A method for road full section splicing based on two-dimensional wavelet decomposition

Abstract

This invention discloses a method for stitching together full-section road surfaces based on two-dimensional wavelet decomposition, comprising: S1, installing point cloud data acquisition equipment to acquire continuous point cloud data of the road surface; S2, preprocessing the acquired continuous point cloud data of the road surface; S3, performing two-dimensional rasterization processing on the preprocessed point cloud data; S4, performing two-dimensional wavelet decomposition on the point cloud data to extract the texture features of the point cloud data within each raster; S5, comparing the similarity of the feature vectors of each raster between point cloud data, and extracting matching point pairs in the overlapping areas of the point cloud data based on the similarity; S6, using the matching point pairs as a basis, solving the rotation and translation matrix of the point cloud data based on singular value decomposition (SVD), and stitching together a series of continuously captured point clouds. According to this invention, by performing two-dimensional wavelet decomposition on the acquired continuous point cloud data to obtain the feature vectors of the point cloud, and using these to obtain matching point pairs between point clouds, accurate stitching of road surface point cloud data is achieved without the need for manual placement of markers.

Description

Technical Field

[0001] This invention relates to the technical field of three-dimensional reconstruction of road surface structures, and in particular to a method for splicing full-section road sections based on two-dimensional wavelet decomposition. Background Technology

[0002] Reconstructing the 3D morphology of road surfaces using point cloud data is crucial for refined detection of road defects and for data-driven road maintenance strategies. While existing 3D cameras, such as structured light cameras, can acquire point cloud data of road surfaces, their limited single-shot range hinders the complete capture of the entire road surface's 3D morphology. Therefore, it is necessary to stitch the acquired point cloud data to achieve 3D road reconstruction. Existing 3D reconstruction methods primarily address scenarios related to autonomous driving or buildings, where distinguishable feature points, such as edge and corner points, are readily available. However, flat road surfaces lack unique feature points, posing a challenge to 3D reconstruction in road surface scenarios. Current research involves manually placing markers and manually or automatically extracting visible marker points from at least two scans as feature points for point cloud registration. However, this process is often limited by real-world conditions and may obstruct or damage the true point cloud data of the target object's surface, increasing post-processing time. Therefore, compared to the low efficiency and large errors of existing stitching and reconstruction methods, a road full-section stitching method based on 2D wavelet decomposition offers the possibility of rapid, automatic, and low-cost 3D road reconstruction. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide a road full-section stitching method based on two-dimensional wavelet decomposition. This method involves performing two-dimensional wavelet decomposition on collected continuous point cloud data to obtain feature vectors of the point clouds, and then using these vectors to obtain matching point pairs between point clouds, achieving accurate stitching of road surface point cloud data without the need for manual marker placement. To achieve the above-mentioned objective and other advantages of the present invention, a road full-section stitching method based on two-dimensional wavelet decomposition is provided, comprising the following steps:

[0004] S1. Install point cloud data acquisition equipment to obtain continuous point cloud data of the road surface;

[0005] S2. Preprocess the collected continuous point cloud data of the road surface;

[0006] S3. Perform two-dimensional rasterization processing on the x,y plane of the preprocessed point cloud data;

[0007] S4. Perform two-dimensional wavelet decomposition on the rasterized point cloud data, extract the texture features of the point cloud data in each raster, and represent them in the form of feature vectors.

[0008] S5. Compare the similarity of the feature vectors of each grid in the point cloud data, and extract matching point pairs in the overlapping areas of the point cloud data based on the similarity.

[0009] S6. Using the matching point pairs as a basis, solve the rotation and translation matrix of the point cloud data based on singular value decomposition (SVD) to stitch together a series of continuously captured point clouds.

[0010] Preferably, the continuous point cloud data of the road surface in step S1 includes depth information, that is, the distance between the collected point and the collection device, and there is a certain overlap area between two consecutive point cloud frames.

[0011] Preferably, the preprocessing of the point cloud data in step S2 includes slope removal, specifically including the following steps:

[0012] S21. Obtain the 3D coordinate set O{(x1,y1,z1),(x2,y2,z2),…,(x...} of all point cloud data. n ,y n ,z n )};

[0013] S22. Use multiple linear regression to fit the depth coordinate z in the point cloud data and obtain the regression coefficients b0, b1, b2.

[0014] S23. Calculate the set of three-dimensional coordinates of the point cloud data after slope removal, as follows:

[0015] O'{(x1,y1,z1-b0-b1x1-b2y1),(x2,y2,z2-b0-b1x2-b2y2),…,(x n ,y n ,z n -b0-b1x n -b2y n )}.

[0016] Preferably, step S3 specifically involves establishing a grid in the x and y directions of the point cloud data at intervals of a certain coordinate step size, mapping the continuous point cloud data into a discrete grid space, thereby dividing the space into a series of regular grids, and performing linear interpolation on the z coordinate of the point falling within each grid to obtain the representative value of that grid.

[0017] Preferably, in step S4, the point cloud data is decomposed into two-dimensional wavelet decomposition to extract the texture features of the point cloud data in each grid cell. The specific steps are as follows: select wavelet basis functions; use the Matlab wavelet toolbox to decompose the rasterized matrix-form point cloud data into two-dimensional wavelet decomposition, decompose the road texture information into N scales in the x and y directions, where the first N-1 scales represent detail signals and the last scale represents residual signals, thus forming an N×N wavelet decomposition structure; each grid cell of each point cloud data cell obtains a 1*9 feature vector.

[0018] Preferably, the extraction of matching point pairs in the overlapping area of ​​the point cloud data in step S5 specifically includes the following steps:

[0019] S51. Use Euclidean distance, Manhattan distance, cosine similarity, and similarity coefficient as similarity indicators.

[0020] S52. For every two point cloud data, calculate four similarity indices between each raster feature vector of the source point cloud data and all raster feature vectors of the target point cloud data to obtain four similarity index matrices.

[0021] S53. For each grid, the four similarity index matrices are reasonably filtered, and the point pairs that simultaneously meet the threshold of the four similarity indexes are extracted as matching point pairs.

[0022] Preferably, step S53 specifically includes the following steps:

[0023] S531. Set reasonable thresholds for the four indicators. The thresholds should meet the following requirements: obtain matching point pairs in the overlapping area as accurately as possible to reduce the possibility of matching wrong points; the number of matching point pairs that meet the conditions should not be less than the minimum number of 6.

[0024] S532. During the threshold setting process, when it is necessary to adjust the threshold, the specific adjustment steps include: when there are too many matching point pairs that meet the conditions, tighten the threshold in order to reduce the error of false matching; when there are insufficient matching point pairs that meet the conditions, relax the threshold.

[0025] S533. For a certain grid cell in the source point cloud data, if there is more than one point in the target point cloud data that meets the threshold condition, then the point with the higher cosine similarity is taken as the matching point.

[0026] Preferably, step S6 specifically includes the following steps:

[0027] S61. Obtain the set of matching point pairs P = {p1, p2, ..., p...} of the source point cloud data and the target point cloud data through step S5. n} and Q={q1,q2,…,q n}, where (p i ,qi () represents a matching point pair;

[0028] S62. Decentralization: Subtract the centroid of each point from the set of matching point pairs in the source and target point cloud data. The centroid calculation method is as follows:

[0029] S63. Construct the covariance matrix H = P'Q' T ;

[0030] S64. Perform singular value decomposition on the covariance matrix H to obtain the singular value matrix, the eigenvector matrix U of the source point cloud data, and the eigenvector matrix V of the target point cloud data: H = U∑V T ;

[0031] S65. Calculate the rotation matrix R = VU T Translation matrix

[0032] Compared with existing technologies, the advantages of this invention are as follows: It uses a 3D camera to acquire point cloud data of a certain length of road surface through continuous shooting; it transforms the point cloud data into matrix form using a 2D rasterization method; it extracts feature vectors from the point cloud raster through 2D wavelet decomposition; and it uses the similarity of feature vectors as a basis to solve for matching point pairs between point cloud data, thus achieving point cloud stitching. This invention enables rapid and low-cost 3D reconstruction of road surface morphology using low-cost point cloud acquisition methods, eliminating the need for manually setting stitching targets and reducing reliance on manual labor or high-precision equipment. This is of great significance for high-frequency digital detection of road surface performance. Attached Figure Description

[0033] Figure 1 The flowchart below shows the road full-section splicing method based on two-dimensional wavelet decomposition according to the present invention.

[0034] Figure 2 This is a schematic diagram of point cloud data for slope removal, based on an embodiment of the road full-section stitching method based on two-dimensional wavelet decomposition according to the present invention.

[0035] Figure 3 This is a schematic diagram of the local rasterization result of point cloud data according to an embodiment of the road full-section stitching method based on two-dimensional wavelet decomposition according to the present invention;

[0036] Figure 4 This is a schematic diagram of the two-dimensional wavelet decomposition result according to an embodiment of the road full-section splicing method based on two-dimensional wavelet decomposition according to the present invention;

[0037] Figure 5 This is a schematic diagram of four similarity index matrices for an embodiment of the road full-section splicing method based on two-dimensional wavelet decomposition according to the present invention;

[0038] Figure 6 This is a schematic diagram showing the comparison before and after the completion of the road full-section splicing method based on two-dimensional wavelet decomposition according to an embodiment of the present invention. Detailed Implementation

[0039] 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.

[0040] Reference Figure 1-6 A method for stitching together the full cross-section of a road based on two-dimensional wavelet decomposition includes the following steps: Step 1, data acquisition: Install a 3D camera or other acquisition equipment on the inspection vehicle and take pictures of the road surface from a top-down angle to acquire continuous road surface point cloud data. The road surface point cloud data must contain at least depth information, i.e., the distance between the acquired point and the acquisition equipment, and there must be a certain overlap area between two consecutive frames of point cloud.

[0041] The second step, preprocessing: The preprocessing of the road surface point cloud data for slope removal specifically includes the following steps:

[0042] Take the three-dimensional coordinate set O{(x1,y1,z1),(x2,y2,z2),…,(x n ,y n ,z n )};

[0043] By fitting set O using the least squares method, the three-dimensional plane parameters are obtained as b0 = 485.573, b1 = 0.027, and b2 = 0.005.

[0044] Calculate the three-dimensional coordinate set O'{(x1,y1,z1-b0-b1x1-b2y1),(x2,y2,z2-b0-b1x2-b2y2),…,(x n ,y n ,z n -b0-b1x n -b2y n )}, comparing point cloud data before and after removing slope, for example Figure 2 As shown.

[0045] The third step is rasterization: a grid is created in the x and y directions with an interval of 1 coordinate. The preprocessed point cloud data is mapped to a discrete grid space, thus dividing the space into a series of regular grids. The z-coordinate of the points falling within each grid is linearly interpolated. The local result after rasterization is as follows: Figure 3 As shown.

[0046] The fourth step is two-dimensional wavelet decomposition and feature vector extraction: sym4 is selected as the wavelet basis function for wavelet decomposition. Using the Matlab wavelet toolbox, the rasterized matrix-form point cloud data is decomposed into two-dimensional wavelet decomposition, decomposing the road texture information into 9 scales in the x and y directions. Eight of these scales represent detail signals, and the last scale represents the residual signal, thus forming a 9x9 wavelet decomposition structure. The two-dimensional wavelet decomposition result is shown below. Figure 4 As shown, each raster of each point cloud data point obtains a 1*9 feature vector.

[0047] The fifth step involves comparing the similarity of the feature vectors of each grid cell in the point cloud data, and extracting matching point pairs from the overlapping areas of the point cloud data based on the similarity. The calculation method is as follows:

[0048] Euclidean distance, Manhattan distance, cosine similarity, and similarity coefficient are chosen as similarity indicators. The formulas for calculating Euclidean distance E_d, Manhattan distance M_d, cosine similarity C_S, and similarity coefficient C_C are as follows:

[0049]

[0050]

[0051] C_S=(a·b)(||a||*||b||)

[0052]

[0053] Where a and b are the feature vectors of the point clouds in the two frames, respectively.

[0054] For every two point cloud data sets, four similarity indices are calculated between the feature vector of each raster in the source point cloud data and all raster feature vectors in the target point cloud data, resulting in four similarity index matrices, such as... Figure 5 As shown;

[0055] For each raster, the four similarity index matrices are reasonably filtered, and point pairs that simultaneously meet the thresholds of the four similarity indices are extracted as matching point pairs. Specifically, reasonable thresholds are set for the four indices, which need to meet the following requirements: to obtain matching point pairs as accurately as possible in the overlapping area and reduce the possibility of matching false points; the number of matching point pairs that meet the conditions is not less than the minimum number of 6; during the threshold setting process, it may be necessary to adjust the thresholds. The adjustment steps include: when there are too many matching point pairs that meet the conditions, tighten the threshold to reduce the error of false matching; when there are insufficient matching point pairs that meet the conditions, relax the threshold; for a certain raster in the source point cloud data, if there is more than one point in the target point cloud data that meets the threshold conditions, the point with the higher cosine similarity is selected as the match.

[0056] The sixth step involves using the matched point pairs as a basis to solve for the rotation and translation matrix of the point cloud data based on Singular Value Decomposition (SVD), and then stitching together a series of continuously captured point clouds. The calculation method is as follows:

[0057] Decentralization involves matching the source point cloud data and the target point cloud data into a point set P = {p1, p2, ..., p...}. n} and Q={q1,q2,…,q n Subtract the centroids of each i ∈ {1,2,…,n}: The centroid calculation method is as follows:

[0058] Construct the covariance matrix H = P'Q' T ;

[0059] Singular value decomposition (SVD) is performed on the covariance matrix H to obtain the singular value matrix, the eigenvector matrix U of the source point cloud data, and the eigenvector matrix V of the target point cloud data: H = U∑V T ;

[0060] Calculate the rotation matrix R = VU T Translation matrix

[0061] The completed splicing result is as follows Figure 6 As shown.

[0062] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention, and applications, modifications and variations thereof will be apparent to those skilled in the art.

[0063] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A method for road full section splicing based on two-dimensional wavelet decomposition, characterized in that, Includes the following steps: S1. Install point cloud data acquisition equipment to obtain continuous point cloud data of the road surface; S2. Preprocess the collected continuous point cloud data of the road surface; S3. Perform two-dimensional rasterization processing on the x and y planes on the preprocessed point cloud data. The specific steps of step S3 are as follows: establish a grid in the x and y directions of the point cloud data with a certain coordinate step size, map the continuous point cloud data into a discrete grid space, thereby dividing the space into a series of regular grids, and perform linear interpolation on the z coordinate of the point falling in each grid to obtain the representative value of the grid. S4. Perform two-dimensional wavelet decomposition on the rasterized point cloud data to extract the texture features of the point cloud data in each raster and represent them in the form of feature vectors. The specific steps for performing two-dimensional wavelet decomposition on the point cloud data in step S4 to extract the texture features of the point cloud data in each raster are as follows: Select wavelet basis functions; Use the Matlab wavelet toolbox to perform two-dimensional wavelet decomposition on the rasterized matrix-form point cloud data, decompose the road texture information into N scales in the x and y directions, where the first N-1 scales represent detail signals and the last scale represents residual signals, thus forming an N×N wavelet decomposition structure; Each raster of each point cloud data obtains a 1*9 feature vector. S5. Compare the similarity of the feature vectors of each grid in the point cloud data, and extract matching point pairs in the overlapping areas of the point cloud data based on the similarity. S6. Using the matched point pairs as a basis, solve the rotation and translation matrix of the point cloud data based on Singular Value Decomposition (SVD) to stitch together a series of continuously captured point clouds; Step S6 specifically includes the following steps: S61, obtaining the matched point pair set of the source point cloud data and the target point cloud data through step S5 with wherein is a matched point pair; S62, decentralization, subtract the matching point pair set of the source point cloud data and the target point cloud data from the respective centroids: , ; wherein the centroid calculation method is , ; S63, construct the covariance matrix ; S64, singular value decomposition is performed on the covariance matrix H to obtain a singular value matrix, a feature vector matrix U of the source point cloud data, and a feature vector matrix V of the target point cloud data: ; S65. Calculate the rotation matrix Translation matrix .

2. The method for stitching together full road cross-sections based on two-dimensional wavelet decomposition as described in claim 1, characterized in that, In step S1, the continuous point cloud data of the road surface includes depth information, that is, the distance between the collected point and the collection device, and there is a certain overlap area between two consecutive frames of point cloud.

3. The method for stitching together full road cross-sections based on two-dimensional wavelet decomposition as described in claim 1, characterized in that, Step S2 involves preprocessing the point cloud data, including slope removal. Specifically, slope removal includes the following steps: S21. Obtain the three-dimensional coordinate set of all point cloud data. ; S22. Use multiple linear regression to fit the depth coordinate z in the point cloud data and obtain the regression coefficients. , , ; S23. Calculate the set of three-dimensional coordinates of the point cloud data after slope removal, as follows: 。 4. The method for stitching together full road cross-sections based on two-dimensional wavelet decomposition as described in claim 1, characterized in that, Step S5, the extraction of matching point pairs in the overlapping areas of the point cloud data, specifically includes the following steps: S51. Use Euclidean distance, Manhattan distance, cosine similarity, and similarity coefficient as similarity indicators. S52. For every two point cloud data, calculate four similarity indices between each raster feature vector of the source point cloud data and all raster feature vectors of the target point cloud data to obtain four similarity index matrices. S53. For each grid, the four similarity index matrices are reasonably filtered, and the point pairs that simultaneously meet the threshold of the four similarity indexes are extracted as matching point pairs.

5. The method for splicing full road cross-sections based on two-dimensional wavelet decomposition as described in claim 4, characterized in that, Step S53 specifically includes the following steps: S531. Set reasonable thresholds for the four indicators. The thresholds should meet the following requirements: obtain matching point pairs in the overlapping area as accurately as possible to reduce the possibility of matching wrong points; the number of matching point pairs that meet the conditions should not be less than the minimum number of 6. S532. During the threshold setting process, when it is necessary to adjust the threshold, the specific adjustment steps include: when there are too many matching point pairs that meet the conditions, tighten the threshold in order to reduce the error of false matching; when there are insufficient matching point pairs that meet the conditions, relax the threshold. S533. For a certain grid cell in the source point cloud data, if there is more than one point in the target point cloud data that meets the threshold condition, then the point with the higher cosine similarity is taken as the matching point.

Images