CN-121658919-B - Automatic reinforcing mesh identification method based on unmanned aerial vehicle laser point cloud data

CN121658919BCN 121658919 BCN121658919 BCN 121658919BCN-121658919-B

Abstract

The invention relates to the technical field of electric digital data processing, and particularly discloses an automatic identification method of a reinforcing mesh based on unmanned aerial vehicle laser point cloud data, which comprises the following steps: the method comprises the steps of obtaining reinforcing steel bar mesh point cloud data in the construction period of a tunnel or an underground structure, preprocessing original point cloud, dividing a plane, detecting a reinforcing steel bar layer, extracting a reinforcing steel bar axis, identifying a reinforcing steel bar direction, calculating interlayer row spacing geometric parameters, outputting a result and the like. According to the invention, full-automatic identification is realized by combining an unmanned plane laser scanning technology with an intelligent point cloud processing algorithm, the accuracy of a detection result is improved, the problems of low efficiency, insufficient accuracy and the like existing in the traditional method due to the dependence on manual measurement are overcome, the high-precision and high-efficiency identification of the reinforcing mesh in the construction period of a tunnel or an underground structure is facilitated, and a more reliable quality evaluation basis is provided.

Inventors

CHEN XIN
LI ZHIYUAN
LI PENG
ZHANG QINGSONG
DENG JIA
Cheng Tianshuo
LI JUNCHANG
Chang Luyi
YANG RUI

Assignees

山东大学
青岛地铁集团有限公司西海岸建设分公司

Dates

Publication Date: 20260508
Application Date: 20260206

Claims (8)

1. The automatic identification method for the reinforcing mesh based on the unmanned aerial vehicle laser point cloud data is characterized by comprising the following steps of: s1, acquiring original point cloud data of a reinforcing mesh by using laser scanning equipment carried by an unmanned aerial vehicle for standby; S2, preprocessing the original point cloud data by adopting a voxel grid downsampling technology to obtain preprocessed point cloud with reduced density, and then calculating the normal vector of the preprocessed point cloud to obtain a unit normal vector Filtering to remove outliers, and calculating a transformation matrix from the coordinate system of the residual point cloud to the local construction coordinate system after the outliers are completed; S3, iteratively fitting tunnel side wall planes from the preprocessed point cloud data, calculating the distance d 0 between each point in the preprocessed point cloud and a candidate plane, and obtaining a plane model with the largest number of internal points Standby; S4, calculating the plane point of each non-tunnel side wall to the plane model Forming a one-dimensional distance data set, then carrying out clustering treatment on the data set to obtain clusters, filtering, and then carrying out depth sequencing on the rest clusters to obtain a pure reinforcement layer point cloud set for later use; s5, carrying out single steel bar separation treatment on the pure steel bar layer point cloud set to obtain point cloud clusters of single steel bars, carrying out validity verification on each point cloud cluster, carrying out principal component analysis on the point cloud clusters passing verification, and calculating a principal direction vector of a steel bar axis and a linear equation of the axis in a three-dimensional coordinate system; s6, calculating an included angle cosine between the principal direction vector and the base vector of the axis of the steel bar, and outputting the three-dimensional axis coordinate, the principal direction vector and the distribution direction type of the steel bars of each layer of steel bars; S7, respectively calculating the spacing of the reinforcing steel bars on the same layer and the spacing of the reinforcing steel bars between layers based on the calculation result of the S6, then outputting a visual result showing the original point cloud, the reinforcing steel bar axis and the hierarchical relation, and outputting the spacing qualification rate, the spacing distribution histogram, the position deviation, the spacing deviation and the direction deviation of the reinforcing steel bars; in step S3, the planar model The method comprises the following steps: s3.1, randomly selecting three non-collinear points from the preprocessing point cloud, and obtaining a candidate plane according to the three points, wherein a plane equation is expressed as follows: ; s3.2, calculating the linear distance d 0 between each point in the preprocessing point cloud and the candidate plane by adopting the following (6): ; in the above formula (6), a, b, c, d are plane models As shown in step S3.1 above; S3.3, the linear distance d 0 obtained by statistics is smaller than a threshold value The points form an ' inner point set ' of the iteration, and the total inner point number is contained in the ' inner point set The number of the optimal interior points serving as the current candidate plane; S3.4, comparing the number of the optimal interior points of the current candidate plane with the number of the optimal interior points obtained by the last iteration, if the number of the optimal interior points is more than the number of the optimal interior points, replacing the optimal interior points by the former, further adopting a plane model obtained by the current iteration to replace the plane model of the last iteration, and iterating the steps After that, the final optimal number of interior points is obtained, and the corresponding candidate plane is used as the plane model ; The step S4 specifically comprises the following steps: s4.1, the Is the three-dimensional coordinates of (a) To the plane model The linear distance d i of (1) is calculated by the following formula (9) (i=1) 2 3.....n.), finally obtaining the one-dimensional distance data set ; ; In the above formula (9), a, b, c, d are plane models As shown in step S3.1 above; S4.2, adopting DBSCAN algorithm to make the one-dimensional distance data set The direct clustering is carried out, and the calculation formula of the clustering radius is shown as the following formula (10): (10); In the above formula (10), k is a proportionality coefficient, the value of this embodiment is k=0.5, and S d is the design distance of the reinforcing steel bars; s4.3, eliminating class clusters with the number of point clouds of less than a threshold value of 30 in the class clusters obtained after the clustering treatment of the S4.2, wherein the rest class clusters are the effective reinforcement layer point cloud sets; In the step S5, the basis for the validity verification is that the number of points in the cluster is not less than 3, and the length of the cluster point cloud calculated based on the principal component analysis is not less than 0.01m; in S5, the standby method is triggered when the number of the rebars extracted by the principal component analysis is less than 60% of the expected number, specifically, the method comprises the following steps: s5.5, firstly, carrying out systematic statistics on the number of the steel bars successfully extracted by the main method And the number of the expected steel bars estimated based on the coverage area of the point cloud and the design distance Comparing when the condition is satisfied < 0.6 When the main method is judged to be invalid, the standby method is started immediately, and all point clouds of the current steel bar layer are projected to the plane model Sequentially establishing 5cm×5cm, 3cm×3cm and 7cm×7cm grids on the plane for scanning detection, and then judging the effectiveness of each grid, wherein the judging formula is shown in the following formula (12); ; In the above formula (12) The number of projection points contained in the current grid is the number; The plane model is formed by the point cloud of the reinforcing steel bar layer The average unit area point cloud density is calculated by The unit of the area of the current adopted grid is m 2 ; s5.6, calculating the direction of a first principal component of an internal point cloud or a neighborhood point of each effective grid by utilizing principal component analysis to serve as a local principal direction of the grid, then constructing a set of candidate point sets with directions by using the central points of all the effective grids and the calculated principal directions, combining by using direction histogram peak analysis, namely dividing a 360-degree direction range into a region of 10 degrees, counting the histogram distribution of the principal directions of all the candidate points, identifying the principal peak value of the histogram to represent the dominant direction of a reinforcing steel bar in the region, combining grids with the principal directions falling within the same peak value +/-15 degrees and adjacent to each other in space into the same "reinforcing steel bar candidate region", mapping all grid central points or original projection points in the region into a Hough parameter space rho-theta space, setting peak angle theta +/-30 degrees, voting in the parameter space, searching for the local maximum value with the highest ticket number, and taking the corresponding rho and theta, fitting a straight line equation: the direction of the straight line is the direction of the reinforcing steel bar on the projection plane, and the two-dimensional straight line equation obtained by fitting the Hough transformation is combined with the direction of the reinforcing steel bar And finally, outputting all the axes of the reinforcing steel bars identified by the standby method and the direction information thereof, keeping the same with the result format of the main method, and carrying out geometric parameter calculation by the S7.
2. The automated mesh reinforcement recognition method based on unmanned aerial vehicle laser point cloud data of claim 1, wherein in step S2, the voxel grid downsampling technique is adaptive to voxel size Calculated using the following formula (1): (1); in the above-mentioned formula (1), For the adjustment factor, the value is 1.5-3.0, Is the original point cloud density.
3. The automatic recognition method of the reinforcing mesh based on the unmanned aerial vehicle laser point cloud data according to claim 1, wherein in the step S2, the method for calculating the normal vector of the preprocessing point cloud comprises the following steps: S2.1, determining the neighborhood of each point to be calculated in the preprocessing point cloud through a K neighbor algorithm to obtain a neighborhood point set; S2.2, calculating a covariance matrix C of the neighborhood point set through the following (2): ; In the above formula (2), k=30 to 50, which is the number of neighboring points considered in estimating the normal vector of a single point, As the three-dimensional coordinate point of the j-th point in the neighborhood point set, The three-dimensional arithmetic average center of each point in the neighborhood point set is obtained by the following formula (3): ; s2.3, then carrying out characteristic decomposition on the covariance matrix C through the following (4) to obtain three characteristic values And corresponding three mutually orthogonal eigenvectors ; ; S2.4, minimum said eigenvalue Corresponding feature vector Namely the normal vector direction of the preprocessing point cloud, and then the preprocessing point cloud is processed Normalizing to obtain the unit normal vector 。
4. The method for automatically identifying the reinforcing mesh based on the unmanned aerial vehicle laser point cloud data according to claim 1, wherein in the step S2, the outlier is determined by the method comprising the following steps if the unit normal vector is adopted The included angle between the neighbor average normal vector is more than or equal to 30 degrees, and the point is judged Is an outlier.
5. The automated mesh reinforcement recognition method based on unmanned aerial vehicle laser point cloud data according to claim 1, wherein in step S3, the number of times required for the iteration is calculated by the following equation (5) Then iteratively fitting the tunnel side wall plane by adopting a RANSAC algorithm; (5); in the above equation (5), P is a desired confidence probability, p=0.95 to 1 is usually taken as an estimated interior point rate, w=0.3 to 0.6 is usually taken as a minimum number of sample points required for fitting the model.
6. The unmanned aerial vehicle laser point cloud data-based reinforcement mesh automatic identification method according to claim 1, wherein in step S5, a DBSCAN algorithm is adopted to perform two-dimensional clustering on each reinforcement layer point cloud L k in the pure reinforcement layer point cloud set, and the clustering projects the reinforcement layer point cloud to the plane model And performing the two-dimensional coordinate obtained later.
7. The automated mesh reinforcement recognition method based on unmanned aerial vehicle laser point cloud data of any of claims 1-6, wherein in step S6, the angle cosine comprises a principal direction vector of a bar axis Respectively at the basis vectors 、 Cosine value of projection included angle on upper surface It is calculated by the following formula (9): ; in the above formula (9), the basis vector In order to be parallel to the design axis direction of the tunnel, the base vector Through the plane model Unit normal vector of (2) And base vector Is obtained by cross-product of (i.e.) Thereby ensuring Constituting an orthogonal basis in a plane.
8. The unmanned aerial vehicle laser point cloud data-based mesh reinforcement automated identification method of claim 7, wherein if Judging the distribution direction type of the reinforcing steel bar is transverse, if Judging that the distribution direction type of the reinforcing steel bar is vertical, if so And marking the steel bar as an oblique steel bar, and outputting a construction deviation alarm.

Description

Automatic reinforcing mesh identification method based on unmanned aerial vehicle laser point cloud data Technical Field The invention relates to the technical field of electric digital data processing, in particular to an automatic identification method for a reinforcing mesh based on unmanned aerial vehicle laser point cloud data. Background The disclosure of this background section is only intended to increase the understanding of the general background of the invention and is not necessarily to be construed as an admission or any form of suggestion that this information forms the prior art already known to those of ordinary skill in the art. In the construction of lining structures of underground projects such as tunnels, subway stations and the like, the installation quality of a reinforcing mesh directly and deeply influences the overall safety and long-term durability of the structure. The reinforcing mesh is used as a core framework in a concrete structure, and the accuracy of positioning, the uniformity of spacing and the firmness of lapping and binding all determine whether the lining can effectively bear soil pressure and water pressure from surrounding rocks and long-term dynamic load brought by train operation. If the installation is deviated, such as overlarge spacing, infirm binding or insufficient thickness of the protective layer, concrete cracking and bearing capacity reduction are easily caused, even local spalling or structural damage is caused, and engineering operation safety and service life are seriously threatened. Therefore, the reinforcing mesh must be installed as a key procedure to be strictly controlled, thus laying a solid foundation for the safety and stability of the whole structure. The traditional steel bar mesh installation quality detection mainly relies on manual sampling measurement, but has the defects of low detection efficiency, long time consumption and easy accident initiation in high-altitude operation due to the fact that scaffolds are required to be erected for point-by-point measurement. Secondly, coverage rate is insufficient, and the quality is difficult to comprehensively evaluate because the sampling rate is often lower than 10%. At present, some detection schemes adopt an automatic technology, including (1) an image-based method, wherein the method is easy to be interfered by illumination and shielding, and the error rate of steel bar identification exceeds 30%. (2) The method based on the point cloud has higher precision, but faces the technical bottleneck that characteristic confusion is caused by uneven point cloud density in the steel bar intersection area. (3) The traditional clustering algorithm has the advantages that the adaptability of the method to the sparse point cloud is poor, the omission ratio exceeds 25%, and the direction identification depends on manual preset parameters and cannot be used for adaptively distinguishing the transverse and vertical distribution conditions of the reinforcing steel bars. Disclosure of Invention The invention provides an unmanned aerial vehicle laser point cloud data-based automatic identification method for a reinforcing mesh, which realizes full-automatic identification by combining an unmanned aerial vehicle laser scanning technology with an intelligent point cloud processing algorithm, improves the accuracy of a detection result, overcomes the problems of low efficiency, insufficient accuracy and the like caused by the fact that the traditional method relies on manual measurement, facilitates high-precision and high-efficiency identification of the reinforcing mesh in the construction period of a tunnel or an underground structure, and provides a more reliable quality evaluation basis. Specifically, the technical scheme of the invention is as follows. An unmanned aerial vehicle laser point cloud data-based automatic identification method for a reinforcing mesh comprises the following steps: s1, acquiring original data of the reinforcing steel bar mesh point cloud by using laser scanning equipment carried by an unmanned aerial vehicle for standby. S2, preprocessing the original point cloud data by adopting a voxel grid downsampling technology to obtain preprocessed point cloud with reduced density. Then calculating the normal vector of the preprocessing point cloud to obtain a unit normal vector. And filtering to remove outliers, and calculating a transformation matrix from the coordinate system of the residual point cloud to the local construction coordinate system after the completion. S3, iteratively fitting tunnel side wall planes from the preprocessed point cloud data, and calculating the distance from each point in the preprocessed point cloud to a candidate planeThen obtaining the plane model with the largest number of inner pointsAnd (5) standby. S4, calculating the plane point of each non-tunnel side wall to the plane modelAnd (3) forming a one-dimensional distance data set, then carrying out clus