Slope dangerous rock identification and monitoring method based on unmanned aerial vehicle inspection

Abstract

The present application relates to the technical field of geological engineering, and especially provides a slope dangerous rock identification and monitoring method based on unmanned aerial vehicle inspection, comprising the following steps: step 1: collecting slope image information by adopting unmanned aerial vehicle inspection to obtain a first point cloud model, and performing vegetation removal pretreatment on the first point cloud model; step 2: extracting a main joint normal vector direction of the slope based on the first point cloud model; step 3: determining a dangerous rock position of the slope based on the main joint normal vector direction of the slope and by using an equatorial projection analysis method and a normal vector matching method; and step 4: repeatedly performing steps 1 to 3 on the same slope according to a preset time interval to obtain a second point cloud model, and obtaining a slope change result after point cloud model registration and relative distance calculation are performed on the second point cloud model and the first point cloud model. The purpose is to solve the problem of insufficient non-contact rapid identification of slope dangerous rocks and achieve the technical effect of subsequent slope change monitoring.

Description

Technical Field

[0001] This invention relates to the field of geological engineering technology, and more specifically, to a method for identifying and monitoring dangerous rocks on slopes based on unmanned aerial vehicle (UAV) inspection. Background Technology

[0002] Currently, traditional engineering applications for identifying unstable rock slopes primarily rely on on-site surveys by engineers. This involves conducting field measurements of slope discontinuities and performing field and laboratory tests on rock properties, weathering levels, and strength. Combined with the engineer's experience, unstable rock formations are identified, and slope stability is assessed. However, the potential instability of rock slope areas poses safety hazards to on-site surveys. Furthermore, for higher and steeper slopes, manual measurement is further limited, making it impossible to obtain comprehensive information about the slope's structural surfaces.

[0003] Clearly, methods based on UAV oblique photography to acquire structural surface information and perform rapid identification and monitoring of unstable rock masses can solve these problems, thereby improving the efficiency of unstable rock mass identification and reducing the risks associated with on-site investigation. Therefore, to address the above shortcomings, this invention provides a method for identifying and monitoring unstable rock masses on slopes based on UAV inspection. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying and monitoring dangerous rocks on slopes based on unmanned aerial vehicle (UAV) inspection, in order to solve the problem of insufficient research on non-contact rapid identification of dangerous rocks on slopes and to achieve the technical effect of subsequent monitoring of slope changes.

[0005] This invention provides a method for identifying and monitoring dangerous rockfalls on slopes based on unmanned aerial vehicle (UAV) inspection, comprising the following steps:

[0006] Step 1: Use drones to inspect and collect slope images to obtain the first point cloud model, and perform vegetation removal preprocessing on the first point cloud model;

[0007] Step 2: Extract the normal vector directions of the main joints of the slope based on the first point cloud model;

[0008] Step 3: Determine the location of dangerous rocks on the slope based on the normal vector direction of the main joints and using stereographic projection analysis and normal vector matching methods;

[0009] Step 4: Repeat steps 1 to 3 for the same slope according to the preset time interval to obtain the second point cloud model. After registering the second point cloud model with the first point cloud model and calculating the relative distance, the result of the slope change is obtained.

[0010] Furthermore, the specific steps of acquiring the first point cloud model by using unmanned aerial vehicle (UAV) inspection and collecting slope image information include:

[0011] Select a drone equipped with differential GNSS to fly along a preset path, and the drone will acquire information from two adjacent images of an overlapping area with a set overlap ratio.

[0012] A sparse point cloud model is obtained by aerial triangulation based on image information collected by UAV, and a 3D mesh model of slope with texture information is obtained based on 3D reconstruction.

[0013] The first point cloud model was obtained by using CloudCompare software and sampling based on density.

[0014] Furthermore, the process of performing vegetation removal preprocessing on the first point cloud model also includes:

[0015] Establish a Kd-tree for the first point cloud model, and calculate the K nearest neighbor set of each point in the first point cloud model based on the Kd-tree;

[0016] Construct the covariance matrix of the neighborhood point set and perform eigenvalue decomposition;

[0017] The PCA principal component analysis method is used to calculate the eigenvalues ​​and corresponding eigenvectors of each point in the neighborhood point set.

[0018] Furthermore, the vegetation removal pretreatment specifically includes:

[0019] Calculate the curvature corresponding to each point based on the characteristic values ​​of each point;

[0020] Convert the RGB values ​​corresponding to the colors of each point to HSV values;

[0021] Filter and remove points whose HSV values ​​are within a preset range and whose curvature is greater than a preset value.

[0022] Furthermore, step 2 specifically includes:

[0023] Manually select areas with obvious slope joint characteristics;

[0024] Based on the region growing algorithm, the slope joint surface is segmented according to the preset normal vector angle threshold and curvature threshold;

[0025] The RANSAC algorithm is used to fit the plane to each segmented point to obtain the three-dimensional normal vector of each joint surface;

[0026] Convert the three-dimensional normal vectors of each joint surface into unit vectors and make their Z coordinates negative;

[0027] Using the lower spherical stereographic projection, the three-dimensional normal vector is converted into a two-dimensional stereographic projection map;

[0028] Based on the coordinates of two-dimensional projection points on a two-dimensional stereographic projection map, KDE kernel density estimation and meanshift clustering are used to determine the direction of the normal vector of the main joints of the slope.

[0029] Furthermore, the stereographic projection analysis method also includes calculating the dip and dip angle of the main joint surfaces based on the direction of the normal vector of the main joint of the slope, and obtaining the normal vector of the unstable joint surface using the stereographic projection analysis method.

[0030] Further, step 3 includes:

[0031] The Z-coordinate value of the normal vector direction of the main joint of the slope is corrected to a negative value;

[0032] Calculate the angle between the normal vector at each point after correction and the normal vector of the unstable joint surface;

[0033] If the included angle is less than the threshold, it is judged as a dangerous rock, and then the DBSCAN clustering algorithm is used to determine the final location of the dangerous rock.

[0034] Furthermore, the point cloud model registration specifically includes:

[0035] Manually select feature point pairs for initial registration;

[0036] The ICP algorithm is used to further register the data based on the initial registration.

[0037] Furthermore, the results of obtaining slope changes specifically include:

[0038] Calculate the relative distance of each point in the registered second point cloud model to the nearest point in the first point cloud model and assign a new color to the registered point cloud model to obtain the third point cloud model;

[0039] The Quickhull algorithm was used to calculate the coordinates of the convex hull vertices of the dangerous rock locations in the second point cloud model.

[0040] The portion of the convex hull vertex coordinates that corresponds to the third point cloud model is the part of the point cloud region that changes.

[0041] The beneficial effects of this invention include:

[0042] 1. This invention extracts joint features by selecting joint feature regions, which differs from the existing technology of whole slopes. It only applies stereographic projection analysis results to the identification of dangerous rocks in the whole slope, which can reduce the calculation time for dangerous rock identification and improve the efficiency of dangerous rock identification.

[0043] 2. This invention involves conducting multiple inspections of the same slope. The point cloud model obtained during subsequent inspections can be used for point cloud registration and distance calculation to obtain information about slope changes. On the other hand, by extracting the convex hull of dangerous rocks, it is possible to observe whether the dangerous rock area has changed. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 A flowchart of a slope dangerous rock identification and monitoring method provided in an embodiment of the present invention;

[0046] Figure 2 A flowchart of a slope dangerous rock identification and monitoring method provided in an embodiment of the present invention;

[0047] Figure 3 This is a flowchart of vegetation removal provided in an embodiment of the present invention;

[0048] Figure 4 This is a slope effect diagram before vegetation removal provided in an embodiment of the present invention;

[0049] Figure 5 This is a slope effect diagram after vegetation removal provided in an embodiment of the present invention;

[0050] Figure 6 This is a stereographic projection of the lower hemisphere provided in an embodiment of the present invention;

[0051] Figure 7 This is a diagram illustrating the KDE kernel density estimation results provided in an embodiment of the present invention.

[0052] Figure 8 The diagram shows the meanshift clustering results provided in an embodiment of the present invention. Detailed Implementation

[0053] The technical solutions of the present invention will now be described with reference to the accompanying drawings in the embodiments of the present invention.

[0054] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0055] Please refer to Figure 1 This invention provides a method for identifying and monitoring dangerous rockfalls on slopes based on unmanned aerial vehicle (UAV) inspection, comprising the following steps:

[0056] Step 1: Select a UAV equipped with differential GNSS and fly it along a pre-set detection path. During the UAV's image capture, there should be a 60% to 70% overlap between adjacent images to facilitate subsequent 3D reconstruction. Then, based on the image information collected by the UAV, perform aerial triangulation to obtain a sparse point cloud model. Next, based on 3D reconstruction, obtain a 3D mesh model of the slope with texture information. Finally, use CloudCompare software on the computer and sample based on density. During the sampling process, there should be at least 100 points per unit area to obtain the first point cloud model.

[0057] The process includes the following steps before performing vegetation removal preprocessing on the first point cloud model:

[0058] Establish a Kd-tree for the first point cloud model, and calculate the K nearest neighbor set of each point in the first point cloud model based on the Kd-tree;

[0059] Construct the covariance matrix of the neighborhood point set and perform eigenvalue decomposition;

[0060] Principal component analysis (PCA) is used to calculate the eigenvalues ​​and corresponding eigenvectors of each point in the neighborhood point set.

[0061] The following is a simple example of point cloud preprocessing according to the present invention:

[0062] To reduce the number of points in the first point cloud model, this disclosure requires voxel downsampling of the first point cloud model before building the Kd-tree, where the voxel size is chosen to be 0.05; using a neighborhood set N of the first point cloud model... i ={p k Let's take} as an example to illustrate:

[0063] First, construct the neighborhood covariance matrix C. For any point Pi, the neighborhood covariance matrix can be calculated according to the following formula:

[0064]

[0065] Wherein, K is the number of nearest neighbors of point Pi. In this embodiment, K is selected as 40. After the neighborhood covariance matrix is ​​constructed, eigenvalue decomposition is performed based on the neighborhood covariance matrix to calculate the eigenvalues ​​λ1, λ2, λ3 of the neighborhood point set and the corresponding eigenvectors ξ1, ξ2, ξ3. The curvature of this point can be calculated by the following formula:

[0066]

[0067] Step 2: Extract the normal vector directions of the main joints of the slope based on the first point cloud model;

[0068] Step 3: Determine the location of dangerous rocks on the slope based on the normal vector direction of the main joints and using stereographic projection analysis and normal vector matching methods;

[0069] Step 4: Repeat steps 1 to 3 for the same slope according to the preset time interval to obtain the second point cloud model. After registering the second point cloud model with the first point cloud model and calculating the relative distance, the result of the slope change is obtained.

[0070] For example, such as Figures 3 to 5 As shown, the vegetation removal pretreatment includes:

[0071] The curvature corresponding to each point is calculated based on the characteristic values ​​of each point. The calculation process of curvature V can refer to the curvature calculation method in the above embodiment.

[0072] The RGB values ​​corresponding to the colors of each point are converted into HSV values. RGB to HSV conversion is a conventional technique in existing technology and will not be elaborated here.

[0073] Filter and remove points whose HSV values ​​are within a preset range and whose curvature is greater than a preset value.

[0074] In this embodiment, since vegetation on slopes is often green and protruding, the points belonging to vegetation in the first point cloud model have a large curvature and are located in green areas. Before removing these points, such as Figure 3 As shown, firstly, the preset curvature value is set to T, where T is the 60th percentile of the curvature of all points; the preset range of HSV values ​​is set to R, which is R60 to R160. Then, points with curvature greater than T and HSV values ​​within the range of R60 to R160 are removed. Next, outlier removal methods from statistics are used to calculate the average distance from each point to its neighbors. Assuming the result is a Gaussian distribution whose shape is determined by the mean and standard deviation, points with average distances outside the preset standard range can be defined as outliers and removed from the point cloud data. For example, the number of neighboring points is set to 200, and the standard deviation is set to 0.8. Finally, the DBSCAN algorithm is used for clustering, and categories with fewer than 200 points in each category are removed, completing the vegetation removal operation.

[0075] For example, such as Figures 6 to 8 As shown, the extraction of the main joint normal vector directions of the slope based on the first point cloud model also includes:

[0076] Manually select areas with obvious slope joint characteristics;

[0077] Based on the region growing algorithm, and according to the preset threshold T of the angle between the normal vectors... A and seed point curvature threshold V A Segmenting of slope joint surfaces;

[0078] The RANSAC algorithm is used to fit the plane to each segmented point to obtain the three-dimensional normal vector of each joint surface;

[0079] Convert the three-dimensional normal vectors of each joint surface into unit vectors and make their Z coordinates negative;

[0080] Using the lower spherical stereographic projection, the three-dimensional normal vector is converted into a two-dimensional stereographic projection map;

[0081] Based on the coordinates of two-dimensional projection points on a two-dimensional stereographic projection map, KDE kernel density estimation and meanshift clustering are used to determine the direction of the normal vector of the main joints of the slope.

[0082] For example, the stereographic projection analysis method further includes calculating the dip and dip angle of the main joint surfaces based on the direction of the normal vector of the main joints of the slope, and obtaining the normal vector of the unstable joint surface using the stereographic projection analysis method; wherein the normal vector of the unstable joint surface includes possible aspects of the failure types such as slope plane sliding, wedge sliding, and collapse; specifically, firstly, the Z coordinate value of the direction of the normal vector of the main joint of the slope is corrected to a negative value; specifically, if the Z coordinate is positive, then the X, Y, and Z coordinates are all multiplied by -1; then the angle between the normal vector of each point after correction and the normal vector of the unstable joint surface is calculated; finally, it is determined whether the angle is less than the angle threshold T. A If the rock is identified as a dangerous rock, the DBSCAN clustering algorithm is used to determine the final location of the dangerous rock.

[0083] The following is a simple example of stereographic projection analysis in this embodiment:

[0084] In this embodiment, the dip direction (dipdir), the tilt angle (d), and the angle between the normal vectors can be calculated as follows:

[0085]

[0086] d = arccos(cosγ);

[0087] If cosα>0,

[0088] dipdir = arccos(cosβ);

[0089] If cosα < 0,

[0090] dipdir = 360 - arccos(cosβ);

[0091] Where, N x N y N z , where Z is the normal vector of the main joint of the slope with negative Z-coordinate. As the normal phase modulus, in this embodiment, although the dip and tilt angle may differ from the actual direction, since only the relative relationship between structural surfaces is considered in the dynamic analysis, this embodiment can ultimately determine the direction of the joint surfaces that may lead to slope instability based on the dynamic analysis of stereographic projection.

[0092] For example, the point cloud model registration includes the following steps:

[0093] Manually select feature point pairs for initial registration;

[0094] The ICP algorithm is used to further register the data based on the initial registration.

[0095] like Figure 2 As shown, this embodiment is implemented based on point cloud 2 obtained after a period of time following the identification of dangerous rocks in point cloud 1 in the previous embodiment. The manual coarse registration specifically involves manually identifying the matching relationships between these markers in point cloud 1 and point cloud 2 based on selected photogrammetric control points or road markers, such as tunnel outlines, building outlines, and road inflection points, to obtain initial transformation values. Subsequently, based on the manual registration and using ICP, further optimization is performed to obtain more accurate transformations.

[0096] For example, obtaining the results of slope change specifically includes: calculating the relative distance of each point in the registered second point cloud model to the nearest point in the first point cloud model and assigning a new color to the registered point cloud model to obtain the third point cloud model; using the Quickhull algorithm to calculate the coordinates of the convex hull vertex of the dangerous rock location in the second point cloud model; extracting the part of the convex hull vertex coordinates corresponding to the point cloud region changes in the third point cloud model, and finally obtaining the results of slope change.

[0097] In summary, the method for identifying and monitoring dangerous rocks on slopes based on UAV inspection provided in this disclosure uses UAVs as a carrier, based on oblique photography and point cloud processing technology, and combined with geological engineering knowledge, to achieve rapid identification of dangerous rocks on slopes and subsequent monitoring.

[0098] Finally, it should be noted that the above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for identifying and monitoring dangerous rockfalls on slopes based on unmanned aerial vehicle (UAV) inspection, characterized in that, Includes the following steps: Step 1: Use drones to inspect and collect slope images to obtain the first point cloud model, and perform vegetation removal preprocessing on the first point cloud model; Step 2: Extract the normal vector directions of the main joints of the slope based on the first point cloud model, including: Manually select areas with obvious slope joint characteristics; Based on the region growing algorithm, the slope joint surface is segmented according to the preset normal vector angle threshold and curvature threshold; The RANSAC algorithm is used to fit the plane to each segmented point to obtain the three-dimensional normal vector of each joint surface; Convert the three-dimensional normal vectors of each joint surface into unit vectors and make their Z coordinates negative; Using the lower spherical stereographic projection, the three-dimensional normal vector is converted into a two-dimensional stereographic projection map; KDE kernel density estimation and meanshift clustering are performed based on the coordinates of two-dimensional projection points on a two-dimensional stereographic projection map to determine the direction of the normal vector of the main joint of the slope. Step 3: Determine the location of unstable rock formations on the slope based on the normal vector direction of the main joints and using stereographic projection analysis and normal vector matching methods, including: The Z-coordinate value of the normal vector direction of the main joint of the slope is corrected to a negative value; Calculate the angle between the normal vector at each point after correction and the normal vector of the unstable joint surface; If the included angle is less than the threshold, it is judged as a dangerous rock, and then the DBSCAN clustering algorithm is used to determine the final location of the dangerous rock. Step 4: Repeat steps 1 to 3 for the same slope according to a preset time interval to obtain a second point cloud model. Register the second point cloud model with the first point cloud model and calculate the relative distance to obtain the slope change results, including: Manually select feature point pairs for initial registration; The ICP algorithm is used to further register the data based on the initial registration. Calculate the relative distance of each point in the registered second point cloud model to the nearest point in the first point cloud model and assign a new color to the registered point cloud model to obtain the third point cloud model; The Quickhull algorithm was used to calculate the coordinates of the convex hull vertices of the dangerous rock locations in the second point cloud model. The portion of the convex hull vertex coordinates that corresponds to the third point cloud model is the part of the point cloud region that changes.

2. The method for identifying and monitoring dangerous rockfalls on slopes according to claim 1, characterized in that, The process of acquiring the first point cloud model by collecting slope image information through drone inspection specifically includes: Select a drone equipped with differential GNSS to fly along a preset path, and the drone will acquire information from two adjacent images of an overlapping area with a set overlap ratio. A sparse point cloud model is obtained by aerial triangulation based on image information collected by UAV, and a 3D mesh model of slope with texture information is obtained based on 3D reconstruction. The first point cloud model was obtained by using CloudCompare software and sampling based on density.

3. The method for identifying and monitoring dangerous rockfalls on slopes according to claim 1, characterized in that, Before performing vegetation removal preprocessing on the first point cloud model, the following also includes: Establish a Kd-tree for the first point cloud model, and calculate the K nearest neighbor set of each point in the first point cloud model based on the Kd-tree; Construct the covariance matrix of the neighborhood point set and perform eigenvalue decomposition; The PCA principal component analysis method is used to calculate the eigenvalues ​​and corresponding eigenvectors of each point in the neighborhood point set.

4. The method for identifying and monitoring dangerous rockfalls on slopes according to claim 3, characterized in that, The vegetation removal pretreatment specifically includes: Calculate the curvature corresponding to each point based on the characteristic values ​​of each point; Convert the RGB values ​​corresponding to the colors of each point to HSV values; Filter and remove points whose HSV values ​​are within a preset range and whose curvature is greater than a preset value.

5. The method for identifying and monitoring dangerous rockfalls on slopes according to claim 1, characterized in that, The stereographic projection analysis method also includes calculating the dip and dip angle of the main joint surfaces based on the direction of the normal vector of the main joint of the slope, and obtaining the normal vector of the unstable joint surface using the stereographic projection analysis method.

Images