CN115619977A - High-order dangerous rock monitoring method based on airborne laser radar - Google Patents
High-order dangerous rock monitoring method based on airborne laser radar Download PDFInfo
- Publication number
- CN115619977A CN115619977A CN202211410024.5A CN202211410024A CN115619977A CN 115619977 A CN115619977 A CN 115619977A CN 202211410024 A CN202211410024 A CN 202211410024A CN 115619977 A CN115619977 A CN 115619977A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- point
- dangerous rock
- plane
- obtaining
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/17—Terrestrial scenes taken from planes or by drones
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/10—Image acquisition
- G06V10/16—Image acquisition using multiple overlapping images; Image stitching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/80—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level
- G06V10/803—Fusion, i.e. combining data from various sources at the sensor level, preprocessing level, feature extraction level or classification level of input or preprocessed data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Remote Sensing (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of high-level dangerous rock monitoring, in particular to a high-level dangerous rock monitoring method based on an airborne laser radar, which comprises the following steps: obtaining sparse point cloud data according to a scanning result of the laser radar on the high-level dangerous rock area; obtaining dense matching point cloud according to the image shot for the high dangerous rock area; carrying out denseness processing on the sparse point cloud data by using the dense matching point cloud to obtain a fused point cloud; acquiring the structural surface displacement of the dangerous rock according to the fused point cloud; and (4) carrying out high-position dangerous rock monitoring according to the structural plane position of the dangerous rock. Compared with the prior art that the high-position dangerous rock monitoring is directly carried out by using the sparse point cloud data, the data of the fused point cloud is more complete and dense, so that more dangerous rock characteristics can be represented, more accurate displacement of the structural surface of the dangerous rock can be obtained according to the fused point cloud, and the accuracy of the high-position dangerous rock monitoring is improved.
Description
Technical Field
The invention relates to the technical field of high-level dangerous rock monitoring, in particular to a high-level dangerous rock monitoring method based on an airborne laser radar.
Background
After some geological activities, the mountain body shatters and the rock mass loosens to form a large amount of loose substances and loose rock mass, which aggravates the induction degree of geological disasters and brings great threat to the safe operation of roads. The disaster types mainly include landslide, collapse and debris flow, wherein the landslide and the debris flow have good monitoring means at present, but the collapse does not form an effective monitoring method, and the collapse dangerous rock mass accounts for a considerable proportion, about 27.36 percent and is mainly high-level dangerous rock mass. The dangerous rock masses are often positioned on the tops of mountains or high slopes outside highway areas on two sides of the roadbed, are high in concealment, difficult to reach by personnel and high in danger degree. Dangerous rocks have extremely strong uncontrollable property, and cause a great amount of geological disasters or secondary disasters of highways and railways every year, so the method has very important social and scientific significance for early identification, monitoring and early warning research of high-level high-energy rock collapse in order to guarantee the property safety of people and reduce the national economic property loss.
In recent years, with the rapid development of surveying and mapping technology, new surveying and mapping technology is emerging continuously, and many scholars begin to measure dangerous rock mass structural planes by photogrammetry. Technologies such as satellite-borne synthetic aperture radar interferometry (InSAR), unmanned aerial vehicle photography (UAV), ground-based synthetic aperture radar (D-Insar), three-dimensional laser scanning, unmanned aerial vehicle three-dimensional laser radar (LiDAR) and the like are successively used for geological structure characteristic analysis, continuous high-precision deformation monitoring, early identification and early warning of dangerous rock masses. Because each laser point of the point cloud obtained by scanning of the laser radar is a real point on the surface of the measured object, the point cloud has the characteristics of accurate position, reliable data and no influence of ambient light, but has the defects of sparsity, incompleteness and the like, so that the accuracy of monitoring and early warning by directly analyzing the geological structure characteristics of dangerous rock masses according to the point cloud obtained by scanning of the laser radar is lower.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a high-level dangerous rock monitoring method based on an airborne laser radar, and the accuracy of high-level dangerous rock monitoring is improved.
The invention adopts the technical scheme that the high-level dangerous rock monitoring method is based on an airborne laser radar.
In a first implementation manner, the high-altitude dangerous rock monitoring method based on the airborne laser radar comprises the following steps: obtaining sparse point cloud data according to a scanning result of the laser radar on the high-level dangerous rock area; obtaining dense matching point cloud according to the image shot for the high dangerous rock area; performing densification processing on the sparse point cloud data by using the dense matching point cloud to obtain a fusion point cloud; acquiring the structural plane displacement of the dangerous rock according to the fusion point cloud; and (4) carrying out high-position dangerous rock monitoring according to the structural plane position of the dangerous rock.
With reference to the first implementable manner, in a second implementable manner, performing densification processing on sparse point cloud data to obtain a fused point cloud, including: preprocessing the sparse point cloud data and the dense matching point cloud data to obtain two types of heterogeneous point clouds; performing point cloud segmentation processing on each heterogeneous point cloud to obtain two classification labels; and obtaining a fused point cloud according to the binary classification label and the point cloud fusion algorithm.
With reference to the second implementable manner, in a third implementable manner, point cloud segmentation processing is performed according to each heterogeneous point cloud to obtain two classification labels, including:
by calculation of
Obtaining a binary classification label;
wherein, L = { L = i Is a binary labelset for the sparse point cloud, and l i E {0,1}, the label is 0 to indicate that the sparse point cloud has substitution dense matching point cloud, and lambda is a regularization factor, pi [ ·]Is a discriminant function, Π [ true]＝1，Π[false]=0, sparse point cloud P = { P = i }。
With reference to the second implementation manner, in a fourth implementation manner, obtaining a fused point cloud according to a binary tag and a point cloud fusion algorithm includes: splicing the two types of heterogeneous point clouds to form spliced point clouds; removing overlapped redundant point clouds of the spliced point clouds by using the binary classification labels to obtain initial fusion point clouds; and smoothing the initial fusion point cloud according to a point cloud fusion algorithm to obtain a fusion point cloud.
In combination with the first implementable manner, in a fifth implementable manner, obtaining the structural plane displacement of the dangerous rock according to the fused point cloud includes: extracting key points in the fused point cloud; acquiring local point cloud characteristics according to the key points; matching the key points according to the local point cloud characteristics to obtain a dangerous rock point cloud set; acquiring a dangerous rock structural surface according to the dangerous rock point cloud set; and acquiring the structural plane displacement of the dangerous rock according to the structural plane of the dangerous rock.
With reference to the fifth implementable manner, in a sixth implementable manner, the extracting key points in the fused point cloud includes: preprocessing the fused point cloud to obtain initial point cloud data; processing the initial point cloud data by using a point cloud partitioning strategy of a self-adaptive expansion receiving domain to obtain point cloud enhanced data; and inputting the point cloud enhanced data into a key point extraction network to obtain key points in the fused point cloud.
With reference to the sixth implementable manner, in a seventh implementable manner, the method for preprocessing the fused point cloud to obtain initial point cloud data includes: performing ground segmentation processing on the fused point cloud; selecting a sampling point of the fused point cloud; constructing a local neighborhood of each sampling point; obtaining a least square plane of a local neighborhood; and acquiring a normal vector of the least square plane, and taking the normal vector of the least square plane as a normal vector of a sampling point.
With reference to the fifth implementable manner, in an eighth implementable manner, the obtaining local point cloud features according to the key points includes: extracting a plurality of local point sets from the key points by utilizing a point cloud division strategy of a self-adaptive expansion receiving domain; carrying out centralized operation on each local point set, and calculating the distance from each point in the local point set to a key point; adding the normal vector of the key points and the distance between each point and each key point to the spatial information of each key point; and inputting the spatial information into a local feature extraction network to obtain local point cloud features.
With reference to the fifth implementable manner, in a ninth implementable manner, the key points are paired according to the local point cloud features to obtain a dangerous rock point cloud set, including: obtaining a matching relation between the key points and the source point cloud by utilizing a nearest neighbor search algorithm according to the key points and the local point cloud characteristics; matching the key points with the source point cloud according to the matching relation to obtain a plurality of matching point clouds; and obtaining a dangerous rock point cloud set according to the matched point cloud.
In combination with the fifth implementable manner, in a tenth implementable manner, the key points are paired according to the local point cloud characteristics to obtain a dangerous rock point cloud set, including: screening local point cloud characteristics and key points by adopting a parameter estimation algorithm to obtain point cloud data conforming to parameter transformation; inputting point cloud data into an initial transformation matrix to obtain point cloud transformation parameters; performing initial matching on point cloud transformation parameters by adopting an ICP (inductively coupled plasma) algorithm to obtain initial matching parameters; inputting the initial matching parameters into a fine registration matrix for matching again to obtain a matching point cloud; and extracting and dividing the target according to the matching point cloud to obtain a dangerous rock point cloud set.
With reference to the fifth implementable manner, in an eleventh implementable manner, obtaining a dangerous rock structural plane according to the dangerous rock point cloud set includes:
and performing aggregation classification on the dangerous rock point cloud sets by using a clustering algorithm to obtain a dangerous rock structural surface.
With reference to the fifth implementable manner, in a twelfth implementable manner, obtaining the structural plane displacement of the dangerous rock according to the structural plane of the dangerous rock includes: grouping a plurality of structural surfaces of the dangerous rock to form a plurality of structural surface groups; forming a structural block of the rock mass according to the plurality of structural surface groups; obtaining the displacement condition of each plane in the structure block in the respective normal vector direction; and acquiring the structural plane displacement of the dangerous rock according to each displacement condition.
With reference to the twelfth implementation manner, in a thirteenth implementation manner, the obtaining a displacement condition of each plane in the structural block in the corresponding normal vector direction includes: taking each plane of the structure block at the previous moment as a reference plane; and obtaining each distance change value of each current plane of the structural block and the reference plane in the normal vector direction, wherein each distance change value is the displacement of each plane in the structural block in the corresponding self-normal vector direction.
With reference to the thirteenth implementable manner, in a fourteenth implementable manner, obtaining structural plane displacement of the dangerous rock according to each displacement condition includes: and comparing the displacement of each plane in the structural block in the direction corresponding to each normal vector to obtain the displacement of the structural surface of the dangerous rock.
According to the technical scheme, the beneficial technical effects of the invention are as follows:
1. the dense matching point cloud data is used for processing the sparse point cloud data, dense fusion point cloud is finally obtained, advantages of the sparse point cloud data and the dense matching point cloud data are fully exerted, the problems of single data missing, data sparseness and point cloud cavities are solved, advantages of the sparse point cloud data and the dense matching point cloud are complemented, accordingly, more dangerous rock characteristics can be represented, more accurate displacement structural surface information of dangerous rocks can be obtained according to the fusion point cloud, and accuracy of high-order dangerous rock monitoring is improved.
2. The traditional laser point cloud processing method does not consider monitoring in combination with the characteristics of the dangerous rock mass, the scheme is combined with the characteristics of the structural surface of the dangerous rock mass to extract the structural surface, the integral displacement condition of the point cloud surface is adopted to replace the displacement of a plurality of discrete points, and the precision and the stability are improved.
3. And performing aggregation classification on the structural surface point clouds by adopting a clustering algorithm to extract the point clouds of the same group of structural surfaces from the whole rock mass point cloud, so that the loss of details on the rock mass surface and the cleaning of vegetation on the rock mass surface are avoided, and the accuracy of high-level dangerous rock identification is improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below. Throughout the drawings, like elements or portions are generally identified by like reference numerals. In the drawings, elements or portions are not necessarily drawn to scale.
Fig. 1 is a schematic diagram of a high-level dangerous rock monitoring method based on an airborne laser radar provided by the invention;
FIG. 2 is a schematic diagram of a point cloud partitioning strategy for adaptively expanding a receiving domain according to the present invention;
FIG. 3 is a schematic diagram of a method for extracting cloud sets of dangerous rock points according to the present invention;
FIG. 4 is a schematic diagram of a point cloud coloring according to the present invention;
fig. 5 is a flow chart of the high-level dangerous rock displacement measurement provided by the invention.
Detailed Description
Embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and therefore are only examples, and the protection scope of the present invention is not limited thereby.
It is to be noted that, unless otherwise specified, technical or scientific terms used herein shall have the ordinary meaning as understood by those skilled in the art to which the invention pertains.
Referring to fig. 1, the embodiment provides a high-level dangerous rock monitoring method based on an airborne laser radar, including:
s01, obtaining sparse point cloud data according to a scanning result of a laser radar on a high-position dangerous rock area;
s02, obtaining dense matching point cloud according to an image shot in a high dangerous rock area;
s03, performing densification processing on the sparse point cloud data by using dense matching point cloud to obtain fusion point cloud;
s04, obtaining the structural surface displacement of the dangerous rock according to the fusion point cloud;
and S05, performing high dangerous rock monitoring according to the structural plane position of dangerous rocks.
Optionally, performing densification processing on the sparse point cloud data to obtain a fused point cloud, including: preprocessing the sparse point cloud data and the dense matching point cloud data to obtain two types of heterogeneous point clouds; carrying out point cloud segmentation processing on each different source point cloud to obtain two classification labels; and obtaining a fused point cloud according to the binary classification label and the point cloud fusion algorithm.
In some embodiments, an energy function is constructed based on the distance and normal vector angle of the airborne LiDAR point cloud and the dense matching point cloud, and a smooth term is constructed by utilizing color difference and geometric relationship in the point cloud; obtaining two classification labels by adopting a graph cutting algorithm, and removing overlapped redundant point clouds according to the classification labels; and conducting guided point cloud filtering smoothing processing on the boundary point cloud, finally combining to generate heterogeneous point cloud, and finally realizing LiDAR sparse point cloud densification processing.
Optionally, preprocessing the sparse point cloud data and the dense matching point cloud data to obtain two types of heterogeneous point clouds, including: and removing outliers from the sparse point cloud data and the dense matching point cloud data respectively to obtain two types of heterogeneous point clouds. The two types of heterogeneous point clouds are respectively a sparse point cloud and a dense matching point cloud with outliers removed.
In some embodiments, the airborne laser radar scans a high-level dangerous rock area to obtain laser radar sparse point cloud data, and each laser point of the point cloud obtained by scanning of the laser radar is a real point on the surface of a measured object, so that the point cloud has the characteristics of accurate position, reliable data and no influence of ambient light, but has the defects of sparsity, incompleteness and the like. Aerial photography is carried out on the high dangerous rock area through an unmanned aerial vehicle, and dense matching point cloud data are obtained according to aerial photography images; the dense matching point cloud data is obtained by a motion recovery structure and a multi-view stereo matching algorithm, has the characteristics of dense and rich point cloud and color and texture information, but has limited matching point cloud precision and is influenced by shooting conditions, illumination conditions, image quality and the like. Due to different acquisition methods, the heterogeneous point cloud data to be fused have great differences, such as point cloud distribution uniformity, point cloud density, noise conditions and the like. According to the scheme, the dense matching point cloud of the visible light image and the sparse point cloud data of the laser radar are fused, the advantages of the visible light image and the sparse point cloud data are fully played, the problems of single data loss, data sparseness and point cloud cavities are solved, the advantages of the visible light image and the sparse point cloud cavities are complemented, more dangerous rock characteristics can be represented, more accurate structural surface displacement of dangerous rocks can be obtained according to the fused point cloud, and therefore the accuracy of high-position dangerous rock monitoring is improved.
Optionally, preprocessing the sparse point cloud data and the dense matching point cloud data includes: and carrying out coordinate conversion on the sparse point cloud data and the dense matching point cloud data, and uniformly converting the sparse point cloud data and the dense matching point cloud data to a geodetic coordinate system.
Optionally, the point cloud segmentation processing is performed according to the heterogeneous point clouds to obtain two classification labels, including:
by calculating
Obtaining a binary classification label;
wherein L = { L i Is a binary labelset for the sparse point cloud, and l i E {0,1}, the label is 0 to indicate that the sparse point cloud has substitution dense matching point cloud, and lambda is a regularization factor, pi [ ·]Is a discriminant function, Π [ true]＝1，Π[false]=0, sparse point cloud P = { P = i }。
In some embodiments, the approximate solution obtained by the two-class label through multiple iterations of the graph cut algorithm is the two-class label set L. The graph cut algorithm comprises two types of heterogeneous point clouds including sparse point cloud P = { P = { (P) } i And dense matching point cloud Q = { Q } i H, according to sparse point cloud P = { P = i And dense matching point cloud Q = { Q } i The geometric relationship, color relationship and similarity can define the energy function
In some embodiments, two types of heterogeneous point clouds are spliced through a registration algorithm, and point location offset exists in the overlapping part of the dense matching point cloud and the sparse point cloud, namely the splicing point cloud is easy to have a 'double-wall' redundant point cloud layering problem. The two classification labels are calculated through a graph cut algorithm, similar overlapping redundant parts between the two types of heterogeneous point clouds are removed in an optimized mode, and therefore the two types of heterogeneous point clouds can be fused and densified in a high-precision mode, and more accurate high-order dangerous rock information can be obtained.
Optionally, obtaining a fused point cloud according to a binary label and a point cloud fusion algorithm, including: splicing the two types of heterogeneous point clouds to form spliced point clouds; removing overlapped redundant point clouds of the spliced point clouds by using the binary classification labels to obtain initial fusion point clouds; and smoothing the initial fused point cloud according to a point cloud fusion algorithm to obtain a fused point cloud.
In some embodiments, the point cloud fusion algorithm comprises: and carrying out spatial rasterization processing on the initial fusion point cloud. And judging whether the number of the point clouds in the grid is less than 2, if so, deleting all data in the grid, thus deleting sparse points which deviate from the point clouds and are suspended above the point clouds, and if not, performing region growth on the grid to obtain grids belonging to the same piece of point cloud. Judging whether the grid number of each piece of point cloud is the maximum, if not, deleting the piece of point cloud data, thus deleting the point clouds which are far away from the center of the large piece of point cloud, small and dense, if not, calculating the average Euclidean distance of the point clouds of the grid and the average Euclidean distance in the point cloud field, and obtaining a point cloud distance value according to the average Euclidean distance of the point clouds of the grid and the average Euclidean distance in the certain point cloud field. And judging whether the point cloud distance value is larger than a preset threshold value, if so, calculating a bilateral filtering factor by adopting the point cloud in the neighborhood range, otherwise, calculating a bilateral filtering factor A by adopting the point cloud in the grid, and filtering by adopting C = P-An, wherein C is the filtered point cloud, P is the original point cloud, and n is the vector direction of the normal vector.
In some embodiments, after the sparse point cloud and the dense matching point cloud are spliced, a fused point cloud is formed, two types of heterogeneous point clouds are processed by using two classification labels, a part of the sparse point cloud is substituted for the dense matching point cloud, and at the moment, overlapping redundant parts similar to the dense matching point cloud in the sparse point cloud are well eliminated. However, in the process of obtaining dense matching point clouds by aerial picture images of the unmanned aerial vehicle, the non-rigid deformation of the image matching point clouds can be caused by inherent algorithm defects and attitude estimation errors of camera depth estimation, and particularly, the phenomena of obvious concave-convex fluctuation and uneven thickness of structures and dangerous rock point clouds are generated. This results in the fusion point cloud still having layering in the complex dangerous rock surface area, and there may be fracture and noise after segmentation. And the guide point cloud fusion algorithm is adopted to carry out smoothing treatment on the junction of the sparse point cloud and the dense matching point cloud, so that gaps and layering at the fusion are reduced, and more accurate fusion point cloud is finally obtained, thereby obtaining more accurate high-level dangerous rock information.
In some embodiments, filtering of the point cloud is accomplished by invoking individual filter objects. The main filters are a straight-through filter, a voxel grid filter, a statistical filter, a radius filter and the like. Filters with different characteristics form a complete point cloud pretreatment family, and can be combined to be used for finishing the smoothing treatment on the junction of the sparse point cloud and the dense matching point cloud.
Optionally, obtaining the structural surface displacement of the dangerous rock according to the fused point cloud includes: extracting key points in the fused point cloud; acquiring local point cloud characteristics according to the key points; matching the key points according to the local point cloud characteristics to obtain a dangerous rock point cloud set; acquiring a dangerous rock structural surface according to the dangerous rock point cloud set; and acquiring the structural plane displacement of the dangerous rock according to the structural plane of the dangerous rock.
Optionally, extracting key points in the fused point cloud includes: preprocessing the fused point cloud to obtain initial point cloud data; processing the initial point cloud data by using a point cloud partitioning strategy of a self-adaptive expansion receiving domain to obtain point cloud enhanced data; and inputting the point cloud enhancement data into a key point extraction network to obtain key points in the fused point cloud.
Optionally, preprocessing the fused point cloud to obtain initial point cloud data, including: performing ground segmentation processing on the fused point cloud; selecting a sampling point of the fused point cloud; constructing a local neighborhood of each sampling point; obtaining a least square plane of a local neighborhood; and acquiring a normal vector of the least square plane, and taking the normal vector of the least square plane as a normal vector of the sampling point.
In some embodiments, the amount of information in the fused point cloud is increased using a normal-calculation-based data enhancement method, which includes: selecting a sampling point of the fused point cloud; constructing a local neighborhood of each sampling point; acquiring a least square plane of a local neighborhood; and acquiring a normal vector of the least square plane, and taking the normal vector of the least square plane as a normal vector of the sampling point.
In some embodiments, the normal vector of the sample points is obtained based on PCA (principal component analysis). Selecting a sampling point of the fused point cloud, constructing a local neighborhood of each sampling point, replacing a single sampling point by constructing the local neighborhood, then fitting a least square plane of each local neighborhood, then obtaining a normal vector of the least square plane, and taking the plane normal vector as a normal vector of the sampling point.
Optionally, an arbitrary sampling point p is selected i Obtaining a sampling point p i Obtaining the mass center p of k local neighborhood points, and passing through the sampling point p i The centroid p of the k local neighborhood points fits the sampling point p i A least squares plane of the corresponding local neighborhood.
In some embodiments, a covariance matrix of a local neighborhood is obtained, eigenvalue decomposition is performed on the covariance matrix, and an eigenvector corresponding to the minimum eigenvalue of the covariance matrix is obtained, where the eigenvector is the sampling point p i The normal vector n.
Optionally, the eigenvector corresponding to the covariance matrix Σ minimum eigenvalue is obtained by the following formula:
wherein k represents a sampling point p i The number of points in the local neighborhood of (a),
is the centroid of the local neighborhood, λ j Is the jth eigenvalue of the covariance matrix, and the eigenvalue satisfies lambda 0 ≤λ 1 ≤λ 2 ，
For the eigenvector corresponding to the jth eigenvalue, sample point p i Has a surface curvature δ of
j is an integer.
Optionally, ground points in the fused point cloud are removed by using a point cloud segmentation method, so that ground segmentation processing is performed on the fused point cloud.
In some embodiments, since it is not necessary to solve the precise ground plane, only the point set belonging to the ground plane needs to be found, so that the limited interior points can be directly used to solve the linear equation set to obtain the ground model parameters, and then the point set of the ground plane is obtained through the ground model.
In some embodiments, in the key point extraction algorithm process, firstly, the normal solution and ground segmentation technology are used to complete the preprocessing of the fused point cloud, so as to obtain initial point cloud data containing more effective information; then, point cloud division is completed by using a point cloud division strategy based on a self-adaptive expansion receiving domain, and a plurality of local point cloud sets are obtained; performing centralization operation on each point cloud set, calculating the distance from each point to a central point, and adding a normal vector and the distance to point cloud information to obtain point cloud enhanced data; and finally, inputting the point cloud enhanced data into a key point extraction network to obtain key points with expression capability and corresponding unreliability.
Optionally, the point cloud partitioning strategy for adaptively expanding the receiving domain includes: and (3) dividing the fused point cloud space to obtain a plurality of local subspaces, extracting the features of the subspaces, and selecting space coordinates according to the extracted features.
In some embodiments, the concept of expansion (scaling) is first proposed by two-dimensional expansion convolution, and the algorithm expands the size of the convolution kernel by adding holes in the middle of the convolution kernel, which enables the convolution operation to have a larger perception range without increasing the amount of calculation, and can acquire features containing more interrelations. The point cloud partitioning strategy of the self-adaptive expansion receiving domain firstly partitions a fused point cloud space into a plurality of local subspaces, extracts features for each subspace, and calculates the space coordinates which are most suitable as key points for registration in the space based on the features.
In some embodiments, in the point cloud partitioning policy for the adaptive expanded receiving domain, if a certain node encounters other nodes during the proximity search, the policy may first determine whether the receiving domain of the node satisfies the minimum size, if so, stop the search, and if not, continue the proximity search to expand the receiving domain until M nodes are included. As shown in FIG. 2, the receiving domain of the node on the left side stops expanding due to the nearest neighbor searching of other nodes, and the receiving domain of the node on the right side expands to M due to the absence of other nodes D In (c) is used. The self-adaptive expansion receiving domain point cloud partitioning strategy solves the problem of adjacent nodes brought by random sampling, reduces the overlapping degree of local regions, and enables extracted key points to be more dispersed and more effective.
Optionally, the key point extraction network is obtained by unsupervised learning of the PointNet feature extraction network. In some embodiments, in the point cloud correlation problem, pointNet has been an initiative, which solves the disorder and rigid invariance of point cloud, and its feature extraction network is also quite universal, and most point cloud processing algorithms based on deep learning at present use PointNet as a feature extraction network.
In some embodiments, in the point cloud data, the structural information amount of different points is different from the probability of matching points existing in the target point cloud, for example, in the point cloud data of dangerous rock masses, the rock masses are continuously irregular surfaces and obviously different from the characteristics of ground leveling structural surfaces such as highway subgrades and the like. Based on the method, extraction of the key points is realized by adopting a key point extraction network based on unsupervised learning and a self-adaptive expansion receiving domain point cloud partitioning strategy. Unsupervised learning is a new trend of deep learning development at present. On one hand, the unsupervised learning utilizes the characteristics of an autoencoder, and can realize the spanning generation from one type to another type; on the other hand, the unsupervised learning can realize the feature learning only by a small amount of labeled data and a large amount of unlabeled data.
In some embodiments, a point cloud partitioning strategy for adaptively expanding a receiving domain is designed based on the idea of expanding convolution in two-dimensional convolution, a fused point cloud is properly partitioned into a plurality of local regions, and the perception domain of a local point set is increased while the calculation amount is slightly increased. Obtaining the weight of each point through the fusion of symmetric functions, and weighting each point in the local space to obtain a key point; and finally, embedding the key point extraction network into a Simese network structure, and using the chamfering distance combined with probability as a target function to finish unsupervised training of the network.
Optionally, obtaining local point cloud features according to the key points includes: extracting a plurality of local point sets from key points by using a point cloud division strategy of a self-adaptive expansion receiving domain, carrying out centralization operation on each local point set, calculating the distance between each point in the local point sets and each key point, adding the normal vector of each key point and the distance between each point and each key point to the spatial information of each key point, and inputting the spatial information into a local feature extraction network to obtain local point cloud features.
In some embodiments, the key points are used as random sampling nodes, two local point sets with small size and large size are extracted from each key point by using a point cloud division strategy of a self-adaptive expansion receiving domain, the two local point sets are respectively subjected to centralization operation, the distances between each point in the local point sets and the key points are respectively calculated, the normal vectors of the key points and the distances are added to the spatial information of each key point to enhance data, and the spatial information is input into a local feature extraction network to obtain local point cloud features with hierarchy information fused. The implementation carries out unsupervised network training on the Siamese structure to obtain a feature extraction network of local point clouds, the Siamese network structure is completely suitable for the problem of point cloud registration extraction with two approximate inputs, different inputs are projected into the same vector space, and the difficulty of constructing an unsupervised target function is reduced.
Optionally, the matching of the key points according to the local point cloud features to obtain a dangerous rock point cloud set includes: obtaining a matching relation between the key points and the source point cloud by utilizing a nearest neighbor search algorithm according to the key points and the local point cloud characteristics; matching the key points with the source point cloud according to the matching relation to obtain a plurality of matching point clouds; and obtaining a dangerous rock point cloud set according to the matched point cloud.
In some embodiments, the parameter estimation method is a modified algorithm based on the RANSAC algorithm framework. And screening point cloud data which accord with parameter transformation from the key point set and the local point cloud characteristics by a parameter estimation method. And (3) calculating the unreliability of the key point set, the local point cloud characteristics and the key points by using an RANSAC (random sample consensus) improved algorithm to obtain point cloud data conforming to parameter transformation, inputting the point cloud data into a transformation matrix to obtain point cloud transformation parameters, calculating the point cloud transformation parameters by using an ICP (inductively coupled plasma) algorithm to obtain primary matching parameters, inputting the primary matching parameters into a fine registration matrix for re-matching to obtain matched point cloud, and performing target extraction and segmentation according to the matched point cloud to obtain a dangerous rock point cloud set.
Optionally, the matching the key points according to the local point cloud features to obtain a dangerous rock point cloud set, including: screening local point cloud characteristics and key points by adopting a parameter estimation algorithm to obtain point cloud data conforming to parameter transformation; inputting point cloud data into an initial transformation matrix to obtain point cloud transformation parameters; performing initial matching on point cloud transformation parameters by adopting an ICP (inductively coupled plasma) algorithm to obtain initial matching parameters; inputting the initial matching parameters into a fine registration matrix for matching again to obtain a matching point cloud; and extracting and dividing the target according to the matching point cloud to obtain a dangerous rock point cloud set.
In some embodiments, RANSAC (randomasmampreleconsenssus, random sampling consistency) may find a sample set conforming to some mathematical models from a set of observation data, and estimate parameters of the mathematical models. ICP (Iterative Closest Point, closest Iterative algorithm) is the most classical data registration algorithm. And the ICP calculates a corresponding point pair between the key point and the source point cloud, constructs a rotational translation matrix based on the corresponding point pair, transforms the key point to the coordinate system of the source point cloud by using the calculated matrix, estimates an error function of the transformed key point and the source point cloud, and iterates the operation until the given error requirement is met if the error function value is greater than a threshold value.
In some embodiments, according to the obtained key points and the corresponding local cloud point features, a closest search algorithm is used to find a matching relationship between the source point cloud and the key points, and a sufficient number of matching point pairs can more accurately estimate a transformation relationship between the point clouds.
Optionally, obtaining a dangerous rock structural plane according to the dangerous rock point cloud set includes: and performing aggregation classification on the dangerous rock point cloud sets by using a clustering algorithm to obtain a dangerous rock structural surface.
In some embodiments, in the current research on identification and extraction of a three-dimensional point cloud rock structural surface, a point cloud triangular mesh reconstruction method is mostly adopted, and a normal vector of the structural surface is calculated by fitting a plane equation through a least square method so as to obtain the attitude. However, this method has the following disadvantages: (1) Because the rock mass structural plane is not an ideal plane, the surface is usually uneven, and the rock mass point cloud is limited by a triangulation network reconstruction algorithm in the triangulation network reconstruction process, partial details of the rock mass surface are lost, so that the identification and extraction accuracy is reduced. (2) The important task of triangular mesh reconstruction is to clean noise points of point cloud data, and the cleaning of vegetation points covering the surface of a rock mass is a difficult problem. The clustering algorithm is adopted to carry out aggregation classification on the structural surface point cloud, so that the point cloud of the same group of structural surfaces is extracted from the whole rock mass point cloud, the loss of details on the rock mass surface is avoided, the vegetation on the rock mass surface is cleaned, and the accuracy of high-level dangerous rock identification is improved.
In some embodiments, there are many ways to aggregate classification, and the K-means algorithm is one of the most widely used clustering algorithms. The method has the advantages of simplicity, rapidness, capability of processing data of a huge data set, and particular suitability for processing high-density rock mass three-dimensional point cloud data. The K-means algorithm is an algorithm for iteratively solving clustering, and belongs to a classification algorithm of unsupervised learning. Assume that there is a data set X that needs to be divided into K groups. Then K objects are randomly chosen as the initial cluster centers μ 1, μ 2, \8230;, μ K, and then each object X in the data set X is computed i With each initial cluster center u k And each object is assigned to the cluster center closest to him. Each cluster comprises a cluster center and an object assigned to this cluster center, a process referred to as a round of computation. When all objects are assigned, the cluster center for each cluster is recalculated based on the existing objects in the cluster, and the process is repeated. The entire calculation process is repeated until it is satisfied that no cluster centers have changed and no objects have been reassigned to different clusters. The calculation process is expressed by the following formula, wherein the smaller the value of E, the smaller the distance of the samples in the cluster is, and the higher the similarity is.
Wherein x is a sample in the cluster, u is the center of the cluster, and E is the average of Euclidean distances from all objects to the respective cluster center.
Optionally, after acquiring the dangerous rock structural plane according to the dangerous rock cloud set, the method further includes: calculating a normal vector of the structural surface; and coloring the point cloud by adopting an HSV color space.
In some embodiments, a point cloud normal vector structural surface recognition extraction based on K-means classification is employed, and then the point cloud is colored using HSV color space. As shown in fig. 4, the HSV color space is a cone, and the circumference of the bottom of the cone represents hue (H), measured by angle, with a value range of 0-360 °; the degree that the color approaches to the spectral color from the center of the circle of the bottom surface to the circumference represents the saturation (S), and the larger the proportion of the spectral color is, the closer the color is to the spectral color, and the higher the saturation is. Lightness (V) describes the shade of a color. Therefore, the fluctuation structure of the three-dimensional rock is rendered by using HSV color, the normal direction of the point cloud is quantized, so that the points on the continuous plane structure are in the same color, the attitude information of the structural plane is represented by the color, the division of the discontinuous structural plane of the three-dimensional rock structure is greatly promoted by the automatic processing mode, and the distribution rule and scale of the rock structure plane can be more visually seen on the point cloud.
Optionally, according to dangerous rock structural plane obtain the structural plane displacement of dangerous rock, include: grouping a plurality of structural surfaces of the dangerous rock to form a plurality of structural surface groups; forming a structural block of the rock mass according to the plurality of structural surface groups; obtaining the displacement condition of each plane in the structure block in the respective normal vector direction; and acquiring the structural plane displacement of the dangerous rock according to each displacement condition.
In some embodiments, after the extraction of the dangerous rock structural surface is completed, two or more groups of dangerous rock structural surfaces are required to be matched to form a structural block, and displacement measurement is performed by matching the structural block according to the blockiness. One group of structural surfaces comprises a plurality of mutually parallel structural surfaces, the criss-cross of the structural surfaces can form an independent structural block, a single structural block can be formed by three or more groups of structural surfaces, and one rock mass often comprises three or more groups of structural surfaces. The volume of the structural mass is represented by the average or distribution of the mass of the structural mass, wherein the mass is a very important parameter for grading the rock mass, for example, the mass is an indispensable input parameter in the geological strength parameter.
In some embodiments, two common methods for estimating the block size are used: exponential estimation and simulation. In the index method, the average size of the structural blocks is used to represent the rock block size, and although the rock block information obtained by the method is not accurate enough, the simple, efficient and low-cost method is still widely applied to a general rock mass evaluation system. The predecessors have conducted a great deal of research on blockiness calculations using the exponential method and have proposed many calculation parameters. The simulation method is mainly used for simulating a structural surface network model according to the existing structural surface information and extracting structural blocks on the basis. In recent years, many algorithms have been proposed to extract the building blocks. These algorithms include two-dimensional and three-dimensional algorithms, where the two-dimensional algorithms assume structural planes as straight lines and can compute polygons where these lines intersect. The two-dimensional algorithm mainly comprises theories such as a graph theory, a matrix, a directed graph, a vector, a numerical method and the like. The three-dimensional algorithm assumes that the structural planes are planar in space, and the planes are interlaced with each other to form a polyhedron.
Optionally, obtaining the displacement condition of each plane in the structural block in the corresponding normal vector direction includes: taking each plane of the structure block at the previous moment as a reference plane; and obtaining each distance change value of each current plane of the structural block and the reference plane in the normal vector direction, wherein each distance change value is the displacement of each plane in the structural block in the corresponding self-normal vector direction.
Optionally, obtaining structural plane displacement of the dangerous rock according to each of the displacement situations comprises: and comparing the displacement of each plane in the structural block in the direction corresponding to each normal vector to obtain the displacement of the structural surface of the dangerous rock.
In some embodiments, the structural blocks in actual engineering are generally stable whole bodies, and when the translation of the structural blocks is calculated, the structural blocks can be regarded as rigid bodies for analysis. Therefore, the displacement measurement of the dangerous rock can be converted into the displacement measurement of the rigid body formed by the point cloud of the rock mass. For a rigid body, the displacements of the rigid body can be obtained by adding the displacements of a plurality of planes forming the rigid body along the normal vector of the rigid body. Considering that the fitting accuracy of the several surfaces is higher, compared with the prior art in which point clouds are directly subtracted, the error can be greatly reduced by analyzing the translation of the structural block through the distance change condition of a plurality of planes along the normal vector of the structural block. In this embodiment, the distance variation condition of the plane in the normal vector direction of the plane is statistically analyzed through the distances from all points on the plane of the structural block to the matching plane. And then the distance change is converted into the displacement of the rigid body in the X, Y and Z directions, so that more accurate displacement of the structural plane of the dangerous rock is obtained, and the accuracy of monitoring the high-level dangerous rock is improved.
In some embodiments, after the heterogeneous point cloud fusion densification, the point cloud target identification and extraction are completed, the displacement measurement of the high dangerous rock is performed on the extracted target point cloud set, and then the high dangerous rock is monitored. For displacement measurement, direct subtraction of two sets of measurement target point clouds (point cloud subtraction) is the most common method. And calculating the variation of the two groups of point clouds in the given direction to obtain the dangerous rock deformation condition of the area. However, under natural conditions, the accuracy of the point cloud data acquired by the scanner is almost in the centimeter level, and the deformation measurement is directly affected by the error brought by the equipment. The traditional point cloud subtraction method does not fully utilize the maximum characteristics of the point cloud: due to the high-precision and high-density point cloud data, the displacement measurement of the high-level dangerous rock is not accurate enough. According to the scheme, a method of extracting a structural plane is adopted for dangerous rocks, the dangerous rock plane is tracked and measured, and high-precision and stable plane displacement measurement is realized by adopting filtering algorithm processing, so that the small displacement and the stability of dangerous rock bodies are reflected.
In some embodiments, fig. 5 is a flow chart of the displacement measurement of the high-level dangerous rock provided by this embodiment, and as shown in fig. 5, the point cloud after the segmentation processing of the dangerous rock in the periods T1 and T2 is respectively subjected to rock structure surface extraction, then two extracted rock structure surfaces are subjected to intelligent matching, and finally, the rock displacement calculation and analysis are performed according to the intelligent matching result, so as to realize the tracking measurement of the high-level dangerous rock target. According to the high-order dangerous rock monitoring method based on the airborne laser radar, the structure blocks are extracted by utilizing the high-density point cloud fitting plane, the specific structure blocks are used as feature objects to conduct contrastive analysis, the displacement monitoring of high-order dangerous rocks can be conducted with higher precision, and meanwhile, the change condition of more-dimensional dangerous rocks can be analyzed.
Optionally, the high-altitude dangerous rock monitoring is carried out according to the structural plane position of dangerous rocks, and comprises: and under the condition that the displacement of the dangerous rock is greater than a preset threshold value, performing high-level dangerous rock early warning. In some embodiments, performing high-level dangerous rock forewarning comprises: sending preset early warning information to a preset client, displaying a red danger exclamation mark or sending a preset danger warning sound.
In some embodiments, the dangerous rock and soil part is identified and extracted from the fused point cloud, and the point cloud of the dangerous rock body to be observed is extracted from a larger three-dimensional model point cloud containing a plurality of environments around the dangerous rock, such as trees, slopes, roads and the like. After the heterogeneous point cloud fusion densification, the point cloud target identification and extraction are completed, displacement measurement needs to be carried out on the extracted dangerous rock point cloud set. For displacement measurement, direct subtraction (point cloud subtraction) of two sets of measurement target point clouds is the most common method. The point cloud subtraction method calculates the variation of two groups of point clouds in a given variation direction, and then the dangerous rock deformation condition of the area can be obtained. However, under natural conditions, the accuracy of the point cloud data acquired by the scanner is almost in the centimeter level, and the deformation measurement is directly affected by the error brought by the equipment. The main reason for this problem is that the traditional point cloud subtraction method does not fully utilize the maximum characteristics of the point cloud: high precision and high density point cloud data. According to the scheme, the extracted dangerous rock point cloud is subjected to structural surface extraction, then intelligent matching is carried out on the structural surface, and finally, dangerous rock displacement analysis is completed, so that tracking measurement of a high-level dangerous rock target is realized. Meanwhile, the point cloud data with high precision and high density are utilized to fit the plane so as to extract the blocks, the specific blocks are used as characteristic objects for comparative analysis, the displacement monitoring of high-order dangerous rocks can be carried out with higher precision, and the change condition of the dangerous rocks with more dimensions can be analyzed.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; such modifications and substitutions do not depart from the spirit and scope of the embodiments of the present invention, and they should be construed as being covered by the appended claims and their equivalents.
Claims (14)
1. A high-order dangerous rock monitoring method based on an airborne laser radar is characterized by comprising the following steps:
acquiring sparse point cloud data according to a scanning result of the laser radar on the high dangerous rock area;
obtaining dense matching point cloud according to the image shot for the high dangerous rock area;
performing densification processing on the sparse point cloud data by using the dense matching point cloud to obtain a fusion point cloud;
acquiring the structural surface displacement of the dangerous rock according to the fusion point cloud;
and carrying out high-position dangerous rock monitoring according to the structural plane position of the dangerous rock.
2. The method of claim 1, wherein densifying the sparse point cloud data to obtain a fused point cloud comprises:
preprocessing the sparse point cloud data and the dense matching point cloud data to obtain two types of heterogeneous point clouds;
performing point cloud segmentation processing on the heterogeneous point clouds to obtain two classification labels;
and obtaining a fused point cloud according to the two classification labels and the point cloud fusion algorithm.
3. The method of claim 2, wherein performing point cloud segmentation processing according to each heterogeneous point cloud to obtain two classification labels comprises:
by calculation of
Obtaining a binary classification label;
wherein, L = { L = i Is a binary labelset for the sparse point cloud, and l i E {0,1}, the label is 0 to indicate that the sparse point cloud has substitution dense matching point cloud, and lambda is a regularization factor, pi [ ·]Is a discriminant function, Π [ true]＝1，∏[false]=0, sparse point cloud P = { P = i }。
4. The method of claim 2, wherein obtaining a fused point cloud according to the two-class label and point cloud fusion algorithm comprises:
splicing the two types of heterogeneous point clouds to form spliced point clouds;
removing overlapped redundant point clouds of the spliced point clouds by using the binary classification labels to obtain initial fusion point clouds;
and smoothing the initial fusion point cloud according to a point cloud fusion algorithm to obtain a fusion point cloud.
5. The method of claim 1, wherein obtaining structural surface displacements of the crisis from the fused point cloud comprises:
extracting key points in the fused point cloud;
obtaining local point cloud characteristics according to the key points;
matching the key points according to the local point cloud characteristics to obtain a dangerous rock point cloud set;
acquiring a dangerous rock structural surface according to the dangerous rock point cloud set;
and acquiring the structural plane displacement of the dangerous rock according to the structural plane of the dangerous rock.
6. The method of claim 5, wherein extracting keypoints in the fused point cloud comprises:
preprocessing the fused point cloud to obtain initial point cloud data;
processing the initial point cloud data by using a point cloud partitioning strategy of a self-adaptive expansion receiving domain to obtain point cloud enhanced data;
and inputting the point cloud enhanced data into a key point extraction network to obtain key points in the fused point cloud.
7. The method of claim 6, wherein preprocessing the fused point cloud to obtain initial point cloud data comprises:
performing ground segmentation processing on the fused point cloud;
selecting a sampling point of the fused point cloud;
constructing a local neighborhood of each sampling point;
obtaining a least squares plane of the local neighborhood;
and acquiring a normal vector of the least square plane, and taking the normal vector of the least square plane as a normal vector of a sampling point.
8. The method of claim 5, wherein obtaining local point cloud features from the keypoints comprises:
extracting a plurality of local point sets from the key points by using a point cloud division strategy of a self-adaptive expansion receiving domain;
performing centralization operation on each local point set, and calculating the distance from each point in the local point set to a key point;
adding the normal vector of the key points and the distance between each point and each key point to the spatial information of each key point;
and inputting the spatial information into a local feature extraction network to obtain local point cloud features.
9. The method of claim 5, wherein pairing the key points according to the local point cloud features to obtain a cloud set of dangerous rock points comprises:
obtaining a matching relation between the key points and the source point cloud by utilizing a nearest neighbor search algorithm according to the key points and the local point cloud characteristics;
matching the key points with the source point cloud according to the matching relation to obtain a plurality of matching point clouds;
and obtaining a dangerous rock point cloud set according to each matching point cloud.
10. The method of claim 9, wherein obtaining a cloud set of crisis points from each of the matching point clouds comprises:
screening local point cloud characteristics and key points by adopting a parameter estimation algorithm to obtain point cloud data conforming to parameter transformation;
inputting point cloud data into an initial transformation matrix to obtain point cloud transformation parameters;
performing initial matching on point cloud transformation parameters by adopting an ICP (inductively coupled plasma) algorithm to obtain initial matching parameters;
inputting the initial matching parameters into a fine registration matrix for matching again to obtain a matching point cloud;
and extracting and dividing the target according to the matching point cloud to obtain a dangerous rock point cloud set.
11. The method of claim 5, wherein obtaining a crisis-rock structure plane from the cloud set of crisis-points comprises:
and performing aggregation classification on the dangerous rock point cloud sets by using a clustering algorithm to obtain a dangerous rock structural surface.
12. The method of claim 5, wherein obtaining the structural plane displacement of the crisis rock from the structural plane of the crisis rock comprises:
grouping a plurality of structural surfaces of the dangerous rock to form a plurality of structural surface groups;
forming a structural block of a rock mass according to the plurality of structural surface groups;
acquiring the displacement condition of each plane in the structural block in the corresponding normal vector direction;
and acquiring the structural plane displacement of the dangerous rock according to the displacement conditions.
13. The method of claim 12, wherein obtaining the displacement of each plane in the structure block in the corresponding normal vector direction comprises:
taking each plane of the structure block at the previous moment as a reference plane;
and obtaining each distance change value of each current plane of the structural block and the reference plane in the normal vector direction, wherein each distance change value is the displacement of each plane in the structural block in the corresponding self-normal vector direction.
14. The method of claim 13, wherein obtaining structural plane displacements of the crisis rock from each of the displacement scenarios comprises:
and comparing the displacement of each plane in the structural block in the direction corresponding to each normal vector for many times to obtain the displacement of the structural surface of the dangerous rock.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211410024.5A CN115619977A (en) | 2022-11-10 | 2022-11-10 | High-order dangerous rock monitoring method based on airborne laser radar |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211410024.5A CN115619977A (en) | 2022-11-10 | 2022-11-10 | High-order dangerous rock monitoring method based on airborne laser radar |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115619977A true CN115619977A (en) | 2023-01-17 |
Family
ID=84878749
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211410024.5A Pending CN115619977A (en) | 2022-11-10 | 2022-11-10 | High-order dangerous rock monitoring method based on airborne laser radar |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115619977A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115861296A (en) * | 2023-02-14 | 2023-03-28 | 湖北工业大学 | Automatic identification method and system for dangerous rock mass with high and steep slope based on unmanned aerial vehicle point cloud |
-
2022
- 2022-11-10 CN CN202211410024.5A patent/CN115619977A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115861296A (en) * | 2023-02-14 | 2023-03-28 | 湖北工业大学 | Automatic identification method and system for dangerous rock mass with high and steep slope based on unmanned aerial vehicle point cloud |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xia et al. | Geometric primitives in LiDAR point clouds: A review | |
| Yang et al. | An individual tree segmentation method based on watershed algorithm and three-dimensional spatial distribution analysis from airborne LiDAR point clouds | |
| Sun et al. | Aerial 3D building detection and modeling from airborne LiDAR point clouds | |
| Zhu et al. | Facade reconstruction using multiview spaceborne TomoSAR point clouds | |
| Biosca et al. | Unsupervised robust planar segmentation of terrestrial laser scanner point clouds based on fuzzy clustering methods | |
| CN102004922B (en) | High-resolution remote sensing image plane extraction method based on skeleton characteristic | |
| Yang et al. | Voxel-based extraction of transmission lines from airborne lidar point cloud data | |
| CN113066162B (en) | Urban environment rapid modeling method for electromagnetic calculation | |
| Xu et al. | Unsupervised segmentation of point clouds from buildings using hierarchical clustering based on gestalt principles | |
| Wang et al. | A bayesian approach to building footprint extraction from aerial lidar data | |
| CN111652241B (en) | Building contour extraction method integrating image features and densely matched point cloud features | |
| Sun et al. | Classification of MLS point clouds in urban scenes using detrended geometric features from supervoxel-based local contexts | |
| Guo et al. | Extraction of dense urban buildings from photogrammetric and LiDAR point clouds | |
| CN115619977A (en) | High-order dangerous rock monitoring method based on airborne laser radar | |
| CN112634447B (en) | Outcrop stratum layering method, device, equipment and storage medium | |
| Chen et al. | An efficient global constraint approach for robust contour feature points extraction of point cloud | |
| CN112070787B (en) | Aviation three-dimensional point cloud plane segmentation method based on opponent reasoning theory | |
| Yu et al. | 3-D feature matching for point cloud object extraction | |
| JP2023530449A (en) | Systems and methods for air and ground alignment | |
| Bretar | Feature extraction from LiDAR data in urban areas | |
| Omidalizarandi et al. | Segmentation and classification of point clouds from dense aerial image matching | |
| Santos et al. | Classification of LiDAR data over building roofs using k-means and principal component analysis | |
| Ghosh et al. | Heuristical feature extraction from LiDAR data and their visualization | |
| Wang et al. | 3D building reconstruction from LiDAR data | |
| Yu et al. | Bidirectionally greedy framework for unsupervised 3D building extraction from airborne-based 3D meshes |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |