Large-span arch bridge arch rib point cloud noise adding method considering density change

Abstract

The invention discloses a large-span arch bridge arch rib point cloud noise adding method considering density change, and relates to the technical field of bridge engineering, and the method comprises the steps: obtaining original arch rib point cloud data, carrying out the resampling of the arch rib point cloud, and generating an interest point set; for any sampling point in the interest point set, a K-nearest neighbor (KNN) algorithm is adopted to determine nearest neighbor points, and a nearest neighbor point set is obtained; calculating the point cloud density between the arch rib point clouds according to the nearest neighbor point set; calculating a point cloud noise adding direction and a point cloud noise size according to the nearest neighbor point set; and according to the point cloud density, the point cloud noise adding direction and the point cloud noise size, adding point cloud noise points conforming to arch rib point cloud density characteristics. And a reliable data basis is provided for accurately evaluating the performance of the point cloud denoising algorithm in different density distribution scenes.

Description

Technical Field

[0001] This invention relates to the field of bridge engineering technology, and more specifically to a method for adding point cloud noise to the arch ribs of a long-span arch bridge that takes into account density variations. Background Technology

[0002] As a typical long-span load-bearing structure, the stability of the arch rib line of a steel-concrete composite arch bridge directly affects the overall operational safety of the bridge. Therefore, regular and precise deformation monitoring of the arch ribs is crucial. Three-dimensional laser scanning technology (also known as real-scene replication technology), with its advantages of non-contact operation, high precision, high sampling rate, and automation, has gradually replaced the traditional total station single-point measurement method, becoming the main technical means for acquiring massive three-dimensional point cloud data of the arch rib surface of steel-concrete composite arch bridges and conducting comprehensive deformation analysis. However, due to the combined influence of factors such as the measurement principle of the equipment, the surface material of the measured object, the atmospheric environment, and multipath effects, the original arch rib point cloud data acquired by ground-based three-dimensional laser scanning inevitably contains various noise points and outliers. To extract accurate structural features from the noisy data, point cloud denoising algorithms must be used for preprocessing. When developing and evaluating the performance of these point cloud denoising algorithms, it is crucial to construct a benchmark dataset containing "known noise" and "pure truth values". This means quantitatively adding noise to an ideal point cloud model without noise to simulate the real environment and accurately calculate the error and robustness of the denoising algorithm.

[0003] Existing methods for adding noise to point clouds of bridge structures mainly follow common computer vision noise models, including Gaussian noise, salt-and-pepper noise, and random outlier noise. Gaussian noise typically adds random, small offsets following a Gaussian distribution to the coordinates of each point in the point cloud to simulate random measurement errors by sensors; salt-and-pepper noise randomly replaces the coordinates of some points with extreme values ​​to simulate data transmission errors; and random outlier noise randomly generates isolated points in the free space within the point cloud bounding box to simulate interference from birds or dust in the environment. While these methods are simple to operate, they have significant limitations when applied to the specific scenario of long-span steel-concrete composite arch bridges. On the one hand, traditional random outliers are usually uniformly distributed in space, lacking the dependence on spatial structures, and cannot simulate the "ghosting" or "mixed pixel" noise common in laser scanning that is attached to the surface of objects; on the other hand, and most importantly, existing methods ignore the density variation characteristics of the point cloud.

[0004] Due to the enormous scale of long-span arch bridges and the limitations imposed by the scanning station's location, the point cloud over the arch ribs exhibits a significant "dense near, sparse far" characteristic. This means the point cloud density is extremely high near the scanning station, while it is sparse further away. Existing noise generation methods typically use globally uniform parameters. However, setting a fixed noise offset distance can lead to excessive noise in the high-density near-field regions, causing it to detach from the surface, while the noise is too small in the sparse far-field regions, becoming lost in the gaps between the point clouds. This fails to accurately reflect the physical characteristic of scanner resolution decreasing with increasing distance.

[0005] Therefore, how to provide a method for adding noise to the point cloud of the arch rib of a long-span arch bridge that takes into account density changes, and how to adaptively add noise points that conform to the distribution law of the arch rib point cloud based on the calculation of the actual local density characteristics of the point cloud, so as to provide high-quality test data that conforms to the actual situation for the point cloud denoising algorithm, is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] In view of this, the present invention provides a method for adding point cloud noise to the arch ribs of long-span arch bridges that considers density variations. Commonly used point cloud noise addition methods do not consider the density characteristics of the point cloud in long-span arch bridges. The point cloud is dense when the arch ribs are close to the scanning station, and sparse when they are far away. The method of the present invention adds point cloud noise points that conform to the density characteristics of the arch rib point cloud based on the calculation of the actual density characteristics of the point cloud.

[0007] To achieve the above objectives, the present invention adopts the following technical solution: a method for adding point cloud noise to the arch ribs of a long-span arch bridge considering density variations, comprising: Obtain the original arch rib point cloud data, resample the arch rib point cloud, and generate a set of points of interest; For any sampling point in the set of interest points The KNN algorithm is used to determine Find the nearest neighbor points to obtain the nearest neighbor set. ; According to the nearest neighbor set Calculate the point cloud density between the point clouds of the arch ribs; According to the nearest neighbor set Calculate the direction of point cloud noise addition and the magnitude of point cloud noise; Add point cloud noise points that match the point cloud density characteristics of the arch rib based on point cloud density, point cloud noise addition direction, and point cloud noise magnitude.

[0008] Preferably, the arch rib point cloud is resampled to generate a set of points of interest, including: the arch rib point cloud is uniformly sampled in layers, and the original point cloud is resampled according to a preset ratio to obtain the points that need to be noise-added, thereby generating a set of points of interest.

[0009] Preferably, based on the nearest neighbor set Calculating the point cloud density between the point clouds of the arch ribs includes: Calculate sampling points and search results The point cloud density is obtained by calculating the average Euclidean distance between the nearest neighbors.

[0010] Preferably, for each sampling point in the point of interest set The K-Nearest Neighbors (KNN) algorithm is used to search for the corresponding point in the original point cloud. The nearest neighbors form the nearest neighbor set. ; Calculate each sampling point Its corresponding nearest neighbor set The average Euclidean distance between all points is defined as the point cloud density. : ; in, For the first point in the neighborhood set One point, The value of reflects the sparsity of the point cloud at that location (the larger the value, the sparser the point cloud).

[0011] Preferably, based on the nearest neighbor set Calculate the direction of noise addition in the point cloud, including: Based on the nearest neighbor set Construct the covariance matrix The formula is as follows: ; in, The number of nearest neighbors. For the nearest neighbor set, the first One point, Let be the centroid of the nearest neighbor set; For the covariance matrix Perform eigenvalue decomposition and select the eigenvector corresponding to the smallest eigenvalue as the initial normal vector for that sampling point. .

[0012] Preferably, based on the nearest neighbor set Calculate the direction (normal vector) of point cloud noise addition, including: Based on the searched nearest neighbor set The covariance matrix was constructed using principal component analysis (PCA). Perform singular value decomposition (SVD) on the covariance matrix and select the eigenvector corresponding to the smallest eigenvalue as the sampling point. unoriented normal vector .

[0013] Preferably, the direction of the initial normal vector is adjusted for consistency, including: Based on sampling points and their nearest neighbor set Construct a local geometric topology model; Determine the external orientation of the structure: Final outward normal vector Opposite to the direction of the curvature center or interior region of the local geometric topology model; Calculate the judgment value : ; in, This is a reference vector pointing in the opposite direction from the sampling point to the internal region; like Then, the initial normal vector is reversed to obtain the final outward normal vector. ;like ,but .

[0014] Preferably, a reference direction vector is set. (For example, the direction vector pointing outward from the center of the structure, or the reverse vector of the viewpoint direction); calculate the dot product of the unoriented normal vector and the reference direction vector. If the dot product result is less than 0, then reverse the direction of the normal vector to ensure that noise is added to the outside of the arch rib structure. ; Preferably, the adjusted normal vector is normalized to obtain a unit normal vector. : ; Preferably, outlier noise points are generated based on local feature scale and unit normal vector, as shown in the following formula: First, calculate the offset magnitude of the noise. The modulus consists of two parts: a basic offset based on local feature scales and a Gaussian random perturbation. ; in, Sampling points The corresponding point cloud density, This is a preset outlier coefficient (used to control the degree to which noise is far from the surface). To follow a Gaussian distribution The random perturbation value; Finally, the coordinates of the generated outlier noise points are calculated. : .

[0015] As can be seen from the above technical solution, compared with the prior art, this invention discloses a method for adding noise to the point cloud of the arch ribs of a long-span arch bridge that considers density variations. This invention solves the problem that existing methods do not consider the "near-dense, far-sparse" characteristics of long-span arch bridges caused by changes in scanning distance. By introducing a local feature scale (average neighborhood distance), this invention can adaptively generate near-range noise in dense regions and far-range noise in sparse regions, and combine Gaussian perturbation to simulate real sensor errors. This method can generate noisy datasets with clear labels, which is beneficial for objectively evaluating the performance of denoising algorithms and providing high-quality data support for the construction of geometric twin models and deformation monitoring of long-span arch bridges. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0017] Figure 1(a) is a point cloud model of a noiseless arch rib provided in an embodiment of the present invention.

[0018] Figure 1(b) is a point cloud model of the 1% noise arch rib provided in the embodiment of the present invention.

[0019] Figure 1(c) is a point cloud model of the 3% noise arch rib provided in the embodiment of the present invention.

[0020] Figure 2 This is a schematic diagram of a method for adding point cloud noise to the arch ribs of a long-span arch bridge, taking into account density variations, as provided in an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] This invention discloses a method for adding point cloud noise to the arch ribs of a long-span arch bridge that takes into account density variations, such as... Figure 2 As shown, it includes: Obtain the original arch rib point cloud data, resample the arch rib point cloud, and generate a set of interest points for adding noise; For any sampling point in the set of interest points Determined by KNN algorithm Find the nearest neighbor points to obtain the nearest neighbor set. ; Based on the nearest neighbor set Calculate sampling points The local average spacing (used to characterize the point cloud density corresponding to any sampling point). Based on the nearest neighbor set Calculate sampling points The normal vector (direction of point cloud noise addition and magnitude of point cloud noise); Based on point cloud density and normal vector, the direction of noise addition and offset magnitude of point cloud are calculated to generate outlier points.

[0023] Specifically, the point cloud of the arch rib is resampled to generate a set of points of interest, including: The point cloud of the arch rib is sampled uniformly in layers, and the original point cloud is resampled according to a preset ratio to generate a set of points of interest.

[0024] Specifically, in the noise-free original arch rib point cloud dataset, random downsampling is performed according to a preset ratio (e.g., 1%, 3%, or 5%), and the sampled points are defined as sampling points. These sampling points will serve as the reference locations for generating outlier noise points.

[0025] Specifically, based on the nearest neighbor set Calculating the point cloud density between the point clouds of the arch ribs includes: Calculate sampling points and search results The point cloud density is obtained by calculating the average Euclidean distance between the nearest neighbors.

[0026] Specifically, the local average spacing of any sampling point is calculated and defined as the point cloud density, including: For any sampling point in the interest point set Using the K-Nearest Neighbors (KNN) search algorithm, the nearest neighbor to the given point is searched in the original point cloud. Points (e.g., take) ), forming the nearest neighbor set ; Calculate sampling points with nearest neighbor set The average Euclidean distance between all points is defined as the point cloud density. (This value reflects the sparsity of the point cloud; a larger spacing indicates a lower density.) ; in, For the first point in the neighborhood set One point, The value of reflects the sparsity of the point cloud at that location (the larger the value, the sparser the point cloud).

[0027] Specifically, based on the nearest neighbor set Calculate the direction of noise addition in the point cloud, including: Based on the nearest neighbor set Construct the covariance matrix The formula is as follows: ; in, The number of nearest neighbors. For the nearest neighbor set, the first One point, Let be the centroid of the nearest neighbor set; For the covariance matrix Perform eigenvalue decomposition and select the eigenvector corresponding to the smallest eigenvalue as the initial normal vector for that sampling point. .

[0028] Specifically, based on the nearest neighbor set Calculate the direction of noise addition in the point cloud, including: Principal component analysis (PCA) is used to estimate sampling points. The normal vector.

[0029] First, calculate the nearest neighbor set. center of mass : ; Next, construct the covariance matrix. : ; For covariance matrix Eigenvalue decomposition yields three eigenvalues. and its corresponding eigenvectors. Select the smallest eigenvalue. The corresponding feature vector is used as the initial normal vector of the sampling point. .

[0030] Specifically, the direction of the initial normal vector is adjusted for consistency, including: Based on sampling points and their nearest neighbor set Construct a local geometric topology model; Determine the external orientation of the structure: Final outward normal vector Opposite to the direction of the curvature center or interior region of the local geometric topology model; Calculate the judgment value : ; in, This is a reference vector pointing in the opposite direction from the sampling point to the internal region; like Then, the initial normal vector is reversed to obtain the final outward normal vector. ;like ,but .

[0031] Specifically, the initial normal vector is adjusted for directional consistency to ensure that noise propagates outward: To ensure that the generated noise points are located outside (or inside, depending on requirements) the arch rib structure, the normal vector direction needs to be redirected. A reference direction vector is defined. (Example: a vector pointing outwards from the center of the arch rib section, or a line-of-sight vector reversed based on the scanner's viewpoint). Calculate the dot product of the initial normal vector and the reference vector. ; like Then the final normal vector ;otherwise .at this time, This is the normalized outward normal vector.

[0032] Specifically, generating point cloud outlier noise includes: Based on the local average spacing and the preset difficulty coefficient, combined with Gaussian random perturbation, the final coordinates of the noise points are calculated.

[0033] First, calculate the noise offset magnitude along the normal vector direction. : ; in, The local average spacing is obtained from the aforementioned calculation; The preset outlier difficulty coefficient ( The larger the value, the farther the outlier is from the surface, and the more difficult it is to denoise. To follow a Gaussian distribution The random perturbation value is used to increase the randomness of the noise.

[0034] Finally, the coordinates of the outlier points are generated based on the offset magnitude and the normal vector. : .

[0035] Repeat the above steps until noise generation is completed for all sampling points. Add the generated outliers to the original point cloud to complete the construction of the noisy dataset.

[0036] In one specific embodiment of the present invention, in order to verify the accuracy and efficiency of the filtering algorithm proposed in the embodiment of the present invention, based on curves... , A noise-free point cloud model of the arch rib with a radius of 0.46 m was generated, as shown in Figures 1(a), 1(b), and 1(c). By controlling the number of points on the circumference and the spacing between sections, the point cloud density of the arch rib was gradually reduced from the arch foot to the arch top, with the maximum point cloud density being 521 points / m. 2 The minimum point cloud density is 110 points / m². 2 The arch rib point cloud model contains 49,732 points. Randomly distributed outliers (1% and 3%) were added to the outer surface of the noise-free arch rib point cloud model, respectively. The outliers were added to the point cloud data using the following method: First, 1% and 3% of the total points were randomly selected from the arch rib point cloud; then, the nearest neighbor set of each sampled point was calculated. The value of k is 10; further, based on the nearest neighbor set... Point cloud density and normal vector are calculated separately; finally, outliers are generated on the surface of the arch rib along the normal vector direction. The outliers conform to the point cloud density distribution characteristics at the corresponding sampling points, and the severity of outlier addition is controlled by adding Gaussian perturbation.

[0037] Compared with conventional point cloud noise addition methods, such as Gaussian noise, salt-and-pepper noise, and outlier noise, the proposed method adds arch rib point cloud noise by considering not only the density variation of the arch rib point cloud, but also the direction of adding outliers and the severity of the outlier addition, resulting in noise that is more consistent with the actual situation.

[0038] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0039] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for adding point cloud noise to the arch ribs of a long-span arch bridge considering density variations, characterized in that, Includes the following steps: Obtain the original arch rib point cloud data, resample the arch rib point cloud, and generate a set of points of interest; For any sampling point in the set of interest points The KNN algorithm is used to determine Find the nearest neighbor points to obtain the nearest neighbor set. ; According to the nearest neighbor set Calculate the point cloud density between the point clouds of the arch ribs; According to the nearest neighbor set Calculate the direction of point cloud noise addition and the magnitude of point cloud noise; Add point cloud noise points that match the point cloud density characteristics of the arch rib based on point cloud density, point cloud noise addition direction, and point cloud noise magnitude.

2. The method for adding point cloud noise to the arch ribs of a long-span arch bridge considering density variations according to claim 1, characterized in that, The arch rib point cloud is resampled to generate a set of points of interest, including: the arch rib point cloud is uniformly sampled in layers, and the original point cloud is resampled according to a preset ratio to generate a set of points of interest.

3. The method for adding point cloud noise to the arch ribs of a long-span arch bridge considering density variations according to claim 1, characterized in that, According to the nearest neighbor set Calculating the point cloud density between the point clouds of the arch ribs includes: Calculate sampling points and search results The point cloud density is obtained by calculating the average Euclidean distance between the nearest neighbors.

4. The method for adding point cloud noise to the arch ribs of a long-span arch bridge considering density variations according to claim 1, characterized in that, According to the nearest neighbor set Calculate the direction of noise addition in the point cloud, including: Based on the nearest neighbor set Construct the covariance matrix The formula is as follows: ； in, The number of nearest neighbors. For the nearest neighbor set, the first One point, Let be the centroid of the nearest neighbor set; For the covariance matrix Perform eigenvalue decomposition and select the eigenvector corresponding to the smallest eigenvalue as the initial normal vector for that sampling point. .

5. The method for adding point cloud noise to the arch ribs of a long-span arch bridge considering density variations according to claim 4, characterized in that, The direction of the initial normal vector is adjusted for consistency, including: Based on sampling points and their nearest neighbor set Construct a local geometric topology model; Determine the external orientation of the structure: Final outward normal vector Opposite to the direction of the curvature center or interior region of the local geometric topology model; Calculate the judgment value : ； in, This is a reference vector pointing in the opposite direction from the sampling point to the internal region; like Then, the initial normal vector is reversed to obtain the final outward normal vector. ;like ,but .

6. The method for adding point cloud noise to the arch ribs of a long-span arch bridge considering density variations according to claim 5, characterized in that, The point cloud noise value is generated based on the normalized normal vector, as shown in the following formula: ； in, The coordinates of the generated outlier points, These are the original sampling point coordinates. This is the normalized outward normal vector; The noise displacement modulus.

7. The method for adding point cloud noise to the arch ribs of a long-span arch bridge considering density variations according to claim 6, characterized in that, The formula for calculating the noise displacement modulus is: ； In the formula, Based on sampling points p The corresponding point cloud density; The preset outlier difficulty coefficient; To follow a Gaussian distribution The random perturbation value.

Images