphological features are identified based on the iterative MAD method;

[0084] (5) According to the recognition result of step (4), the complete pile foundation morphological feature point cloud is mapped back to the original three-dimensional space according to the elevation information, thereby obtaining a three-dimensional pile foundation point cloud with complete morphological features, such as Figure 6 shown.

[0085] The method of the present invention mainly serves the technical field of bridge underwater structure morphology recognition, and the application object is the field measured data of the bridge underwater structure. Therefore, the relevant method steps can be completed by writing codes based on professional software toolkits. In this embodiment, the programming calculation of the relevant method steps is completed based on the Matlab program. The above method steps and basic formula principles can also be loaded onto a computer or other programmable data processing device, so that a series of operation steps are executed on the computer or other programmable device to realize computer processing, thereby providing instructions executed on the computer or other programmable device for realizing the process. Figure 1 The present invention also discloses a computer storage medium for data collection and reception, wherein the computer storage medium stores a computer program that is executed by a processor to execute the method disclosed in the present invention for identifying the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds.

Claims

1. A method for identifying the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds, characterized in that: The steps include: (1) Collecting point cloud data of the three-dimensional shape of the underwater pile foundation of the bridge: Using a three-dimensional real-time sonar system equipped with an underwater survey vessel to conduct multi-directional scanning around the underwater pile foundation of the bridge and splicing it in real time, thereby obtaining point cloud data of the complete shape of the underwater pile foundation; (2) Based on the point cloud data of the complete form of the underwater pile foundation obtained in step (1), the pile foundation is extracted based on the point cloud minimum cut method. The specific steps are as follows: (2.1) Constructing the graph structure of point cloud based on Kd-Tree; (2.2) Assign different weight functions to all types of edges in the point cloud structure and set parameters; (2.3) setting a priori point of the pile foundation to be segmented, and then calculating the minimum value of the sum of weights by the point element minimum cut method to achieve the segmentation of the pile foundation; the specific steps of step (2.3) are: taking the center position of the pile foundation to be segmented as the priori point, because the pile foundation is a cylindrical structure, the center position of the pile foundation to be segmented is selected at the middle position along the elevation direction of the pile foundation; (2.4) Optimize the weight parameters. When the segmentation accuracy meets the extraction requirements, the pile groups are completely segmented and extracted one by one according to the optimized parameters. (3) Segmenting the extracted single pile foundation according to step (2) and projecting the pile foundation point cloud onto the XOY plane to obtain a planar pile foundation point cloud; (4) Based on the plane pile foundation point cloud obtained in step (3), pile foundation morphological features are identified based on the iterative MAD method; The specific steps of step (4) are as follows: (4.1) The Pratt ultra-precise circle fitting method is used to estimate the center and radius parameters (x o ,y o ,r o ); (4.2) Calculate the distance from all point clouds to the center of the fitting circle (x o ,y o ) the square of the distance; (4.3) Use the median absolute deviation to eliminate outliers; (4.4) Iterate steps (4.1) to (4.3). When the calculation converges, the final point set is the pile foundation point cloud that completely eliminates outliers, that is, the complete pile foundation morphological feature point cloud; (5) According to the recognition result of step (4), the complete pile foundation morphological feature point cloud is mapped back to the original three-dimensional space according to the elevation information, thereby obtaining a three-dimensional pile foundation point cloud with complete morphological features.

2. The method for recognizing the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds according to claim 1, characterized in that: The specific steps of constructing the graph structure of the point cloud based on Kd-Tree in step (2.1) are: The graph structure of the point cloud is constructed, which contains two types of vertices and edges, one type of vertex is numbered 0 and the other type of vertex is numbered 1; among them, the two types of vertices are each point in the point cloud data, and a source point s and a sink point t are specified outside the point cloud. The two types of vertices, source points and sink points of the point cloud graph structure are all regarded as the main vertices constituting the point cloud graph structure; the edge is each vertex in the point cloud data and its connection with the nearest neighbor vertex, including the connection between point 0 and point 1, that is, 0-1, the connection between each vertex in the point cloud data and the source point s, that is, s-0 and s-1, and the connection between each vertex in the point cloud data and the sink t, that is, 0-t and 1-t.

3. The method for recognizing the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds according to claim 2, characterized in that: The specific steps of step (2.2) are: assigning different weight values ​​to each type of edge defined in the minimum cut algorithm of the point cloud, namely, the smoothing cost term SC, the background penalty term BGP, and the distance term DTC to the center point; The parameter σ is the average density of all point clouds, the parameter Radius is the distance from the center position of the structure to the non-target points outside it, that is, the horizontal radius of the target point cloud; the parameters centerX and centerY are the center coordinates of the target structure point cloud to be segmented, and dist is the Euclidean distance between point clouds.

4. The method for recognizing the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds according to claim 1 is characterized by: The specific steps of step (2.4) are as follows: after determining the unique prior point position of each pile foundation, the first pile foundation to be segmented is optimized, and then segmented using the point cloud minimum cut method, and finally the point cloud of the first pile foundation extracted by segmentation is compared and evaluated with its point cloud true value, where the point cloud true value is the point cloud true value of the first pile foundation selected; if the segmentation result meets the evaluation index, the other pile foundations to be segmented are segmented and extracted one by one according to the optimized parameters; the evaluation indexes of segmentation are precision Pr, recall rate Re and F1 value, that is, precision Pr, recall rate Re and F1 value all reach their respective standard values, and their calculation formulas are: Among them, TP is the data point that is actually the target and is marked as the target point, FP is the data point that is actually the non-target and is marked as the target point, and FN is the data point that is actually the target and is marked as the non-target point.

5. The method for recognizing the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds according to claim 4 is characterized by: The XOY plane in step (3) is a plane projected along the elevation direction of the pile foundation.

6. The method for recognizing the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds according to claim 5, characterized in that: Any point (x i ,y i ) to the center of the circle is calculated as follows: d i 2 =(x i -x o ) 2 +(y i -y o ) 2 。 7. The method for recognizing the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds according to claim 6, characterized in that: The formula for eliminating outliers in step (4.3) is: MAD=median(|d i 2 -median(d i 2 )|)×1.4826 Where M is d i 2 The median of d; MAD is d i 2 The median of the absolute values ​​of the median deviations; median is the function for calculating the median, and the constant value k is 2.

5.

8. The method for recognizing the morphological features of underwater bridge pile groups based on three-dimensional sonar point clouds according to claim 7, characterized in that: The specific steps of step (4.4) are as follows: the remaining points obtained after executing step (4.3) are regarded as a new fitting point set, and the new point set is iterated according to steps (4.1) to (4.3); after several iterations, when the number of remaining points no longer changes or the parameters of the fitting circle no longer change, the calculation is considered to have converged, thereby determining the final point set.

Images