Steel arch bridge arch rib linear extraction method and system based on improved DBSCAN algorithm

Abstract

The application discloses a steel arch bridge arch rib linear extraction method and system based on an improved DBSCAN algorithm, and comprises the following steps: collecting point cloud data of a steel arch bridge to obtain original point cloud data; preprocessing the original point cloud data to obtain preprocessed point cloud data; preliminarily extracting the preprocessed point cloud data to obtain a preliminary extraction result; and performing secondary extraction on the preliminary extraction result by using an improved DBSCAN algorithm to obtain a final extraction result. The application can more accurately and effectively complete steel arch bridge linear extraction, and significantly improves linear extraction precision.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of steel arch bridge monitoring, in particular to a steel arch bridge arch rib linear extraction method and system based on an improved DBSCAN algorithm. BACKGROUND

[0002] Steel arch bridges, as a structural form, have a long history of development and wide application in bridge construction. The earliest steel arch bridges appeared in the mid-19th century. These bridges quickly gained widespread application due to their superior mechanical properties and long service life. In the mid-20th century, steel arch bridge technology gradually matured, particularly in the application of long-span bridges and complex terrain conditions, leading to the construction of numerous modern steel arch bridges around the world. In the 21st century, with further advancements in materials science and construction technology, steel arch bridges have achieved more optimized and efficient design, construction, and maintenance.

[0003] Common methods for steel arch bridge linear monitoring include total station measurement combined with GPS monitoring, which involves high-precision three-dimensional coordinate measurement of key nodes on the bridge to monitor changes in the bridge's linear shape. Total station measurement has the advantages of high precision and wide applicability, making it suitable for long-term bridge monitoring. However, it is highly dependent on human labor, has high line-of-sight requirements, and has limited coverage. In addition to this, there are two other methods: photogrammetry and laser scanning. Photogrammetry uses high-resolution cameras to take photos of the bridge from different angles, reconstructing a three-dimensional model of the bridge to monitor changes in its linear shape. This method is suitable for short-term monitoring and monitoring in complex environments, but its precision is limited by the photography equipment and processing technology, and data processing is complex. Laser scanning technology can quickly obtain high-density point cloud data of the bridge surface, and by analyzing the point cloud data, the linear shape of the bridge can be monitored. This method is suitable for large-scale and high-precision linear monitoring.

[0004] Although some scholars have successfully extracted bridge linear shapes from point cloud data, the existing methods have certain limitations in dealing with the complex structure and diversity of steel arch bridges, making it difficult to achieve accurate extraction. Therefore, to solve the above problems, a steel arch bridge arch rib linear extraction method and system based on an improved DBSCAN algorithm is needed, which can more accurately and effectively complete the linear extraction of steel arch bridges and significantly improve the linear extraction precision. SUMMARY

[0005] Therefore, the purpose of the present application is to overcome the defects in the prior art and provide a steel arch bridge arch rib linear extraction method and system based on an improved DBSCAN algorithm, which can more accurately and effectively complete the linear extraction of steel arch bridges and significantly improve the linear extraction precision.

[0006] The steel arch bridge arch rib linear extraction method based on an improved DBSCAN algorithm of the present application comprises:

[0007] Point cloud data of a steel arch bridge is collected to obtain original point cloud data;

[0008] The original point cloud data is preprocessed to obtain preprocessed point cloud data;

[0009] The preprocessed point cloud data is preliminarily extracted to obtain a preliminary extraction result;

[0010] The improved DBSCAN algorithm is used to perform secondary extraction on the preliminary extraction result to obtain a final extraction result.

[0011] Further, the preprocessing includes downsampling and denoising;

[0012] The downsampling specifically includes: using the Octree algorithm to downsample the point cloud data, taking the center of gravity of each Octant as a representative point, and constructing a new downsampled point cloud;

[0013] The denoising specifically includes: using bilateral filtering to denoise the downsampled point cloud data.

[0014] Further, the preliminary extraction on the preprocessed point cloud data specifically includes:

[0015] The PCA algorithm is used to estimate the normal of the point cloud, and the eigenvector corresponding to the maximum eigenvalue is selected as the normal direction to obtain the normal direction of each point in the point cloud data set;

[0016] The neighborhood point n is set to n0, and the direction of the normal vector is unified;

[0017] According to the analysis result of the point cloud data, threshold values are set in the Z direction and the X direction, and the point clouds at the bottom and both sides of the arch rib are extracted respectively.

[0018] Further, the improved DBSCAN algorithm is used to perform secondary extraction on the preliminary extraction result, specifically including:

[0019] a. Fitting and segmenting the first straight line: using the RANSAC algorithm to fit a straight line in the current point cloud, finding the inliers of the straight line, extracting these inliers and removing them from the original point cloud;

[0020] b. Continue to fit the remaining point cloud: apply the RANSAC algorithm again on the remaining point cloud to fit the next straight line;

[0021] c. Repeat this process through the above steps until the number of points in the remaining point cloud is less than min_num;

[0022] d. Determine the straight line to which the point belongs: in each RANSAC algorithm fitting, determine whether each point belongs to the current straight line by calculating the distance from the point to the fitted straight line, wherein only the inliers are segmented out and removed from the original point cloud;

[0023] e. Represent each straight line as a set of feature vectors;

[0024] f. Cluster the feature vectors of these straight lines using the DBSCAN algorithm, and if the number of neighbors of a certain straight line is greater than or equal to MinPts within a given ∈-neighborhood, the straight line is considered to be a core straight line, and the core straight line and the density-reachable straight lines form a cluster;

[0025] g. Continue to cluster the remaining straight lines until all straight lines are classified;

[0026] h. According to the classification result, screen the bridge part point cloud, delete the interference point cloud of each type, and only keep the point cloud at the arch rib;

[0027] i. Extract and fit the key point coordinates of the screened point cloud.

[0028] An improved DBSCAN algorithm-based steel arch bridge arch rib line extraction system, comprising a collection unit, a preprocessing unit, a preliminary extraction unit, and a secondary extraction unit;

[0029] The collection unit is configured to collect point cloud data of the steel arch bridge to obtain original point cloud data;

[0030] The preprocessing unit is configured to preprocess the original point cloud data to obtain preprocessed point cloud data;

[0031] The preliminary extraction unit is configured to preliminarily extract the preprocessed point cloud data to obtain a preliminary extraction result;

[0032] The secondary extraction unit is configured to use an improved DBSCAN algorithm to secondarily extract the preliminary extraction result to obtain a final extraction result.

[0033] Further, the preprocessing includes downsampling and denoising;

[0034] The downsampling specifically includes: using an Octree algorithm to downsample the point cloud data, taking the center of gravity of each Octant as a representative point, and constructing a new downsampled point cloud;

[0035] The denoising specifically includes: using bilateral filtering to denoise the downsampled point cloud data.

[0036] Further, the preliminary extraction of the preprocessed point cloud data specifically includes:

[0037] The PCA algorithm is used to estimate the normal of the point cloud, and the eigenvector corresponding to the maximum eigenvalue is selected as the normal direction to obtain the normal direction of each point in the point cloud data set;

[0038] The neighborhood point n is set to n0, and the normal vector is direction-consistent;

[0039] According to the analysis result of the point cloud data, threshold values are set in the Z direction and the X direction, and the point clouds at the bottom and both sides of the arch rib are extracted respectively.

[0040] Further, the improved DBSCAN algorithm is used for secondary extraction of the preliminary extraction result, which specifically includes:

[0041] a. Fitting and segmenting the first straight line: using the RANSAC algorithm to fit the straight line in the current point cloud, finding the inliers of the straight line, extracting these inliers and removing them from the original point cloud;

[0042] b. Continue to fit the remaining point cloud: apply the RANSAC algorithm again on the remaining point cloud to fit the next straight line;

[0043] c. Repeat the process through the above steps until the number of points in the remaining point cloud is less than min_num;

[0044] d. Determine the straight line to which the point belongs: in each RANSAC algorithm fitting, determine whether it belongs to the current straight line by calculating the distance of each point to the fitted straight line, wherein only the inliers will be segmented and removed from the original point cloud;

[0045] e. Express each straight line as a set of eigenvectors;

[0046] f. Use the DBSCAN algorithm to cluster the eigenvectors of these straight lines. If the number of neighbors of a straight line is greater than or equal to MinPts within a given ∈-neighborhood, the straight line is considered to be a core straight line, and the core straight line and the density-reachable straight lines form a cluster;

[0047] g. Continue to cluster the remaining straight lines until all straight lines are classified;

[0048] h. According to the classification result, the bridge part point cloud is screened, and the interference point cloud is deleted, and only the point cloud at the arch rib is reserved;

[0049] i. Extract and fit the key point coordinates of the screened point cloud.

[0050] The beneficial effects of the present application are: the steel arch bridge rib linear extraction method and system based on the improved DBSCAN algorithm are disclosed, the large-span steel arch bridge with a complex structure is taken as an analysis object, the three-dimensional laser scanning technology is used for scanning the steel arch bridge, the PCA algorithm is used for the preliminary extraction of the key structure point cloud of the steel arch bridge according to the structural characteristics of the steel arch bridge, and the improved DBSCAN algorithm fused with RANSAC is used to realize the accurate extraction of the arch rib of the steel arch bridge according to the structures of the interference linear extraction results of the bridge deck, cross brace, cross link and web member and the like in the preliminary extraction results. BRIEF DESCRIPTION OF DRAWINGS

[0051] The present application will be further described below in combination with the drawings and embodiments:

[0052] Figure 1 It is a schematic diagram of the original point cloud data of the steel arch bridge of the present application;

[0053] Figure 2 It is a schematic diagram of the original point cloud data of the steel arch bridge of the present application after pretreatment;

[0054] Figure 3 It is a schematic diagram of the key part point cloud of the arch rib after preliminary extraction of the present application;

[0055] Figure 4a It is a schematic diagram of the point cloud screening result of the two sides of the arch rib of the present application;

[0056] Figure 4b It is a schematic diagram of the point cloud screening result of the bottom of the arch rib of the present application;

[0057] Figure 5a It is a schematic diagram of the point cloud error distribution of the bottom of the arch rib of the present application;

[0058] Figure 5b It is a schematic diagram of the point cloud error distribution of the two sides of the arch rib of the present application;

[0059] Figure 6 It is a schematic diagram of the steel arch bridge rib linear extraction method flow of the present application;

[0060] Figure 7a It is a schematic diagram of the clustering effect of the DBSCAN algorithm without improvement;

[0061] Figure 7b It is a schematic diagram of the clustering effect of the improved DBSCAN algorithm fused with RANSAC of the present application. DETAILED DESCRIPTION

[0062] The present application will be further described below in combination with the drawings and embodiments, as shown in the drawings:

[0063] The embodiment discloses a steel arch bridge arch rib linear extraction method based on an improved DBSCAN algorithm, and comprises the following steps:

[0064] Point cloud data of the steel arch bridge is collected to obtain original point cloud data.

[0065] The original point cloud data is preprocessed to obtain preprocessed point cloud data.

[0066] The preprocessed point cloud data is preliminarily extracted to obtain a preliminary extraction result.

[0067] The preliminary extraction result is secondarily extracted by using the improved DBSCAN algorithm to obtain a final extraction result.

[0068] In the embodiment, a RiGel frame station type scanner is used to perform three-dimensional laser scanning on a steel arch bridge in a construction stage. Two stations are respectively erected on the two banks of the bridge, and three stations are erected on the bridge. The original point cloud data has a total of 178946265 points, as shown in the table. Figure 1 The three-dimensional laser point cloud data has the advantages of comprehensiveness and detail presentation, and can completely present the shape and deformation information (displacement, deformation, settlement and other parameters) of the bridge structure.

[0069] Because the amount of original point cloud data is too large and there are a small amount of noise points, the original point cloud data is first down-sampled and denoised. For down-sampling: the Octree algorithm is used to down-sample the point cloud. The Octree algorithm is an algorithm commonly used for point cloud data processing and analysis, mainly used to reduce the density and complexity of point cloud data while maintaining the main features and structure of the point cloud data. The Octree algorithm effectively partitions and organizes point cloud data by recursively dividing a three-dimensional space into eight sub-cubes (Octants). Because Octree is based on spatial division, it can preserve the main structural features of the point cloud data. This is because the center of gravity of each Octant is calculated from all the points in that region, which can better preserve the distribution and structural features of the original data. The algorithm steps are as follows:

[0070] Construct an Octree to recursively divide the entire point cloud data space into eight sub-cubes until the number of points in each sub-cube is less than a predetermined threshold or reaches a maximum depth.

[0071]

[0072] In the formula: P sub is the point in the current Octant.

[0073] For each non-empty Octant, calculate the center of gravity of all points inside it as the representative point.

[0074]

[0075] where p i - points within Octant; n - number of points within Octant;

[0076] Down-sampling, taking the center of gravity of each Octant as the representative point, a new down-sampled point cloud is constructed.

[0077] For denoising: considering the structural characteristics of the bridge, the bilateral filter which is more adaptive to the complex structure of the bridge is selected to denoise the down-sampled point cloud data, as shown in the pre-processed original point cloud number.Bilateral filtering is a nonlinear filtering technique in point cloud processing, which can handle nonlinearly changing point cloud data and has better robustness to nonlinear noise. Since it considers the distance in both spatial and intensity domains, the edges and details of the filtered point cloud are better preserved. Figure 2

[0078] In this embodiment, the pre-processed point cloud data is preliminarily extracted, specifically including:

[0079] The PCA algorithm is used to estimate the normal of the point cloud, and the normal vector is processed to be consistent in direction:

[0080] The PCA algorithm regards the local point cloud of each point as a smooth surface, then estimates a best tangent plane in the neighborhood using the least squares algorithm, and finally sets the normal vector of the best tangent plane as the normal vector of the sampling point. The algorithm steps are as follows:

[0081] For a point cloud dataset containing n three-dimensional points, where each point is represented by three coordinate values (x i ,y i ,z i ), these points can be represented as a 3x n matrix X, whose column vectors are all the points in the point cloud. First, calculate the mean vector

[0082]

[0083] where X i is the i-th point in the point cloud.

[0084] Decenter the point cloud dataset:

[0085]

[0086] Calculate the covariance matrix C:

[0087]

[0088] The covariance matrix C is a symmetric matrix, and the eigenvector thereof is the normal direction. For each point in the point cloud, the PCA algorithm calculates the normal by solving the eigenvalue and eigenvector of the covariance matrix. The eigenvalue corresponding to the eigenvector represents the variance in the direction, and therefore, the eigenvector corresponding to the maximum eigenvalue is the main direction of the point cloud data, i.e., the normal direction. The eigenvalue and eigenvector of the covariance matrix are solved as follows:

[0089] Cv = λv;

[0090] In the formula, v is the eigenvector, and λ is the eigenvalue.

[0091] The eigenvector corresponding to the maximum eigenvalue is selected as the normal direction.

[0092] Through the above steps, the PCA algorithm can calculate the normal direction of each point in the point cloud data set.

[0093] The neighborhood point n = 5 is set, and the direction of the normal vector is unified.

[0094] According to the analysis result of the point cloud data, the threshold is set, the threshold in the Z direction is set to 0.85-1.000, the threshold in the X direction is set to -1.00--0.99 and 0.99-1.00, and the point clouds at the bottom and both sides of the arch rib are extracted, respectively. After processing, the point cloud data has 865105 points, as shown in Figure 3 .

[0095] In this embodiment, the normal feature of the point cloud has been used to realize the preliminary extraction of the point clouds at the bottom and both sides of the arch rib. However, due to the complexity of the steel arch structure, there are still many discrete points, points on the arch bridge deck, cross braces, cross ties, and web members, and points on the retaining wall in the extraction result, which will affect the extraction of the arch bridge alignment. In order to remove the interference of these points on the extraction result, the RANSAC algorithm is used for linear fitting classification and removal of part of the discrete points, and the fitted straight line is subjected to DBSCAN clustering to remove the points on the arch bridge deck, cross braces, cross ties, and web members, and the points on the retaining wall.

[0096] The improved DBSCAN algorithm is used for secondary extraction of the preliminary extraction result, specifically including:

[0097] a. fitting and segmenting the first straight line: the RANSAC algorithm is used to fit a straight line in the current point cloud, and the inliers (i.e., points conforming to the straight line model) of the straight line are found, and these inliers are extracted and removed from the original point cloud; wherein the straight line can be represented as: L(A, B): y = A + Bx;

[0098] b. continue to fit the remaining point cloud: the RANSAC algorithm is applied again on the remaining point cloud to fit the next straight line;

[0099] c. Repeat the process by the above steps until the number of points in the remaining point cloud is less than min_num;

[0100] d. Determine the straight line to which the point belongs: in each RANSAC algorithm fitting, determine whether each point belongs to the current straight line (i.e., whether it is an inlier) by calculating the distance of each point to the fitted straight line, wherein only inliers will be segmented out and removed from the original point cloud;

[0101] e. Represent each straight line as a set of feature vectors; for example, the feature vectors are as follows:

[0102]

[0103] f. Use the DBSCAN algorithm to cluster the feature vectors of these straight lines, and if the number of neighbors of a certain straight line is greater than or equal to MinPts within a given ∈-neighborhood, the straight line is considered a core straight line, and the core straight line and the density-reachable straight lines form a cluster;

[0104] g. Continue to cluster the remaining straight lines until all straight lines are classified; wherein the clustering results are shown in FIG. 7.

[0105] h. According to the classification results, filter the bridge part point cloud, delete the interference point cloud of each type, and only keep the point cloud at the arch rib; wherein the filtering results can be as shown in FIG. 4.

[0106] i. Extract and fit the coordinates of the key points after filtering the point cloud. Wherein the key points refer to points with special significance in the point cloud, such as corner points, edge points or feature points. Extract these key points and fit their coordinates to identify these key points from the point cloud, and use mathematical models or fitting methods (such as curve fitting or surface fitting, etc.) to estimate their precise positions.

[0107] Wherein, the RANSAC algorithm is an iterative algorithm for estimating the parameters of a mathematical model, especially suitable for cases where there are a large number of noisy data. It is a robust parameter estimation algorithm that can estimate model parameters from a set of data containing a large number of outliers. Its basic principle is as follows:

[0108] Given a two-dimensional data set containing N points Randomly select two points from the data set, because the smallest data subset for line fitting is two points, these two points define a candidate straight line. For the two randomly selected points (x1, y1), (x2, y2), the equation of the straight line can be written as:

[0109] y = mx + c

[0110] Wherein, the slope m and the intercept c can be calculated by the following formula:

[0111]

[0112] c = y1 - mx1

[0113] For each data point (x i ,y i ), an inlier evaluation is performed to calculate its perpendicular distance to the fitted line:

[0114]

[0115] A threshold d th is set, and if d i ≤ d th , the point is considered an inlier. Then the fitted model is evaluated to count the number of inliers, and if the number of inliers exceeds a pre-set threshold (such as the percentage of inliers exceeding a certain percentage), a reasonable straight line is considered to be found. The straight line model with the most inliers is retained. Repeat the above steps k times (the number of iterations can be adjusted according to actual situation and data set size), and select the straight line model with the most inliers as the final fitting result.

[0116] DBSCAN algorithm is a density-based clustering algorithm, which is particularly suitable for discovering clusters of arbitrary shape and handling noisy data. The basic idea of DBSCAN is to define clusters by density, that is, the density of points in a cluster is greater than a certain threshold. The basic principle and algorithm flow are as follows:

[0117] Given a point p and radius ∈, the ∈-neighborhood of p contains all points whose distance from p does not exceed ∈.

[0118] N ∈ (p) = {q ∈ D | dist(p, q) ≤ ∈}

[0119]

[0120] In the formula: D——data set; dist(p, q)——distance between point p and point q (usually using Euclidean distance).

[0121] If the ∈-neighborhood of a point p contains at least MinPts points (including p itself), p is called a core point; if a point p is not a core point itself, but it falls within the ∈-neighborhood of a core point, p is called a boundary point; if a point p is neither a core point nor a boundary point, p is called a noise point. Core point determination:

[0122] N ∈ (p) > MinPts

[0123] Select any unvisited point, if the point is a core point, take the point as the starting point, find all directly density reachable points based on the epsilon neighborhood to form a cluster, if the point is a boundary point or a noise point, mark it as noise (if it is included in the epsilon neighborhood of other points in the subsequent step, it may be re-marked as a boundary point). Starting from the core point, recursively add all directly density reachable points to the current cluster. If point p is a core point and point q is within the epsilon neighborhood of p, q is directly density reachable from p. Repeat the above steps until all points are visited.

[0124] Through the above processing, (1) structured segmentation: RANSAC can effectively fit and segment points with specific geometric shapes, thereby providing structural constraints in preliminary segmentation. (2) Reduce the influence of noise: RANSAC can effectively reduce the sensitivity to noise points through iteration and model verification.

[0125] For steel arch bridges with complex structures but obvious straight line structures, the RANSAC algorithm effectively extracts these structures and constrains the clustering results of the DBSCAN algorithm. Through DBSCAN, the extracted straight lines are further segmented and clustered, and the final clustering results can effectively separate discrete points and bridge decks, cross braces, cross links and web members, and accurately extract point clouds of key parts of the steel arch bridge arch rib.

[0126] The application also relates to a steel arch bridge arch rib line extraction system based on the improved DBSCAN algorithm, which corresponds to the steel arch bridge arch rib line extraction method of the above-mentioned embodiments and can be understood as a system for realizing the above-mentioned steel arch bridge arch rib line extraction method. The system comprises a collection unit, a preprocessing unit, a preliminary extraction unit and a secondary extraction unit.

[0127] The collection unit is used for collecting point cloud data of the steel arch bridge to obtain original point cloud data.

[0128] The preprocessing unit is used for preprocessing the original point cloud data to obtain preprocessed point cloud data.

[0129] The preliminary extraction unit is used for preliminarily extracting the preprocessed point cloud data to obtain a preliminary extraction result.

[0130] The secondary extraction unit is used for secondarily extracting the preliminary extraction result by using the improved DBSCAN algorithm to obtain a final extraction result.

[0131] In order to better understand the technical scheme of the application, the application will be further described as follows:

[0132] If the DBSCAN clustering is directly performed on the point cloud, the extraction of the part point cloud of the arch rib cannot be realized, wherein for the point cloud on both sides of the arch rib, only a few discrete points are filtered out, but the connecting component between the arch ribs cannot be filtered out; for the point cloud at the bottom of the arch rib, no effect is achieved, and even some data points are filtered out. The improved DBSCAN algorithm fusing RANSAC can well achieve the purpose of deleting the discrete points and the point cloud of the bridge deck and cross rod.

[0133] In the results of different parameters, for the point cloud on both sides of the arch rib, when the minimum fitting point number of the RANSAC linear fitting is 25, a large number of useful data points are filtered out, and when the minimum fitting point number of the RANSAC linear fitting is 15 or 20, the extraction effect is better, and the part point cloud of the arch rib can be separated out; for the point cloud at the bottom of the arch rib, when the minimum fitting point number of the RANSAC linear fitting is 25, almost all the point cloud is filtered out, and when the minimum fitting point number of the RANSAC linear fitting is 15 or 20, the extraction effect is better, wherein when the minimum fitting point number of the RANSAC linear fitting is 20, the cross link and the web can be filtered out except the cross brace, and the effect is the best.

[0134] According to the improved DBSCAN algorithm fusing RANSAC, the key point fitting of the extracted point cloud is performed, and the result is compared with the artificial extraction result, the key point error of the point cloud on both sides and at the bottom of the arch rib is in the millimeter level, wherein the maximum difference of the extracted point cloud at the bottom of the arch rib is 9.2 mm, the minimum difference is 0.1 mm, the proportion of the absolute value of the difference below 5.0 mm is 92.05%, and the proportion of the absolute value of the difference below 2.0 mm is 82.95%; the maximum difference of the extracted point cloud on both sides of the arch rib is 8.0 mm, the minimum difference is 0.1 mm, the proportion of the absolute value of the difference below 5.0 mm is 96.74%, and the proportion of the absolute value of the difference below 2.0 mm is 56.82%. The improved DBSCAN algorithm fusing RANSAC can accurately and quickly realize the arch rib linear automatic extraction overcoming the influence of the complex structure of the steel arch bridge, and the extraction precision can meet the actual production demand.

[0135] The improved DBSCAN algorithm fusing RANSAC provided by the application can objectively and scientifically extract the arch rib linear of the steel arch bridge, and the normal feature data of the steel arch bridge is obtained by using the three-dimensional laser scanning technology and the PCA algorithm. The normal feature data is extracted by processing the three-dimensional point cloud data of the steel arch bridge collected by the three-dimensional laser scanner. The point cloud is extracted based on the different vertical characteristics of the normal of the steel arch bridge at different positions, and the feature point cloud of the key part of the arch rib is obtained. The steel arch bridge linear extraction can be more accurately and effectively completed. The linear extraction precision reaches the millimeter level by using the normal feature and the structure feature of the arch rib of the steel arch bridge and the three-dimensional laser point cloud. The human consumption and the time consumption are greatly reduced. The process is more simple, the result is more accurate, the complex structure of the steel arch bridge has better robustness, and the actual production demand can be well met.

[0136] Finally, it is to be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application but not to limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced without departing from the purpose and scope of the present application, and all of them should be covered in the scope of the claims of the present application.

Claims

1. A steel arch bridge arch rib linear extraction method based on an improved DBSCAN algorithm, characterized by: The method comprises the following steps: Point cloud data of a steel arch bridge is collected to obtain original point cloud data; The original point cloud data is preprocessed to obtain preprocessed point cloud data; The preprocessed point cloud data is preliminarily extracted to obtain a preliminary extraction result; The preliminary extraction result is secondarily extracted by using an improved DBSCAN algorithm to obtain a final extraction result; The preliminary extraction result is secondarily extracted by using an improved DBSCAN algorithm, specifically including: a. fitting and segmenting a first straight line: using a RANSAC algorithm to fit a straight line in the current point cloud, finding the inliers of the straight line, extracting these inliers and removing them from the original point cloud; b. continuing to fit the remaining point cloud: applying the RANSAC algorithm again on the remaining point cloud to fit the next straight line; c. repeating the process through the above steps until the number of points in the remaining point cloud is less than min_num; d. judging the straight line to which the point belongs: in each RANSAC algorithm fitting, whether each point belongs to the current straight line is judged by calculating the distance of each point to the fitted straight line, wherein only the inliers are segmented and removed from the original point cloud; e. expressing each straight line as a set of feature vectors; f. using the DBSCAN algorithm to cluster the feature vectors of these straight lines, if the number of neighbors of a straight line is greater than or equal to MinPts within a given ∈-neighborhood, the straight line is considered to be a core straight line, and the core straight line and the density-reachable straight lines constitute a cluster; g. continuing to cluster the remaining straight lines until all straight lines are classified; h. according to the classification result, screening the bridge part point cloud, deleting various interference point clouds, and only retaining the arch rib point cloud; i. extracting and fitting the key point coordinates of the screened point cloud.

2. The steel arch bridge arch rib linear extraction method based on the improved DBSCAN algorithm according to claim 1, characterized in that: The preprocessing comprises downsampling and denoising; The downsampling specifically comprises: using an Octree algorithm to downsample the point cloud data, taking the center of gravity of each Octant as a representative point to construct a new downsampled point cloud; wherein the Octree algorithm recursively divides a three-dimensional space into eight sub-cubes Octants; The denoising specifically comprises: using bilateral filtering to denoise the downsampled point cloud data.

3. The method for extracting the linear shape of the steel arch bridge rib based on the improved DBSCAN algorithm according to claim 1, characterized in that: The preprocessed point cloud data is preliminarily extracted, specifically including: using a PCA algorithm to estimate the normal of the point cloud, selecting the eigenvector corresponding to the maximum eigenvalue as the normal direction to obtain the normal direction of each point in the point cloud data set; setting the neighborhood point n = n0, and performing direction unification processing on the normal vector; according to the analysis result of the point cloud data, setting a threshold in the Z direction and the X direction, and extracting the point cloud at the bottom and both sides of the arch rib, respectively.

4. A steel arch bridge arch rib linear extraction system based on an improved DBSCAN algorithm, characterized in that: The method comprises a collecting unit, a preprocessing unit, a preliminary extraction unit and a secondary extraction unit; The collecting unit is configured to collect point cloud data of a steel arch bridge to obtain original point cloud data; The preprocessing unit is configured to preprocess the original point cloud data to obtain preprocessed point cloud data; The preliminary extraction unit is configured to preliminarily extract the preprocessed point cloud data to obtain a preliminary extraction result; The secondary extraction unit is configured to secondarily extract the preliminary extraction result by using an improved DBSCAN algorithm to obtain a final extraction result. The secondary extraction unit is configured to perform secondary extraction on the preliminary extraction result by using an improved DBSCAN algorithm to obtain a final extraction result. The improved DBSCAN algorithm is used to perform secondary extraction on the preliminary extraction result, and the method specifically comprises the following steps: a. fitting and segmenting the first straight line: using the RANSAC algorithm to fit a straight line in the current point cloud, finding the inliers of the straight line, extracting the inliers and removing them from the original point cloud; b. continue to fit the remaining point cloud: applying the RANSAC algorithm again on the remaining point cloud to fit the next straight line; c. repeat the process through the above steps until the number of points in the remaining point cloud is less than min_num; d. determine the straight line to which the point belongs: in each RANSAC algorithm fitting, determine whether the point belongs to the current straight line by calculating the distance from each point to the fitted straight line, wherein only the inliers will be segmented and removed from the original point cloud; e. express each straight line as a set of feature vectors; f. use the DBSCAN algorithm to cluster the feature vectors of these straight lines, and if the number of neighbors of a straight line is greater than or equal to MinPts within a given ∈-neighborhood, the straight line is considered to be a core straight line, and the core straight line and the density-reachable straight lines form a cluster; g. continue to cluster the remaining straight lines until all straight lines are classified; h. according to the classification result, filter the bridge part point cloud, delete various interference point clouds, and only keep the arch rib point cloud; i. extract and fit the key point coordinates of the filtered point cloud.

5. The steel arch bridge arch rib linear extraction system based on the improved DBSCAN algorithm according to claim 4, characterized in that: The preprocessing includes downsampling and denoising; The downsampling specifically comprises: using an Octree algorithm to downsample the point cloud data, taking the center of gravity of each Octant as a representative point, and constructing a new downsampled point cloud; wherein the Octree algorithm recursively divides a three-dimensional space into eight sub-cubes Octants; The denoising specifically comprises: using bilateral filtering to denoise the downsampled point cloud data.

6. The steel arch bridge arch rib linear extraction system based on the improved DBSCAN algorithm according to claim 4, characterized in that: The preliminary extraction on the preprocessed point cloud data specifically comprises: using the PCA algorithm to estimate the normal of the point cloud, selecting the eigenvector corresponding to the maximum eigenvalue as the normal direction to obtain the normal direction of each point in the point cloud data set; setting the neighborhood point n = n0, and performing direction unification processing on the normal vector; according to the analysis result of the point cloud data, setting a threshold in the Z direction and the X direction, and extracting the point cloud at the bottom and both sides of the arch rib, respectively.

Images