A dangerous rock slope vegetation filtering method based on multi-dimensional features

Abstract

The present application provides a kind of dangerous rock slope vegetation filtering method based on multi-dimensional features, comprising: step 1, obtaining the initial three-dimensional model of dangerous rock slope;Step 2, generating a reference surface based on the initial three-dimensional model;Step 3, preliminary filtering the vegetation of dangerous rock slope;Step 4, secondary filtering the vegetation of dangerous rock slope.The present application uses the hierarchical filtering strategy of "preliminary filtering and secondary filtering" to filter the plants of dangerous rock slope, which greatly improves the filtering accuracy, effectively avoids the problems of "over stripping" and "under stripping" in traditional methods.

Description

Technical Field

[0001] This invention relates to a vegetation filtering method for dangerous rock slopes based on multidimensional features, belonging to the field of information processing technology. Background Technology

[0002] In the raw point cloud data acquired by UAV photogrammetry, vegetation cover is one of the key interference factors affecting the accuracy of engineering-scale terrain modeling and the stability of numerical calculations. The vegetation point cloud of the dangerous rock slope and the real surface point cloud are highly overlapping in spatial location. If they are not effectively distinguished, it is easy to exaggerate the terrain undulations and offset the local geometric boundaries, which in turn leads to systematic misjudgments in the simulation of the movement path and collision behavior of the dangerous rock.

[0003] Existing technologies for vegetation stripping from point clouds on slopes mainly fall into two categories: manual trimming and simple automatic filtering. Manual trimming relies on operator experience for interactive removal, which is not only inefficient but also struggles to guarantee consistency and repeatability of results under large-scale, high-density point cloud conditions, contradicting the standardization requirements of digital engineering. Automatic filtering methods based on height thresholds or simple morphology rely solely on single features such as elevation difference and slope, making it difficult to distinguish between low-lying vegetation and actual rock surfaces on slopes. This can easily lead to either over- or under-stripping of the terrain, and the filtering accuracy cannot meet the refined modeling needs of dangerous rock slopes, let alone adapt to the stringent requirements of BIM parametric modeling for data accuracy and format. These inherent defects result in traditional methods consistently facing bottlenecks such as "difficulty in balancing efficiency and accuracy, poor digital adaptability, weak versatility, and low repeatability," failing to meet the digital application needs of complex mountain engineering scenarios. Summary of the Invention

[0004] The purpose of this invention is to provide a vegetation filtering method for dangerous rock slopes based on multi-dimensional features, so as to solve the problems existing in the prior art.

[0005] To address the aforementioned technical problems, this invention provides a vegetation filtering method for dangerous rock slopes based on multi-dimensional features, comprising:

[0006] Step 1: Obtain the initial 3D model of the unstable rock slope;

[0007] Step 2: Generate a reference surface based on the initial 3D model;

[0008] Step 3: Preliminary Filtering of Vegetation on Dangerous Rock Slopes: For each vertex p in the initial 3D model, its Euclidean distance to the reference surface, Gaussian curvature, mean curvature, and color features are fused to obtain its multidimensional feature set. The multidimensional feature set of each vertex in the initial 3D model is then input into the GMM clustering model to obtain the clustering result, i.e., the preliminary filtered point cloud of vegetation on the dangerous rock slope. The preliminary filtered point cloud of vegetation on the dangerous rock slope is then reconstructed into a 3D mesh to obtain the preliminary filtered mesh M of vegetation on the dangerous rock slope. pre ;

[0009] Step 4: Secondary Filtering of Vegetation on Dangerous Rock Slopes: Randomly select point p0, transform it along the X and Z axes to obtain point set A, and then transform it along the Y axis to obtain point set B. Connect corresponding points with the same X and Z coordinates in A and B to generate a set of straight lines parallel to the Y axis. Stretch the set of straight lines along the X axis at a fixed step length to obtain the auxiliary detection grid M. aux1 Based on the auxiliary detection grid and the preliminary filtering grid M pre Spatial intersection operation to obtain the interference grid M missed by clustering noise Multiple feature points are uniformly extracted from the boundary of the initial 3D model and connected sequentially to form a closed boundary curve. The closed boundary curve is then mapped onto the reference plane to obtain the boundary curve C of the dangerous rock slope. rock0 Extract the initial filter mesh M pre The Y and Z coordinates of each vertex v are used to reconstruct a two-dimensional point set. Points located on the boundary curve C of the dangerous rock slope within the reconstructed two-dimensional point set are then selected. rock0 Within the range and not belonging to the interference grid M noise The vertices form a valid vertex set V. in Based on the effective vertex set V in Its normal information n(v), and reconstruct the secondary filtered point cloud P of vegetation on the dangerous rock slope. final , Where λ is the scaling factor in the normal direction, the secondary filtering point cloud P of the vegetation on the dangerous rock slope is used. final Reconstructing the 3D mesh yields the secondary filtered mesh M for vegetation on the dangerous rock slope. final .

[0010] In one specific implementation, the auxiliary detection grid and the preliminary filtering grid M pre Spatial intersection operation to obtain the interference grid M missed by clustering noise Specifically, this involves: adjusting the auxiliary detection grid M... aux1 With the initial filter grid M pre Perform a spatial intersection operation to obtain the first set of intersection lines C, where C=M. pre ∩M aux1 By filtering out the non-closed curves in the first set of intersections C, we obtain the second set of intersections C. closed ; Set the second intersection line set C closed Transformed into a two-dimensional curved surface Sc Then, the two-dimensional surface S c A 3D mesh M is generated by stretching along the Z-axis. aux2 Finally, the three-dimensional mesh M aux2 The interference mesh M missed by clustering is obtained by translating and scaling along the Z-axis. noise .

[0011] In one specific implementation, step 2 specifically includes:

[0012] Step 2.1: Generate a 3D point set D: Randomly select point p1 and perform a spatial composite transformation to generate a 3D point set D. The spatial composite transformation is as follows:

[0013] The first step is to perform a translation transformation along the X-axis on the initial point p1 to generate point p2. The specific formula is: p2=p1+a·i, where a is the overall translation amount along the X-axis and i is the unit basis vector along the X-axis.

[0014] The second step involves performing an array transformation along the Y-axis on point p2 to generate a point set C, where the o-th point p in point set C is... C，o The specific transformation formula is: p C，o =p2+o·n·j(o=0,1,…,L-1), where n is the step distance between adjacent points in the Y-axis direction, j is the unit basis vector of the Y-axis, and L is the total number of Y-axis arrays;

[0015] The third step is to perform an array transformation on each point in point set C along the Z-axis to generate point set D. The (o, k)th point p in point set D... D，o，k The specific formula is: p D，o，k =p C,o +k·m·K (k=0,1,…,K-1), where m is the step distance between adjacent points in the Z-axis direction, k is the unit basis vector of the Z-axis, and K is the total number of Z-axis arrays;

[0016] Step 2.2: Obtain the projection point set Q base Remove null and invalid values ​​from the 3D point set D to obtain the valid point set. Project each point of the valid point set along the X-axis onto the initial 3D model to obtain the projected point set Q. base ;

[0017] Step 2.3: Project the point set Q base Generate the reference surface S using NURBS interpolation. base .

[0018] In one specific implementation, the Gaussian curvature and mean curvature of each vertex p on the initial 3D model are obtained according to the following steps: based on the spatial position of the initial 3D model and the reference surface, the initial 3D model located on the reference surface S... base The outer vertices are divided into upper side point sets. The initial 3D model is located on the reference surface S. base The inner vertices are divided into the lower set of vertices. Obtain the maximum principal curvature k1 and minimum principal curvature k2 of each vertex p of the initial 3D model to obtain the Gaussian curvature K of each vertex of the initial 3D model. G (p) and mean curvature H(p), K G (p)=k1·k2, H(p)=(k1+k2) / 2.

[0019] In one specific implementation, the Euclidean distance d(p) from each vertex p on the initial 3D model to the reference surface is calculated according to the following formula: ,in Let vertex p lie on the reference surface S base The vertical projection point on the surface.

[0020] In one specific implementation, the color feature c(p) of each vertex p on the initial three-dimensional model is obtained by the following formula: c(p)=(R(p),G(p),B(p)), where R(p), G(p), and B(p) are the red, green, and blue color components of vertex p, respectively.

[0021] In one specific implementation, the multidimensional feature set f(p) of each vertex p on the initial 3D model is obtained by the following formula: .

[0022] In one specific implementation, the initial three-dimensional model of the dangerous rock slope is generated using point cloud data collected by a drone.

[0023] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0024] 1. This invention integrates three types of features—curvature, spatial distance, and RGB color—to obtain a multi-dimensional feature set. Based on this multi-dimensional feature set, vegetation on dangerous rock slopes is initially filtered. This fully utilizes the essential differences between vegetation and dangerous rocks in terms of geometric and visual attributes, overcoming the limitations of traditional clustering methods that rely on single features. This invention then performs secondary filtering of vegetation on dangerous rock slopes based on geometric intersection and reconstruction of effective vertices. This invention employs a hierarchical filtering strategy of "preliminary filtering and secondary filtering" to filter vegetation on dangerous rock slopes, significantly improving the filtering accuracy and effectively avoiding the problems of "over-stripping" and "under-stripping" in traditional methods.

[0025] 2. This invention can automatically execute all steps from benchmark surface construction, feature extraction, initial filtering to secondary filtering without manual intervention. Moreover, the results of each stage can be visualized and verified in real time on the BIM platform, which greatly improves processing efficiency and result reliability while ensuring the consistency and repeatability of the processing results.

[0026] 3. This invention fully adapts to the needs of digital engineering development. The final filtered point cloud can be directly imported into the BIM platform for parametric modeling without additional format conversion. This solves the problem of poor connection between existing methods and BIM digital collaborative processes, greatly improves the efficiency of digital modeling and collaborative analysis of slope engineering, and can effectively promote the digital and intelligent development of slope engineering. Attached Figure Description

[0027] Figure 1 This is a flowchart of a vegetation filtering method for dangerous rock slopes based on multidimensional features.

[0028] Figure 2 This is a schematic diagram of the reference surface in an embodiment of the present invention.

[0029] Figure 3 This is a schematic diagram of the preliminary filtering point cloud of vegetation on a dangerous rock slope according to an embodiment of the present invention.

[0030] Figure 4 This is a schematic diagram of a group of straight lines according to an embodiment of the present invention.

[0031] Figure 5 This is a schematic diagram of the second set of intersection lines in an embodiment of the present invention.

[0032] Figure 6 This is a schematic diagram of the interference grid in an embodiment of the present invention.

[0033] Figure 7 This is a schematic diagram of the boundary curve of a dangerous rock slope according to an embodiment of the present invention.

[0034] Figure 8 Schematic diagram of secondary filtration point cloud of vegetation on dangerous rock slopes according to an embodiment of the present invention.

[0035] Figure 9 This is a comparison diagram of the results of primary and secondary filtration of vegetation on dangerous rock slopes according to an embodiment of the present invention, wherein (a) is the primary filtration grid of vegetation on dangerous rock slopes, and (b) is the secondary filtration grid of vegetation on dangerous rock slopes. Detailed Implementation

[0036] The present invention will now be described in detail with reference to the embodiments and accompanying drawings. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0037] refer to Figure 1 A vegetation filtering method for dangerous rock slopes based on multidimensional features, comprising:

[0038] Step 1: Obtain the initial three-dimensional model of the dangerous rock slope. Specifically, in this embodiment, the initial three-dimensional model is generated by point cloud data collected by a UAV for a certain dangerous rock slope.

[0039] Step 2: Generate the reference surface S based on the initial 3D model base ,refer to Figure 2 The green area in the diagram represents the reference surface. Specifically, step 2 includes the following steps:

[0040] Step 2.1: Generate a 3D point set D: Randomly select point p1 and perform a spatial composite transformation to generate a 3D point set D. The spatial composite transformation is as follows:

[0041] The first step is to perform a translation transformation along the X-axis on the initial point p1 to generate point p2. The specific formula is: p2=p1+a·i, where a is the overall translation amount along the X-axis and i is the unit basis vector along the X-axis.

[0042] The second step involves performing an array transformation along the Y-axis on point p2 to generate a point set C, where the o-th point p in point set C is... C，o The specific transformation formula is: p C，o =p2+o·n·j(o=0,1,…,L-1), where n is the step distance between adjacent points in the Y-axis direction, j is the unit basis vector of the Y-axis, and L is the total number of Y-axis arrays;

[0043] The third step is to perform an array transformation on each point in point set C along the Z-axis to generate point set D. The (o, k)th point p in point set D... D，o，k The specific formula is: p D，o，k =p C,o +k·m·K (k=0,1,…,K-1), where m is the step distance between adjacent points in the Z-axis direction, k is the unit basis vector of the Z-axis, and K is the total number of Z-axis arrays;

[0044] Step 2.2: Obtain the projection point set Q base Remove null and invalid values ​​from the 3D point set D to obtain the valid point set. Project each point of the valid point set along the X-axis onto the initial 3D model to obtain the projected point set Q. base ;

[0045] Step 2.3: Project the point set Q base Generate the reference surface S using NURBS interpolation. base .

[0046] Step 3: Preliminary Filtering of Vegetation on Dangerous Rock Slopes: The Gaussian curvature K of each vertex p on the initial 3D model is... G The multidimensional feature set f(p) of each vertex is obtained by fusing the mean curvature H(p), color feature c(p), and Euclidean distance d(p) to the reference surface. The multidimensional feature set f(p) of each vertex is then input into the GMM clustering model to obtain the clustering result. (Refer to...) Figure 3That is, the preliminary filtered point cloud Ppre of vegetation on the dangerous rock slope is used to reconstruct a three-dimensional mesh from the preliminary filtered point cloud Ppre of vegetation on the dangerous rock slope, resulting in the preliminary filtered mesh M of vegetation on the dangerous rock slope. pre Preliminary vegetation filter grid M for dangerous rock slopes pre refer to Figure 9 (a).

[0047] By integrating Euclidean distance, color features, and curvature features to construct a multi-dimensional feature set, we can make full use of the essential differences between vegetation and dangerous rocks in terms of geometric and visual attributes, thus breaking through the limitations of traditional clustering methods that rely on a single feature.

[0048] Specifically, the Gaussian curvature K of each vertex p in the initial 3D model G (p) and the mean curvature H(p) are obtained according to the following steps: Based on the spatial position of the initial 3D model and the reference surface, the initial 3D model is located on the reference surface S. base The outer vertices are divided into the upper side point set Q. up The initial 3D model is located on the reference surface S. base The inner vertices are divided into the lower set Q. down Obtain the maximum principal curvature k1 and minimum principal curvature k2 of each vertex v of the initial 3D model to obtain the Gaussian curvature K of each vertex of the initial 3D model. G (p) and mean curvature H(p), K G (p)=k1·k2, H(p)=(k1+k2) / 2.

[0049] Specifically, each vertex p on the initial 3D model is connected to the reference surface S. base The Euclidean distance d(p) is calculated using the following formula: ,in Let vertex p lie on the reference surface S base The vertical projection point on the surface.

[0050] Specifically, the color feature c(p) of each vertex p on the initial 3D model is obtained by the following formula: c(p) = (R(p), G(p), B(p)).

[0051] Specifically, the multidimensional feature set f(p) of each vertex p in the initial 3D model is obtained by the following formula: .

[0052] Step 4: Secondary filtration of vegetation on unstable rock slopes: Establish an auxiliary detection grid, based on the auxiliary detection grid M aux1 With the initial filter grid M pre Spatial intersection operation to obtain the interference grid M missed by clustering noiseMultiple feature points are uniformly extracted from the boundary of the initial 3D model and connected sequentially to form a closed boundary curve C. rock , close the boundary curve C rock Mapping to the reference plane, we obtain the boundary curve C of the dangerous rock slope. rock0 ,refer to Figure 7 The green closed curve represents the boundary curve of the dangerous rock slope. In this embodiment, 60 feature points are extracted to generate the boundary curve of the dangerous rock slope; from the initial filtered grid M pre Vertices are extracted to construct a vertex set V. Based on the Y and Z coordinates of any vertex v in vertex set V, a two-dimensional point set is reconstructed. Points located on the boundary curve C of the dangerous rock slope in the reconstructed two-dimensional point set are selected. rock0 Within the range and not belonging to the interference grid M noise The vertices form a valid vertex set V. in Based on the effective vertex set V in Its normal information n(v), and reconstruct the secondary filtered point cloud P of vegetation on the dangerous rock slope. final : Where λ is the scaling factor in the normal direction, the secondary filtering point cloud P of the vegetation on the dangerous rock slope is used. final Reconstructing the 3D mesh yields the secondary filtered mesh M for vegetation on the dangerous rock slope. final Secondary filtering point cloud of vegetation on dangerous rock slope P final refer to Figure 8 Secondary filter grid for vegetation on dangerous rock slopes M final refer to Figure 9 (b).

[0053] The establishment of an auxiliary detection grid, based on the auxiliary detection grid M aux1 With the initial filter grid M pre Spatial intersection operation to obtain the interference grid M missed by clustering noise Specifically, point p0 is randomly selected and spatial transformations along the X and Z axes are performed to generate point set A. A Y-axis transformation is then performed on point set A to generate point set B. Points in point sets A and B with the same X and Z coordinates are then connected sequentially to form a series of spatial line segments parallel to the Y-axis. These line segments together constitute line group L. a，b ,refer to Figure 4 The green lines in the diagram form a group of lines, and the group of lines L... a，b Along the X-axis with step size Stretching generates auxiliary detection mesh M aux1 ; Add auxiliary detection grid M aux1 With the initial filter grid M pre Perform a spatial intersection operation to obtain the first set of intersection lines C, where C=M. pre ∩M aux1 By filtering out the non-closed curves in the first set of intersections C, we obtain the second set of intersections C.closed ,refer to Figure 5 The green closed curve in the diagram represents the second set of intersection lines; let the second set of intersection lines C... closed Transformed into a two-dimensional curved surface S c Then, the two-dimensional surface S c A 3D mesh M is generated by stretching along the Z-axis. aux2 Finally, the three-dimensional mesh M aux2 The interference mesh M missed by clustering is obtained by translating and scaling along the Z-axis. noise ,refer to Figure 6 The red part in the image represents the interference grid.

[0054] This invention employs a layered filtration strategy of "preliminary filtration and secondary filtration" to filter vegetation on dangerous rock slopes, significantly improving filtration accuracy. Figure 9 As can be seen, after secondary filtration, the vegetation remaining from the initial filtration of the dangerous rock slope has been removed.

[0055] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions and substitutions can be made without departing from the inventive concept, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A vegetation filtering method for dangerous rock slopes based on multidimensional features, characterized in that, include: Step 1: Obtain the initial 3D model of the unstable rock slope; Step 2: Generate a reference surface based on the initial 3D model; Step 3, preliminary filtering of vegetation on the dangerous rock slope: for each vertex p on the initial three-dimensional model, fuse its Euclidean distance to the reference surface, Gaussian curvature, mean curvature, color features to obtain its multi-dimensional feature set, input the multi-dimensional feature set of each vertex on the initial three-dimensional model into the GMM clustering model to obtain the clustering result, that is, the dangerous rock slope vegetation preliminary filtering point cloud, reconstruct a three-dimensional grid from the dangerous rock slope vegetation preliminary filtering point cloud to obtain the dangerous rock slope vegetation preliminary filtering grid M pre ; Step 4: Secondary Filtering of Vegetation on Dangerous Rock Slopes: Randomly select point p0, transform it along the X and Z axes to obtain point set A, and then transform it along the Y axis to obtain point set B. Connect corresponding points with the same X and Z coordinates in A and B to generate a set of straight lines parallel to the Y axis. Stretch the set of straight lines along the X axis at a fixed step length to obtain the auxiliary detection grid M. aux1 Based on the auxiliary detection grid and the preliminary filtering grid M pre Spatial intersection operation to obtain the interference grid M missed by clustering noise Multiple feature points are uniformly extracted from the boundary of the initial 3D model and connected sequentially to form a closed boundary curve. The closed boundary curve is then mapped onto the reference plane to obtain the boundary curve C of the dangerous rock slope. rock0 Extract the initial filter mesh M pre The Y and Z coordinates of each vertex v are used to reconstruct a two-dimensional point set. Points located on the boundary curve C of the dangerous rock slope within the reconstructed two-dimensional point set are then selected. rock0 Within the range and not belonging to the interference grid M noise The vertices form a valid vertex set V. in Based on the effective vertex set V in Its normal information n(v), and reconstruct the secondary filtered point cloud P of vegetation on the dangerous rock slope. final , Where λ is the scaling factor in the normal direction, the secondary filtering point cloud P of the vegetation on the dangerous rock slope is used. final Reconstructing the 3D mesh yields the secondary filtered mesh M for vegetation on the dangerous rock slope. final .

2. The vegetation filtering method for dangerous rock slopes based on multi-dimensional features as described in claim 1, characterized in that, The auxiliary detection grid and the preliminary filtering grid M pre Spatial intersection operation to obtain the interference grid M missed by clustering noise Specifically, this involves: adjusting the auxiliary detection grid M... aux1 With the initial filter grid M pre Perform a spatial intersection operation to obtain the first set of intersection lines C, where C=M. pre ∩M aux1 Filter out the non-closed curves in the first set of intersections C to obtain the second set of intersections C. closed ; Set the second intersection line set C closed Transformed into a two-dimensional curved surface S c Then, the two-dimensional surface S c A 3D mesh M is generated by stretching along the Z-axis. aux2 Finally, the 3D mesh M aux2 The interference mesh M missed by clustering is obtained by translating and scaling along the Z-axis. noise .

3. The vegetation filtering method for dangerous rock slopes based on multi-dimensional features as described in claim 2, characterized in that, Step 2 specifically involves: Step 2.1: Generate a 3D point set D: Randomly select point p1 and perform a spatial composite transformation to generate a 3D point set D. The spatial composite transformation is as follows: The first step is to perform a translation transformation along the X-axis on the initial point p1 to generate point p2. The specific formula is: p2=p1+a·i, where a is the overall translation amount along the X-axis and i is the unit basis vector along the X-axis. The second step involves performing an array transformation along the Y-axis on point p2 to generate a point set C, where the o-th point p in point set C is... C，o The specific transformation formula is: p C，o =p2+o·n·j(o=0,1,…,L-1), where n is the step distance between adjacent points in the Y-axis direction, j is the unit basis vector of the Y-axis, and L is the total number of Y-axis arrays; The third step is to perform an array transformation on each point in point set C along the Z-axis to generate point set D. The (o, k)th point p in point set D... D，o，k The specific formula is: p D，o，k =p C,o +k·m·K (k=0,1,…,K-1), where m is the step distance between adjacent points in the Z-axis direction, k is the unit basis vector of the Z-axis, and K is the total number of Z-axis arrays; Step 2.2: Obtain the projection point set Q base Remove null and invalid values ​​from the 3D point set D to obtain the valid point set. Project each point of the valid point set along the X-axis onto the initial 3D model to obtain the projected point set Q. base ; Step 2.3: Project the point set Q base Generate the reference surface S using NURBS interpolation. base .

4. The vegetation filtering method for dangerous rock slopes based on multi-dimensional features as described in claim 3, characterized in that, The Gaussian curvature and mean curvature of each vertex p on the initial 3D model are obtained according to the following steps: Based on the spatial position of the initial 3D model and the reference surface, the initial 3D model located on the reference surface S... base The outer vertices are divided into the upper side point set Q. up The initial 3D model is located on the reference surface S. base The inner vertices are divided into the lower set Q. down Obtain the maximum principal curvature k1 and minimum principal curvature k2 of each vertex p of the initial 3D model to obtain the Gaussian curvature K of each vertex of the initial 3D model. G (p) and mean curvature H(p), K G (p)=k1·k2, H(p)=(k1+k2) / 2.

5. The vegetation filtering method for dangerous rock slopes based on multi-dimensional features as described in claim 4, characterized in that, The Euclidean distance d(p) from each vertex p on the initial 3D model to the reference surface is calculated according to the following formula: ,in Let vertex p lie on the reference surface S base The vertical projection point on the surface.

6. The vegetation filtering method for dangerous rock slopes based on multi-dimensional features as described in claim 5, characterized in that, The color feature c(p) of each vertex p on the initial 3D model is obtained by the following formula: c(p)=(R(p),G(p),B(p)), where R(p), G(p), and B(p) are the red, green, and blue color components of vertex p, respectively.

7. The vegetation filtering method for dangerous rock slopes based on multi-dimensional features as described in claim 6, characterized in that, The multidimensional feature set f(p) of each vertex p in the initial 3D model is obtained by the following formula: .

8. The vegetation filtering method for dangerous rock slopes based on multi-dimensional features as described in claim 7, characterized in that, The initial three-dimensional model of the dangerous rock slope was generated using point cloud data collected by a drone.

Images