A method for automatically identifying a reinforcement mesh based on unmanned aerial vehicle laser point cloud data

Abstract

The application relates to the technical field of electric digital data processing, and particularly discloses a steel mesh automatic recognition method based on unmanned aerial vehicle laser point cloud data, which comprises the following steps: acquiring point cloud data of a steel mesh in a tunnel or underground structure during construction, pre-processing original point cloud, plane segmentation, steel layer detection, steel axis extraction, steel direction recognition, interlayer spacing geometry parameter calculation, result output and the like. The application realizes full-automatic recognition through unmanned aerial vehicle laser scanning technology combined with an intelligent point cloud processing algorithm, improves the accuracy of a detection result, overcomes problems of low efficiency and insufficient accuracy of a traditional method which relies on manual measurement, and facilitates high-precision and high-efficiency recognition of a steel mesh in a tunnel or underground structure during construction, thereby providing a more reliable quality evaluation basis.

Description

Technical Field

[0001] This invention relates to the field of electronic digital data processing technology, and in particular to an automated method for identifying steel mesh based on UAV laser point cloud data. Background Technology

[0002] The information disclosed in this background section is intended only to enhance understanding of the overall background of the invention and is not necessarily to be construed as an admission or in any way implying that such information constitutes prior art known to those skilled in the art.

[0003] In the construction of lining structures for underground engineering projects such as tunnels and subway stations, the installation quality of the reinforcing mesh directly and profoundly affects the overall safety and long-term durability of the structure. As the core skeleton of the concrete structure, the accuracy of the mesh's positioning, the uniformity of its spacing, and the strength of its overlaps and ties all determine whether the lining can effectively withstand the earth pressure, water pressure, and long-term dynamic loads from train operation. If installation deviations occur, such as excessive spacing, insecure tying, or insufficient protective layer thickness, it can easily lead to concrete cracking, reduced load-bearing capacity, and even localized spalling or structural damage, seriously threatening the operational safety and service life of the project. Therefore, the installation of the reinforcing mesh must be strictly controlled as a critical process to lay a solid foundation for the safety and stability of the overall structure.

[0004] The traditional method of inspecting the installation quality of steel mesh mainly relies on manual sampling measurement, but it has the following shortcomings: First, the inspection efficiency is low. Scaffolding needs to be erected for point-by-point measurement, which is not only time-consuming, but also prone to accidents due to high-altitude operations. Second, the coverage is insufficient. Since the sampling rate is often less than 10%, it is difficult to comprehensively assess the quality. At present, some inspection solutions have adopted automated technology, including: (1) Image-based methods: This method is easily affected by lighting and occlusion interference, and the steel bar identification error rate exceeds 30%. (2) Point cloud-based methods: Although the accuracy is high, it faces technical bottlenecks: the uneven density of point clouds in the steel bar intersection area leads to feature confusion. (3) Traditional clustering algorithms: This method has poor adaptability to sparse point clouds, with a false negative rate of over 25%; moreover, the direction recognition depends on manually preset parameters and cannot adaptively distinguish the horizontal and vertical distribution of steel bars. Summary of the Invention

[0005] This invention proposes an automated rebar mesh identification method based on UAV laser point cloud data. It achieves fully automated identification by combining UAV laser scanning technology with intelligent point cloud processing algorithms, improving the accuracy of the detection results and overcoming the problems of low efficiency and insufficient accuracy caused by traditional methods relying on manual measurement. This facilitates high-precision and high-efficiency identification of rebar mesh during tunnel or underground structure construction, providing a more reliable basis for quality assessment. Specifically, the technical solution of this invention is as follows.

[0006] An automated method for identifying rebar mesh based on UAV laser point cloud data includes the following steps:

[0007] S1. Use the laser scanning equipment mounted on the drone to acquire the original data of the steel reinforcement mesh point cloud for later use.

[0008] S2. The original point cloud data is preprocessed using voxel mesh downsampling technology to obtain a preprocessed point cloud with reduced density. Then, the normal vector of the preprocessed point cloud is calculated to obtain the unit normal vector. Then, outliers are filtered out, and the transformation matrix from the coordinate system of the remaining point cloud to the local construction coordinate system is calculated.

[0009] S3. Iteratively fit the tunnel sidewall plane from the preprocessed point cloud data, and calculate the distance from each point in the preprocessed point cloud to the candidate plane. Then, the planar model with the most interior points is obtained. ,spare.

[0010] S4. Calculate the distance from each non-tunnel sidewall plane point to the plane model. The distances are used to form a one-dimensional distance dataset. Then, clustering is performed on this dataset to obtain clusters. After filtering, the remaining clusters are sorted by depth to obtain a pure rebar layer point cloud set for later use.

[0011] S5. Perform single-rebar separation processing on the point cloud set of the pure rebar layer to obtain point cloud clusters of individual rebars. Then, perform validity verification on each point cloud cluster, and perform principal component analysis on the verified point cloud clusters to calculate the principal direction vector of the rebar axis and the linear equation of its axis in the three-dimensional coordinate system.

[0012] S6. Calculate the cosine of the angle between the principal direction vector and the basis vector of the steel bar axis, and output the three-dimensional coordinates of the axis, the principal direction vector, and the distribution direction type of the steel bar in each layer of steel bars.

[0013] S7. Based on the calculation results of S6 above, calculate the spacing of steel bars in the same layer and the spacing of steel bars between layers respectively. Then output and display the original point cloud, steel bar point cloud, and visualization results of steel bar axis and layer relationship, and output the steel bar spacing qualification rate, spacing distribution histogram, and steel bar position deviation, spacing deviation and direction deviation.

[0014] Furthermore, in step S2, the adaptive voxel size of the voxel grid downsampling technique... The following formula (1) is used to calculate:

[0015] (1);

[0016] In the above formula (1), This is an adjustment factor, with a value ranging from 1.5 to 3.0. This represents the original point cloud density.

[0017] Further, in step S2, the method for calculating the normal vector of the preprocessed point cloud includes the following steps:

[0018] S2.1. The neighborhood of each point to be calculated in the preprocessed point cloud is determined by the K-Nearest Neighbors (K-NN) algorithm to obtain the neighborhood point set.

[0019] S2.2 Calculate the covariance matrix of the neighborhood point set using the following formula (2). C :

[0020] (2);

[0021] In the above formula (2), K =30~50, which represents the number of neighboring points considered when estimating the normal vector of a single point. For the th neighborhood point set j The three-dimensional coordinates of each point The three-dimensional arithmetic mean center of each point in the neighborhood point set is obtained by the following equation (3):

[0022] (3);

[0023] S2.3, Then, the covariance matrix is ​​obtained by the following equation (4). C Eigenvalue decomposition yields three eigenvalues. and the corresponding three mutually orthogonal eigenvectors .

[0024] (4);

[0025] S2.4, the smallest eigenvalue Corresponding feature vector That is, the normal vector direction of the preprocessed point cloud; then the... The unit normal vector can be obtained by normalization. .

[0026] Furthermore, in step S2, the outlier point is determined using the following method: if the unit normal vector If the angle between the point and the average normal vector of its neighborhood is ≥30°, then the point is considered a valid point. This is an outlier.

[0027] Furthermore, in step S3, the number of iterations required is first calculated using the following formula (5). Then, the RANSAC algorithm is used to iteratively fit the tunnel sidewall plane.

[0028] (5);

[0029] In the above formula (5), P The expected confidence probability is usually taken as P =0.95~1, w The estimated interior point ratio is usually taken as... w= 0.3~0.6, m This represents the minimum number of sample points required to fit the model.

[0030] Further, in step S3, the planar model It is obtained using the following method:

[0031] S3.1. Randomly select three non-collinear points from the preprocessed point cloud, and then obtain a candidate plane based on these three points. The plane equation is expressed as:

[0032] .

[0033] S3.2 Calculate the straight-line distance from each point in the preprocessed point cloud to the candidate plane using the following formula (6). :

[0034] (6);

[0035] In the above formula (6), the a , b , c , d All are planar models The parameters of the plane equation are as shown in step S3.1 above.

[0036] S3.3, The straight-line distance obtained statistically. Medium less than the threshold The points are used to form the "interior point set" for this iteration, and the total number of interior points contained therein is calculated. The threshold is the optimal number of interior points in the current candidate plane. Optionally, the threshold... The adjustment is adaptive based on engineering design parameters, point cloud density, etc., and is usually selected within the range of 3cm to 10cm.

[0037] S3.4. Compare the current optimal number of interior points in the candidate plane with the optimal number of interior points obtained in the previous iteration. If the former is greater than the latter, replace the latter with the optimal number of interior points, and further use the plane model obtained in this iteration to replace the plane model of the previous iteration; the iteration... After this process, the final optimal number of interior points is obtained, and the corresponding candidate planes are used as the plane model. .

[0038] Further, in step S4, each non-tunnel sidewall plane point... The three-dimensional coordinates are Its to the plane model straight-line distance d i The one-dimensional distance dataset is obtained by calculating using the following formula (7). .

[0039] (7);

[0040] In the above formula (7), the a , b , c , d All are planar models The parameters of the plane equation are as shown in step S3.1 above.

[0041] Furthermore, in step S4, the DBSCAN algorithm is used to directly cluster the one-dimensional distance dataset, and the formula for calculating the cluster radius is shown in the following equation (8):

[0042] (8);

[0043] In the above formula (8), the k This is the proportionality coefficient, and its value ranges from 0.3 to 1. k ≤0.5, the It refers to the design spacing of the reinforcing bars.

[0044] Further, in step S4, clusters with fewer than 30 point clouds are removed / filtered out, and the remaining clusters constitute the effective rebar layer point cloud set. These are then sorted in ascending or descending order of depth to obtain the pure rebar layer point cloud set.

[0045] Furthermore, in step S5, the DBSCAN algorithm is used to process each rebar layer point cloud in the pure rebar layer point cloud set. L k Two-dimensional clustering is performed, and the clusters are projected onto the planar model from the point cloud of the reinforcement layer. The process is performed on the subsequently obtained two-dimensional coordinates.

[0046] Furthermore, in step S5, the validity verification is based on the following criteria: the number of points within a cluster must be no less than 3, and the length of the cluster point cloud calculated based on the principal component analysis must be ≥0.01m. Clusters that cannot meet both of these conditions will be considered noise and removed / filtered out.

[0047] Furthermore, in step S5, when the number of steel bars extracted by the principal component analysis is less than 60% of the expected number, the backup method is triggered.

[0048] Furthermore, in step S6, the included angle cosine includes the principal direction vector of the reinforcing bar axis. On the basis vectors respectively , Cosine value of the projected angle on It is calculated using the following formula (9):

[0049] (9);

[0050] In the above equation (9), the basis vector The basis vectors are parallel to the design axis direction of the tunnel. Through the planar model unit normal vector with basis vectors The cross product is obtained, i.e. Thus ensuring They form an orthogonal basis in the plane.

[0051] like If so, then the distribution direction of the reinforcing bar is determined to be transverse. If so, then the distribution direction of the reinforcing bar is determined to be vertical. ,and If the condition is met, the steel bar will be marked as an inclined steel bar, and a construction deviation alarm will be output.

[0052] Compared with the prior art, the present invention has at least the following beneficial technical effects:

[0053] (1) This invention first adopts dynamic iteration number control, which enables the algorithm to adaptively adjust the computational load according to the noise level of the specific point cloud data. When the point cloud quality is good, it can converge quickly with fewer iterations; when the point cloud quality is poor, it automatically increases the number of iterations to ensure a high probability of finding the optimal solution, thereby optimizing the computational efficiency while ensuring fitting accuracy and avoiding the subjectivity and blindness of the iteration number setting in traditional methods. Then, by adding normal vector consistency constraints, this invention effectively overcomes the problem of "pseudo-interior points" easily generated in areas with complex local geometric features such as steel mesh intersections and metal supports in traditional methods. By introducing the consistency verification of surface geometric features, it can more strictly distinguish between the real tunnel sidewall planar points and the non-planar structural points attached to them, thereby significantly improving the purity and accuracy of the fitted plane. In addition, the key parameter settings in the identification method proposed in this invention adopt adaptive parameter settings rather than fixed values, which enhances the algorithm's adaptability to different projects and working conditions. Finally, the present invention uses the improved algorithm in step S3 above to fit the tunnel sidewall plane, which effectively solves the technical problems of traditional point cloud processing algorithms being easily affected by noise interference and having insufficient fitting accuracy when fitting the sidewall plane in the complex environment of the tunnel, thereby providing an accurate reference plane for subsequent steel reinforcement layer detection.

[0054] (2) This invention abandons the traditional single and isolated recognition algorithm and proposes a brand-new processing flow that progresses from macro to micro. This flow systematically decomposes the steel mesh recognition system into a sequence of steps: establishing a plane benchmark, separating steel layers, extracting individual steel bars, calculating geometric parameters, and assessing quality. This architecture ensures that the output of each step is the accurate input of the next step, realizing the progressive refinement of the processed object and the gradual improvement of positioning accuracy. It solves the technical problems of chaotic process and large error accumulation when existing methods deal with complex scenarios.

[0055] (3) The present invention adopts an adaptive clustering strategy based on physical meaning. In the key steps of steel bar layer detection and single steel bar separation, the main parameter of the clustering algorithm, clustering radius eps, no longer depends on empirical settings. This enables the identification method of the present invention to automatically adapt to various types of steel bars of different specifications, and solves the technical bottleneck of the general clustering algorithm performing poorly on sparse and uneven density point clouds.

[0056] (4) This invention adopts a dual-path collaborative rebar extraction mechanism with primary and backup methods, and innovatively designs a fault-tolerant mechanism for the primary method and backup method to work together. When the primary method fails due to excessively sparse point clouds or severe occlusion, the system automatically triggers the backup method. This mechanism effectively ensures that the algorithm can maintain good recognition ability under harsh data conditions, and improves the robustness and practicality of the entire method system. Attached Figure Description

[0057] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0058] Figure 1 The following is a diagram showing the identification results of the tunnel reinforcement mesh in the embodiments.

[0059] Figure 2 The following diagram shows the calculation results of the spacing and row spacing of the steel mesh on the tunnel sidewall in the embodiments. Detailed Implementation

[0060] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0061] For ease of description, the words "up," "down," "left," and "right" appearing in this invention only indicate that they are consistent with the up, down, left, and right directions of the accompanying drawings themselves. They do not limit the structure and are merely for the purpose of facilitating the description of this invention and simplifying the description. They do not indicate or imply that the device or component referred to needs to have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0062] The automatic steel mesh identification method based on UAV laser point cloud data proposed in this invention is further described below with reference to the accompanying drawings. The method specifically includes the following steps:

[0063] S1. Use a laser scanning device mounted on a drone to obtain the original data of the steel reinforcement mesh point cloud of a tunnel in the third phase of a subway project in Qingdao, for later use.

[0064] S2. First, the original point cloud data from step S1 is preprocessed using voxel grid downsampling technology to obtain a preprocessed point cloud with reduced density. The voxel grid downsampling technology utilizes adaptive voxel sizes. The following formula (1) is used to calculate:

[0065] (1);

[0066] In the above formula (1), the As the adjustment factor, in this embodiment, it is taken as... =2, This represents the original point cloud density.

[0067] Then, the normal vector of the preprocessed point cloud is calculated based on the k-nearest neighbor (K-NN) algorithm, and its unit normal vector is further obtained. Specifically, the steps include the following:

[0068] S2.1. The neighborhood of each point to be calculated in the preprocessed point cloud is determined using the K-nearest neighbor algorithm, resulting in a neighborhood point set. In this embodiment, the calculation takes... K =30, K The number of neighboring points considered when estimating the normal vector of a single point.

[0069] S2.2 Calculate the covariance matrix of the neighborhood point set obtained in S2.1 using the following formula (2). C :

[0070] (2);

[0071] In the above formula (2), K In this embodiment, to estimate the number of neighboring points considered when estimating the normal vector of a single point, K The value is 30. For the th neighborhood point set j The three-dimensional coordinates of each point Within the neighborhood point set The three-dimensional arithmetic mean center of each point is obtained by the following equation (3):

[0072] (3);

[0073] S2.3, Then, the covariance matrix obtained in S2.2 above is processed by the following equation (4). C Performing eigenvalue decomposition yields three eigenvalues, which are then sorted in descending order: At the same time, we obtain its three corresponding mutually orthogonal eigenvectors. .

[0074] (4);

[0075] S2.4, the smallest eigenvalue Corresponding feature vector That is, the normal vector direction of the preprocessed point cloud; then the... The unit normal vector can be obtained by normalization. .

[0076] Furthermore, after completing the above calculations, in order to ensure the quality of the point cloud data, it is also necessary to filter out outliers in the preprocessed point cloud. This specifically includes the following steps: First, select the centroid of the preprocessed point cloud. As the origin of the coordinate system, its calculation formula is as follows (5):

[0077] (5);

[0078] In the above formula (5), the NIt is the total number of points in the point cloud. It is a point i Three-dimensional coordinates in the original global coordinate system.

[0079] Then, to determine whether a point is a noise point, the following formula (6) is used to calculate the value of each point in the preprocessed point cloud. relative coordinates :

[0080] (6);

[0081] Then, a statistical filtering algorithm based on normal vector consistency is used to remove outliers, i.e., when the unit normal vector of a point is equal to the normal vector of the point, the outlier is removed. A point is considered a noise point if the angle between it and the average normal vector of its neighborhood is ≥30°. Outliers are removed. After completion, the transformation matrix from the coordinate system of the remaining point cloud to the local construction coordinate system is calculated to provide a unified coordinate reference for all subsequent spatial location-based calculations.

[0082] S3. First, use the following formula (7) to calculate the number of iterations required. Then, the RANSAC algorithm is used to iteratively fit the tunnel sidewall plane:

[0083] (7);

[0084] In the above formula (7), P For the expected confidence probability, this embodiment takes P =0.99. The aforementioned w For the estimated interior point rate, this embodiment takes... w =0.5. m To minimize the number of sample points required for model fitting, this implementation takes... m =3.

[0085] Then, the straight-line distance from each point in the preprocessed point cloud to the candidate plane is calculated. d 0, then obtain the planar model with the most interior points. Specifically, the steps include the following:

[0086] S3.1. Randomly select three non-collinear points from the preprocessed point cloud described in step S2, and then obtain a candidate plane based on these three points. The plane equation is expressed as:

[0087] .

[0088] S3.2 Calculate the straight-line distance from each point in the preprocessed point cloud to the candidate plane obtained in S3.1 using the following formula (8). d 0:

[0089] (8).

[0090] S3.3, Statistical results d 0 less than the threshold The points are used to form the "interior point set" for this iteration, and the total number of interior points contained therein is calculated. This is used as the optimal number of interior points for the current candidate plane. In this embodiment, the threshold value is used. =5cm.

[0091] S3.4. Compare the current candidate plane's optimal number of interior points with the optimal number of interior points obtained in the previous iteration. If the former is greater than the latter, replace the latter with the optimal number of interior points, and further use the plane model obtained in this iteration instead of the plane model from the previous iteration. (The iteration is described below.) After this process, the final optimal number of interior points is obtained, and the corresponding candidate planes are used as the plane model. .

[0092] S4. Calculate each non-tunnel sidewall plane point. To the planar model The distances are used to form a one-dimensional distance dataset. This dataset is then clustered to obtain clusters. After filtering, the remaining clusters are sorted by depth to obtain a clean set of point clouds representing the rebar layer. Specifically, the steps include:

[0093] S4.1, the aforementioned The three-dimensional coordinates are Its to the plane model straight-line distance d i The following formula (9) is used to calculate ( i =1, 2, 3... n Finally, the one-dimensional distance dataset is obtained. .

[0094] (9);

[0095] In the above formula (9), the a , b , c , d All are planar models The plane equation is shown in step S3.1 above.

[0096] S4.2. The DBSCAN algorithm is used to process the one-dimensional distance dataset. For direct clustering, the formula for calculating the cluster radius is shown in equation (10) below:

[0097] (10);

[0098] In the above formula (10), the k This is a proportionality coefficient; in this embodiment, it is set to a value of [value missing]. k =0.5, the S d It refers to the design spacing of the reinforcing bars.

[0099] S4.3. After the clustering process in S4.2 above, remove clusters with fewer than 30 point clouds. The remaining clusters are the effective reinforcement layer point cloud set. Sort them in ascending or descending order of depth to obtain the pure reinforcement layer point cloud set.

[0100] S5. Perform single-rebar separation processing on the point cloud set of the pure rebar layer to obtain point cloud clusters of individual rebars. Then, verify the validity of each point cloud cluster, and perform principal component analysis on the verified point cloud clusters to calculate the principal direction vector of the rebar axis and the linear equation of its axis in the three-dimensional coordinate system. Specifically, this includes the following steps:

[0101] S5.1. Use the DBSCAN algorithm to process each rebar layer point cloud in the pure rebar layer point cloud set. L k Two-dimensional clustering is performed, and the clusters are projected onto the planar model from the point cloud of the reinforcement layer. The process is performed on the subsequently obtained two-dimensional coordinates.

[0102] S5.2. After clustering, the validity of each generated point cloud cluster is verified: the number of points within the cluster must be no less than 3, and the length of the cluster point cloud calculated based on the principal component analysis (PCA) must be ≥0.01m. Clusters that cannot simultaneously meet both of the above conditions will be considered noise and filtered out.

[0103] S5.3, Verify the point cloud clusters of each valid single rebar. Principal component analysis (PCA) was used to calculate the principal direction vectors of the steel reinforcement axes, including: first, obtaining the point cloud clusters. The three-dimensional centroid coordinates are then used to construct the covariance matrix using the following equation (11). :

[0104] (11);

[0105] In the above formula (11), the , Same as formula (2) above; Represents the outer product of vectors. For the point cloud cluster The number of points in the middle.

[0106] S5.4, Then, the covariance matrix... Performing eigenvalue decomposition yields three eigenvalues, which are then sorted in descending order: At the same time, its corresponding feature vector is obtained. Among them, the largest eigenvalue Corresponding feature vector This is the principal direction vector of the reinforcing bar's axis. The reinforcing bar's axis can be determined by aligning the point cloud cluster along this principal direction vector. The direction projection is determined by taking the sequence of its midpoints.

[0107] Furthermore, in this step, a backup method is triggered when the number of reinforcing bars extracted by the principal component analysis (PCA) is less than 60% of the expected number. Specifically, this includes the following steps:

[0108] S5.5 First, the system counts the number of steel bars successfully extracted by the main method. And compared with the expected number of steel bars estimated based on the point cloud coverage area and design spacing. Compare. When the condition is met. <0.6 If the main method fails, the backup method is immediately activated. All point clouds of the current reinforcement layer are projected onto the aforementioned planar model. On the plane, 5cm×5cm, 3cm×3cm and 7cm×7cm grids are sequentially established for scanning and detection. Then, the validity of each grid is determined, and the determination formula is shown in the following formula (12).

[0109] (12);

[0110] In the above formula (12): the This represents the number of projected points contained within the current grid. The point cloud of the reinforcement layer in the planar model The average point cloud density per unit area is calculated as follows: . The area unit of the currently used grid is square meters (m²). 2 ).

[0111] S5.6. For each effective grid, Principal Component Analysis (PCA) is used to calculate its internal point cloud or, combined with its neighboring points, the first principal component direction, which is taken as the local principal direction of that grid. Then, the center points of all effective grids and their calculated principal directions are used to construct a set of candidate points with directions. Next, merging is performed through histogram peak analysis: the 360° direction range is divided into 10° zones, and the histogram distribution of the principal directions of all candidate points is statistically analyzed. The main peak of the histogram is identified as representing the dominant direction of the reinforcement in that region. Grids whose principal directions fall within ±15° of the same peak and are spatially adjacent are merged into the same "reinforcement candidate region". For each merged "reinforcement candidate region", all grid center points or original projection points within that region are mapped to the Hough parameter space ρ-θ. A peak angle search range of θ ±30° is set to improve fitting efficiency and accuracy. A voting process is performed in the parameter space to find the local maximum value with the highest number of votes, and the corresponding value is taken. ρ , θ Fit the equation of the straight line: The direction of this straight line is the direction of the reinforcing bar at that location on the projection plane. The two-dimensional straight line equation obtained by fitting the Hough transform, combined with its location... The three-dimensional equation is used to perform back projection calculations to reconstruct the axis of the rebar in three-dimensional space. Finally, the axis and direction information of all rebars identified by the backup method are output, maintaining the same format as the main method's results, for subsequent step S7 to calculate geometric parameters.

[0112] S6. Calculate the cosine of the angle between the principal direction vector and the basis vector of the reinforcing bar axis, and output the three-dimensional coordinates of the axis, the principal direction vector, and the distribution direction type of the reinforcing bar for each reinforcing bar in each layer. Specifically, this includes the following steps:

[0113] S6.1, The included angle cosine includes the principal direction vector of the steel bar axis. On the basis vectors respectively , Cosine value of the projected angle on It is calculated using the following formula (13):

[0114] (13);

[0115] In the above equation (13), the basis vector The basis vectors are parallel to the design axis direction of the tunnel. Through the planar model unit normal vector with basis vectors The cross product is obtained, i.e. Thus ensuring They form an orthogonal basis in the plane.

[0116] S6.2, if If so, then the distribution direction of the reinforcing bar is determined to be transverse. If so, then the distribution direction of the reinforcing bar is determined to be vertical. and If the condition is met, the steel bar will be marked as an inclined steel bar, and a construction deviation alarm will be output.

[0117] S6.3 Finally, output the identification results for each rebar layer, including the three-dimensional coordinates of the axis of each rebar, the principal direction vector, and the direction type, such as... Figure 1 As shown.

[0118] S7. Based on the calculation results of S6 above, calculate the spacing of rebars in the same layer and the spacing of rebar rows between layers. Then, output and display the original point cloud, rebar point cloud, and visualization results of rebar axis and layer relationship, and output the rebar spacing qualification rate, row spacing distribution histogram, and rebar position deviation, spacing deviation, and direction deviation. Specifically, this includes the following steps:

[0119] S7.1 Calculation of Spacing Between Reinforcing Bars in the Same Layer: Calculate the midpoint coordinates of the axis segment of each reinforcing bar within the same layer, and project all midpoint coordinates onto a vector perpendicular to the direction of the reinforcing bar. If it is a transverse reinforcing bar, the projection direction is a vertical vector; if it is a vertical reinforcing bar, the projection direction is a transverse vector. After obtaining the projected scalar value of each midpoint coordinate, sort all reinforcing bar midpoint coordinates in ascending or descending order according to the magnitude of this scalar value. For the resulting reinforcing bar sequence... The spacing between adjacent reinforcing bars is calculated sequentially using the following formula (14):

[0120] (14);

[0121] In the above formula (14), the... This represents the projected scalar value of the midpoint of the reinforcing bar. From this, a set of spacing values ​​can be obtained. To eliminate abnormal spacing caused by identification errors or local missing values, a filtering method based on statistical distribution is employed. First, the arithmetic mean of all spacing values ​​in the set is calculated. and standard deviation Remove items falling within the interval. Spacing values ​​other than those specified are considered outliers. The average of the remaining spacing values ​​after filtering is taken as the average spacing of the group of reinforcing bars.

[0122] S7.2 Calculation of interlayer reinforcement spacing: Calculate the minimum and average reinforcement spacing between different reinforcement layers. Plane fitting: For each reinforcement layer point cloud obtained in the pure reinforcement layer point cloud set obtained in step S4 above. An optimal plane is obtained by fitting the least squares method. Then calculate the best-fit plane between two adjacent reinforcement layers. and The normal distance between the two layers of reinforcing bars is the minimum spacing between them. ,like Figure 2 As shown.

[0123] S7.3 To verify the uniformity of row spacing, Hausdorff distance verification and row spacing uniformity evaluation are performed: compare the minimum row spacing. Verification value with Hausdorff The difference between the two is considered; when the difference exceeds a preset threshold, it is determined that there is uneven distribution of inter-layer reinforcement. Specifically: from the layer... to the floor The Hausdorff distance is defined as:

[0124] ;

[0125] Similarly, it can be calculated The final row spacing verification value. Take the maximum of the two:

[0126] ;

[0127] S7.4 Minimum spacing for multiple adjacent reinforcement layers Statistical analysis was performed to calculate the arithmetic mean of all effective inter-layer row spacings, which was then used as the average row spacing parameter.

[0128] S7.5 Finally, the output displays the visualization results of the original point cloud, rebar point cloud, rebar axis and hierarchical relationship. It shows the original point cloud, rebar point cloud, rebar axis and hierarchical relationship, outputs the spacing qualification rate and row spacing distribution histogram, and integrates with the BIM model to perform deviation analysis, outputs the position deviation, spacing deviation and direction deviation of the rebar, and obtains the quality assessment report of the rebar mesh.

[0129] It should be noted that any modifications, equivalent substitutions, or improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention. Although specific embodiments of this invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this invention are still within the scope of protection of this invention.

Claims

1. A method for automated identification of steel mesh based on UAV laser point cloud data, characterized in that, Includes the following steps: S1. Use the laser scanning equipment mounted on the drone to acquire the original point cloud data of the steel mesh for later use; S2. The original point cloud data is preprocessed using voxel grid downsampling technology to obtain a preprocessed point cloud with reduced density; then, the normal vector of the preprocessed point cloud is calculated to obtain the unit normal vector. Then, filter out outliers and calculate the transformation matrix from the coordinate system of the remaining point cloud to the local construction coordinate system. S3. Iteratively fit the tunnel sidewall plane from the preprocessed point cloud data, and calculate the distance from each point in the preprocessed point cloud to the candidate plane. d 0 Then, the planar model with the most interior points is obtained. ,spare; S4. Calculate the distance from each non-tunnel sidewall plane point to the plane model. The distances are used to form a one-dimensional distance dataset. Then, clustering is performed on the dataset to obtain clusters. After filtering, the remaining clusters are sorted by depth to obtain a pure steel layer point cloud set for later use. S5. Perform single-rebar separation processing on the point cloud set of the pure rebar layer to obtain point cloud clusters of single rebars; then perform validity verification on each point cloud cluster, perform principal component analysis on the point cloud clusters that pass the verification, and calculate the principal direction vector of the rebar axis and the linear equation of its axis in the three-dimensional coordinate system. S6. Calculate the cosine of the angle between the principal direction vector and the basis vector of the steel bar axis, and output the three-dimensional coordinates of the axis, the principal direction vector, and the distribution direction type of the steel bar in each layer of steel bars; S7. Based on the calculation results of S6 above, calculate the spacing of steel bars in the same layer and the spacing of steel bars between layers respectively; then output and display the original point cloud, steel bar point cloud and steel bar axis and layer relationship visualization results, and output the steel bar spacing qualification rate, spacing distribution histogram and steel bar position deviation, spacing deviation and direction deviation. In step S3, the planar model It is obtained using the following method: S3.

1. Randomly select three non-collinear points from the preprocessed point cloud, and then obtain a candidate plane based on these three points. The plane equation is expressed as: ; S3.2 Calculate the straight-line distance from each point in the preprocessed point cloud to the candidate plane using the following formula (6). d 0: ； In the above formula (6), the a , b , c , d All are planar models The plane equation parameters are as shown in step S3.1 above; S3.3, The straight-line distance obtained statistically d 0 less than the threshold The points are used to form the "interior set" for this iteration, and the total number of interior points contained in it is calculated. The optimal number of interior points in the current candidate plane; S3.

4. Compare the current optimal number of interior points in the candidate plane with the optimal number of interior points obtained in the previous iteration. If the former is greater than the latter, replace the latter with the optimal number of interior points, and further use the plane model obtained in this iteration to replace the plane model of the previous iteration; the iteration... After this process, the final optimal number of interior points is obtained, and the corresponding candidate planes are used as the plane model. ; Step S4 specifically includes the following steps: S4.1, the aforementioned The three-dimensional coordinates are Its to the plane model straight-line distance d i The following formula (9) is used to calculate ( i =1, 2, 3... n Finally, the one-dimensional distance dataset is obtained. ; ； In the above formula (9), the a , b , c , d All are planar models The plane equation is shown in step S3.1 above; S4.

2. The DBSCAN algorithm is used to process the one-dimensional distance dataset. For direct clustering, the formula for calculating the cluster radius is shown in equation (10) below: （10）； In the above formula (10), the k This is a proportionality coefficient; in this embodiment, it is set to a value of [value missing]. k =0.5, the S d It refers to the design spacing of the reinforcing bars; S4.

3. After removing the clusters obtained from the clustering process in S4.2 above, those clusters with fewer than 30 point clouds are removed. The remaining clusters are the effective set of point clouds for the steel reinforcement layer. Sort them in ascending or descending order according to their depth to obtain the pure set of point clouds for the steel reinforcement layer. In step S5, the basis for validity verification is: the number of points within a cluster must be no less than 3, and the length of the cluster point cloud calculated based on the principal component analysis must be ≥0.01m; if the above two conditions cannot be met, the cluster will be considered as noise and filtered out. In step S5, when the number of reinforcing bars extracted by the principal component analysis is less than 60% of the expected number, a backup method is triggered, specifically including the following steps: S5.5 First, the system counts the number of steel bars successfully extracted by the main method. And compared with the expected number of steel bars estimated based on the point cloud coverage area and design spacing. Compare; when conditions are met < 0.6 If the main method fails, the backup method is immediately activated; all point clouds of the current reinforcement layer are projected onto the aforementioned planar model. On the plane, 5cm×5cm, 3cm×3cm and 7cm×7cm grids are sequentially established for scanning and detection. Then, the validity of each grid is determined, and the determination formula is as shown in the following formula (12). ； In the above formula (12): the This represents the number of projected points contained within the current grid. The point cloud of the reinforcement layer in the planar model The average point cloud density per unit area is calculated as follows: The area unit of the currently used grid is m. 2 ; S5.

6. For each effective grid, use principal component analysis to calculate its internal point cloud or combine the first principal component direction of its neighboring points, and use it as the local principal direction of the grid. Subsequently, the center points of all effective grids and their calculated principal directions are constructed into a set of oriented candidate points. Then, merging is performed using histogram peak analysis: the 360° directional range is divided into 10° zones, and the histogram distribution of the principal directions of all candidate points is statistically analyzed; the main peaks of the histograms are identified as representing the dominant direction of the reinforcement in that region; grids whose principal directions fall within ±15° of the same peak and are spatially adjacent are merged into a single "reinforcement candidate region"; for each merged "reinforcement candidate region," all grid center points or original projection points within that region are mapped to the Hough parameter space ρ-θ; a peak angle search range of θ ±30° is set to improve fitting efficiency and accuracy; voting is conducted in the parameter space to find the local maximum value with the highest number of votes, and the corresponding value is taken. ρ , θ Fit the equation of the straight line: ; The direction of this straight line is the direction of the reinforcing bar at that location on the projection plane; the two-dimensional straight line equation obtained by fitting the Hough transform, combined with its location... The three-dimensional equation is used to perform back projection calculations to reconstruct the axis of the steel bar in three-dimensional space. Finally, the axis and direction information of all steel bars identified by the backup method are output, and the result format is consistent with that of the main method, so that S7 can perform geometric parameter calculations.

2. The automatic steel mesh identification method based on UAV laser point cloud data according to claim 1, characterized in that, In step S2, the adaptive voxel size of the voxel grid downsampling technique The following formula (1) is used to calculate: （1）； In the above formula (1), This is an adjustment factor, with a value ranging from 1.5 to 3.

0. This represents the original point cloud density.

3. The automatic steel mesh identification method based on UAV laser point cloud data according to claim 1, characterized in that, In step S2, the method for calculating the normal vector of the preprocessed point cloud includes the following steps: S2.

1. The neighborhood of each point to be calculated in the preprocessed point cloud is determined by the K-nearest neighbor algorithm to obtain the neighborhood point set; S2.2 Calculate the covariance matrix of the neighborhood point set using the following formula (2). C : ； In the above formula (2), K =30~50, which represents the number of neighboring points considered when estimating the normal vector of a single point. For the th neighborhood point set j The three-dimensional coordinates of each point The three-dimensional arithmetic mean center of each point in the neighborhood point set is obtained by the following equation (3): ； S2.3, Then, the covariance matrix is ​​obtained by the following equation (4). C Eigenvalue decomposition yields three eigenvalues. and the corresponding three mutually orthogonal eigenvectors ; ； S2.4, the smallest eigenvalue Corresponding feature vector That is, the normal vector direction of the preprocessed point cloud; then the... The unit normal vector can be obtained by normalization. .

4. The method for automated identification of steel mesh based on UAV laser point cloud data according to claim 1, characterized in that, In step S2, the outlier is determined using the following method: if the unit normal vector If the angle between the point and the average normal vector of its neighborhood is ≥30°, then the point is considered a valid point. This is an outlier.

5. The method for automated identification of steel mesh based on UAV laser point cloud data according to claim 1, characterized in that, In step S3, the number of iterations required is first calculated using the following formula (5). Then, the RANSAC algorithm is used to iteratively fit the tunnel sidewall plane; （5）； In the above formula (5), P The expected confidence probability is usually taken as P =0.95~1, w The estimated interior point ratio is usually taken as... w= 0.3~0.6, m This represents the minimum number of sample points required to fit the model.

6. The method for automated identification of steel mesh based on UAV laser point cloud data according to claim 1, characterized in that, In step S5, the DBSCAN algorithm is used to process each rebar layer point cloud in the pure rebar layer point cloud set. L k Two-dimensional clustering is performed, and the clusters are projected onto the planar model from the point cloud of the reinforcement layer. The process is performed on the subsequently obtained two-dimensional coordinates.

7. The method for automated identification of steel mesh based on UAV laser point cloud data according to any one of claims 1-6, characterized in that, In step S6, the included angle cosine includes the principal direction vector of the reinforcing bar axis. On the basis vectors respectively , Cosine value of the projected angle on It is calculated using the following formula (9): ； In the above equation (9), the basis vector The basis vectors are parallel to the design axis direction of the tunnel. Through the planar model unit normal vector with basis vectors The cross product is obtained, i.e. Thus ensuring They form an orthogonal basis in the plane.

8. The method for automated identification of steel mesh based on UAV laser point cloud data according to claim 7, characterized in that, like If the distribution direction of the reinforcing bar is determined to be transverse; if If the distribution direction of the reinforcing bar is vertical, then the distribution direction of the reinforcing bar is determined to be vertical; if If the condition is met, the steel bar will be marked as an inclined steel bar, and a construction deviation alarm will be output.

Images