Fine extraction method for point cloud slope step line of strip mine fused with multiple parameters

Abstract

The invention discloses a multi-parameter fused strip mine point cloud slope step line refined extraction method. The method comprises the following steps: S1, collecting point cloud data through an unmanned aerial vehicle, and downsampling and simplifying the point cloud data according to a voxel grid; s2, filtering out non-ground points in the point cloud data by adopting a point cloud filtering algorithm; s3, fitting the point cloud data through a least square method according to a neighborhood point set to obtain a normal vector of a local plane, and extracting potential step feature points through principal component analysis; s4, calculating elevation gradients of topographic points of the point cloud data in neighborhoods of all the potential step feature points, setting an elevation gradient size threshold value, and screening the potential step feature points greater than the elevation gradient size threshold value as step feature points; and S5, sequentially fitting all step feature points in the point cloud data by using a moving least square method to generate a continuous and smooth step line. According to the method, the step line can be efficiently and accurately extracted, and important technical support is provided for intelligent mining and accurate management of the strip mine.

Description

Technical Field

[0001] The present invention relates to the field of open-pit mine UAV mapping and identification, and in particular to a method for extracting step lines from open-pit mine slopes by fusing multi-parameter point clouds. Background Art

[0002] In the process of mining, open-pit mines are generally mined layer by layer from top to bottom, and the minerals and rocks in the mining area are divided into horizontal layers with a certain thickness. The layered working surface is called a step. Step line information is crucial for production activities such as mining design, block division, and blasting design. It is also the basis for automated and intelligent management of open-pit mines. However, traditional step line extraction mostly relies on manual single-point measurement, such as using RTK to collect step line coordinate points. This method not only has a large workload and high labor intensity, but also has a long data collection cycle, making it difficult to ensure measurement accuracy and timeliness. With the development of drone mapping technology, the current research direction is to combine manual vision and drone interpretation or graphic recognition to draw step lines, but it still relies on manual drawing for a long time, and efficiency and timeliness cannot be effectively improved. Summary of the invention

[0003] The purpose of the present invention is to solve the technical problems pointed out by the background technology, and to provide a method for fine-tuning the step line extraction of open-pit mine point cloud slope by integrating multi-parameters. By dividing the voxel grid and combining the voxel grid downsampling and cloth simulation filtering, not only the original features of the point cloud are retained while the data volume is reduced, but also non-ground points are filtered out, effectively reducing their interference with feature point recognition; the present invention also integrates the two parameters of the normal vector angle and the elevation gradient to further improve the recognition accuracy of feature points for the complex terrain of the open-pit mine, and finally, the continuous and complete step line is fitted through the improved moving least squares, which provides strong support for the automated mining and intelligent management of open-pit mines.

[0004] The purpose of the present invention is achieved through the following technical solutions:

[0005] A method for extracting step lines of open-pit mine slopes by integrating multi-parameter point clouds, the method comprising:

[0006] S1. Obtain point cloud data of the research mining area by collecting data with a drone using a set three-dimensional coordinate system, divide the point cloud space of the point cloud data into voxel grids, and simplify the point cloud data by downsampling according to the voxel grids;

[0007] S2, using point cloud filtering algorithm to filter out non-ground points in point cloud data;

[0008] S3, fitting the point cloud data into a local plane by the least square method according to the neighborhood point set, obtaining the covariance matrix and terrain feature vector of the local plane by principal component analysis, extracting the terrain points whose mean angle between the terrain feature vector and the adjacent terrain feature vector is greater than the angle threshold as potential step feature points;

[0009] S4. Calculate the elevation gradient of the terrain points in the neighborhood of all potential step feature points of the point cloud data. The elevation gradient expression is: in Indicates the elevation gradient of terrain point j in the coordinate system, d jx represents the first-order partial derivative of the terrain point j in the x-axis direction in the coordinate system, d jy Represents the first-order partial derivative of terrain point j in the y-axis direction in the coordinate system;

[0010] Set the elevation gradient threshold and select potential step feature points larger than the elevation gradient threshold as step feature points;

[0011] S5. Use the moving least squares method to fit all the step feature points in the point cloud data in sequence to generate continuous and smooth step lines.

[0012] In order to better realize the present invention, the UAV collects point cloud data of the research mining area according to the set three-dimensional coordinate system and flight planning; the voxel grid in the point cloud space of the point cloud data is densely divided, and the down-sampling simplified point cloud data processing method is as follows: the voxel grid is a three-dimensional grid, the point cloud data in the voxel grid is position-encoded, the point cloud data is mapped to the voxel grid unit, the grid is down-sampled in the voxel grid and the representative point of the voxel grid is determined, and the representative points of all voxel grids are aggregated into simplified point cloud data.

[0013] Preferably, the point cloud filtering algorithm adopts CSF filtering, and the method is as follows: the point cloud data is inverted according to the elevation, the cloth grid is initialized at the highest point position, the cloth particles and the point cloud are projected to the same horizontal plane, the point cloud closest to each particle is found as the corresponding point, and the elevation of the corresponding point is defined as the limit height; for all cloth particles, first set them to the "movable" state, then calculate their position affected by gravity and internal forces, and use the position as the current height; compare the current height with the limit height in each iteration, if the former is less than the latter, move the particle back to the position of the corresponding point and set the particle state to "immovable", otherwise it is set to "movable"; when all particles exceed the specified maximum number of iterations, the cloth simulation process is terminated; a particle scene approximating the real terrain point cloud is obtained through cloth simulation, and the distance between the original point cloud and the particle is obtained by using the cloud distance calculation method, and a distance threshold is set, and points less than the distance threshold are classified as terrain points, and the remaining points are non-ground points.

[0014] Preferably, the neighborhood point set of the point cloud data takes the terrain point P as the target and uses the KD tree structure to find K neighborhood points as the neighborhood point set of the terrain point P. The neighborhood point set of the terrain point P is fitted into a local plane M using the least squares method. The centroid P of the local plane M is 0 and the normal vector The local plane is analyzed by principal component analysis to obtain the eigenvalues ​​arranged from small to large, and the eigenvector with the smallest eigenvalue is selected as the terrain eigenvector.

[0015] Preferably, the potential step feature point is the junction between the slope surface and the top of the slope, and the slope surface and the bottom of the slope. The method for obtaining the potential step feature point is as follows: select a terrain feature vector, extract the terrain feature vectors of its neighborhood, calculate the average value of the angle between the selected terrain feature vector and the terrain feature vectors of the neighborhood, if the average value of the angle is greater than the set angle threshold, then the terrain point corresponding to the selected terrain feature vector is determined to be a potential step feature point, and all terrain feature vectors are traversed in turn to obtain all potential step feature points.

[0016] Preferably, the step feature point screening utilizes approximate differential operation and adopts a four-directional Sobel operator template to perform traversal convolution operation on potential step feature points and point cloud data. The elevation gradient size expression of the potential step feature point is: grad(P) represents the elevation gradient of the terrain point or potential step feature point P, F(P) represents the data after the terrain point or potential step feature point P is expressed by a two-dimensional discrete function, and f n (P) represents the data obtained by the Sobel operator template in four different directions, and the step feature points are screened based on the size of the elevation gradient.

[0017] Preferably, step S5 further includes the following method: using the fitting surface method to perform quadratic surface fitting on the step feature points and their neighborhoods in the point cloud data to obtain a slope angle, and the slope angle expression is as follows:

[0018] S x Indicates the slope of the step feature point in the x direction, S y It represents the slope of the step feature point in the y direction, θ represents the slope angle of the step feature point, and the step feature point with a slope angle less than the slope angle threshold is excluded; after exclusion, the step line fitting process is performed.

[0019] Preferably, the step line generation method includes: expressing the area where the target step feature point x is located by a target function polynomial, f(x)=p T (x)α(x),α(x)=[α 1 (x), α 2 (x), …, α m (x)] T , p T(x) = [p 1 (x), p 2 (x), …, p m (x)], where p m (x) represents the basis function of the m-order polynomial space in the region, α m (x) represents the coefficient vector of the m-order polynomial space in the region, defines the weighted error sum J of the fitting function at the target step feature point x in the moving least squares method, and fits to generate a continuous and smooth step line.

[0020] Preferably, the weighted square error sum J is expressed as follows:

[0021]

[0022]

[0023] Where n represents the step feature point x in the area where the target step feature point x is located. i Total,||xx i || represents the target step feature point x and the step feature point x i distance, h represents the smoothing parameter that controls the distance attenuation, Φ represents the cosine value of the angle between the normal vector of the target step feature point and the normal vector of the adjacent step feature point, θ 平滑 is the smoothing parameter that controls the influence of the normal vector, y i Represents the step feature point x i The node value of .

[0024] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0025] (1) The present invention divides the voxel grid and combines voxel grid downsampling and cloth simulation filtering, which not only reduces the data volume while retaining the original features of the point cloud, but also filters out non-ground points, effectively reducing their interference with feature point recognition; the present invention also integrates the two parameters of normal vector angle and elevation gradient to further improve the recognition accuracy of feature points for the complex terrain of open-pit mines, and finally fits a continuous and complete step line through improved moving least squares, providing strong support for the automated mining and intelligent management of open-pit mines.

[0026] (2) The present invention extracts step lines based on point cloud data collected by drones. Compared with traditional GPS measurement or ground laser scanning, the present invention has the advantages of low collection cost, wide coverage, flexible operation, and fast data acquisition speed. By utilizing the high precision and high density of drone point clouds, the problem of decreased step line extraction accuracy caused by data sparsity in traditional methods can be avoided.

[0027] (3) The present invention combines voxel grid downsampling and cloth simulation filtering to pre-process the original point cloud data, which can effectively reduce the amount of data, remove the interference of irrelevant objects such as vegetation and buildings, and reduce the impact of noise and outliers on subsequent feature point extraction.

[0028] (4) The present invention integrates the normal vector angle and elevation gradient dual parameters to extract feature points, which can more accurately identify step feature points; at the same time, the parameters of feature point extraction are dynamically adjusted, which improves the versatility of the algorithm in open-pit mine scenarios and reduces the occurrence of false detection and missed detection.

[0029] (5) The present invention is based on the improved moving least squares method to fit the step line. By introducing a new weight function and combining the distance and normal vector angle changes, the smoothness and fitting accuracy of the step line are significantly improved; the generated step line has good continuity and accuracy under complex terrain conditions.

[0030] (6) The present invention realizes the whole process of refined processing from open-pit mine point cloud data preprocessing, step feature point extraction, and step line fitting. It can extract step lines efficiently and accurately, is suitable for the complex terrain environment of open-pit mines, and provides important technical support for intelligent mining and precise management of open-pit mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 is a flow chart of the method of the present invention;

[0032] Figure 2 It is a schematic diagram of the principle in the embodiment;

[0033] Figure 3 This is a schematic diagram showing the comparison of the first type of voxel grid before and after planarization using an example in the embodiment;

[0034] Figure 4 This is a comparison diagram of the second voxel grid before and after planarization using an example in the embodiment;

[0035] Figure 5 This is a schematic diagram showing the comparison before and after filtering out non-ground points in point cloud data in the first embodiment;

[0036] Figure 6 This is a comparison diagram of the second method of filtering out non-ground points in point cloud data before and after the embodiment;

[0037] Figure 7 This is a schematic diagram of taking an example of cutting out a certain area for preliminary screening of potential step feature points in an embodiment;

[0038] Figure 8 This is a schematic diagram of taking an area for step feature point screening as an example in an embodiment;

[0039] Fig. 9This is a schematic diagram of the step line fitting effect of a certain area taken as an example in the embodiment. DETAILED DESCRIPTION

[0040] The present invention is further described in detail below in conjunction with embodiments:

[0041] Example

[0042] like Figure 1 As shown, a method for extracting step lines of open-pit mine slopes by fusing multi-parameter point clouds is provided, and the method includes:

[0043] S1. The point cloud data of the research mining area is acquired by collecting the drone with a set three-dimensional coordinate system. The drone collects the point cloud data of the research mining area according to the set three-dimensional coordinate system (the drone of the present invention uses the Gauss Kruger projection belt to represent the coordinates, and its three-dimensional coordinate system corresponds to it, where the x-axis and y-axis are the horizontal projection position coordinates, and the z-axis is the elevation coordinate associated with the elevation) and flight planning. For example, P in the point cloud data i The coordinates of (x i ,y i , z i ).

[0044] like Figure 2 As shown, the point cloud space of the point cloud data is divided into voxel grids and the point cloud data is simplified by downsampling according to the voxel grids. The point cloud space of the point cloud data is densely divided into voxel grids, and the point cloud data is mapped to the voxel grid units. The grid is downsampled in the voxel grid and the representative points of the voxel grid are determined. The representative points of all voxel grids are gathered into simplified point cloud data. The downsampling simplified point cloud data processing method is as follows: the voxel grid is a three-dimensional grid, and the point cloud data in the voxel grid is position-encoded. The position encoding of the voxel grid adopts the grid index method. The coordinate (x i ,y i , z i )'s grid index is (I x , I y , I z ), this embodiment takes a three-dimensional grid with a voxel grid size of a cube d×d×d as an example, d is the side length of the voxel grid, and the [] symbol indicates rounding down. In this embodiment, the representative point of the voxel grid is preferably determined by the centroid method. Where P centre represents the centroid of the voxel grid (or center point), N 1 Represents the total number of points in the voxel grid, i represents the number of the point, x i ,y i , z iThey represent the x-axis, y-axis, and z-axis coordinates of point number i, respectively. After downsampling simplifies point cloud data processing, the amount of point cloud data is significantly reduced while retaining the overall geometric shape, and the impact of noise and outliers is also reduced in subsequent processes.

[0045] The present invention takes Jiangjun Gobi Desert No. 2 Mine in Changji Hui Autonomous Prefecture, Xinjiang Uygur Autonomous Region as the research area, and selects DJI Phantom 4RTK UAV to collect data from the mine area. The camera resolution is 0.05m and the scanning area is about 23km 2 The collected point cloud information includes coordinates, intensity, RGB, echo angle, etc. Based on the point cloud data processing library PCL, the voxel downsampling of the point cloud is performed in the C++ environment. The amount of point cloud collected by the drone is about 158.6 million. The voxel grid is set to 0.5×0.5×0.5. The grid index of each point in its voxel is calculated and the point cloud coordinates are mapped to the voxel unit. The representative point of each voxel is determined by the centroid method. Finally, all representative points are combined into a simplified point cloud. This invention takes the General Gobi Desert No. 2 Mine Research Area as an example. Figure 3 , Figure 4 Comparison chart of two intercepted areas before and after downsampling. Figure 3 The middle left picture shows the first cut area before downsampling and the planarization after downsampling. Figure 3 The right figure of (i.e. the schematic diagram of the first intercepted area after downsampling and flattening); Figure 4 The middle left picture shows the second cut area before downsampling and the planarization after downsampling. Figure 4 The right picture of (i.e. the schematic diagram of the second cut area after downsampling and flattening).

[0046] S2. Use a point cloud filtering algorithm to filter out non-ground points in the point cloud data. In some embodiments, the point cloud filtering algorithm uses CSF filtering (CSF filtering on the downsampling of the voxel grid can effectively eliminate the influence of non-ground factors and obtain accurate ground points; the point cloud filtering algorithm is based on the principle of cloth simulation. First, the point cloud data is regarded as terrain, and then the cloth composed of particles is covered on the point cloud and its drooping process is simulated to separate ground points from non-ground points. In the CSF filtering processing of the open-pit mine point cloud, the movement of the cloth particles is restricted to the vertical direction, and collision detection is achieved by comparing the height values ​​of the particles and the point cloud representing the terrain). The method is as follows: the point cloud data is inverted according to the elevation (that is, the z-axis), and at the highest Initialize the cloth grid at the point position (determine the grid resolution according to the scale of the open-pit mine, and the grid resolution is the density of the point cloud after the final filter), project the cloth particles and the point cloud to the same horizontal plane, find the nearest point cloud as the corresponding point for each particle, and define the elevation of the corresponding point as the limit height; for all cloth particles, first set them to the "movable" state, then calculate their position affected by gravity and internal forces, and use this position as the current height; compare the current height with the limit height in each iteration. If the former is less than the latter, move the particle back to the position of the corresponding point and set the particle state to "immovable", otherwise set it to "movable". The above iterative process can be repeated. In some embodiments, the number of repetitions, that is, the number of particle movements, is represented by setting the stiffness. The greater the stiffness, the more rigid the behavior of the cloth. For the large-scale terrain undulations of the open-pit mine terrain, appropriate parameters are selected to adapt to different scenarios. The relationship between stiffness and total particle displacement is as follows: TD represents the total displacement of the particle, r represents the stiffness, and h 1 Indicates the vertical distance between two particles. In the implementation example of the Jiangjun Gobi Desert No. 2 Mine Research Area, the grid resolution is set to 0.5, the stiffness is set to 2, and the distance threshold is set to a fixed value of 0.5m according to the point cloud scale; the point cloud is processed by CSF filtering to obtain the ground points. Figure 5 , Figure 6 Comparison of the two intercepted areas before and after filtering out the point cloud. Figure 5 The middle left picture is a schematic diagram of filtering out the point cloud in the first intercepted area. After filtering by the point cloud filtering algorithm, Figure 5 The right picture of (i.e. the schematic diagram after filtering out the point cloud in the first intercepted area); Figure 6 The middle left picture is a schematic diagram of the second intercepted area before filtering the point cloud. After filtering by the point cloud filtering algorithm, Figure 6 The right picture of (i.e. the schematic diagram after filtering out the point cloud in the second intercepted area).

[0047] When all particles exceed the specified maximum number of iterations, the cloth simulation process terminates. Through cloth simulation, a particle scene that approximates the real terrain point cloud is obtained. The distance between the original point cloud and the particles is obtained using the cloud distance calculation method. A distance threshold is set (this embodiment chooses to set the distance threshold to 0.5m). Points less than the distance threshold are classified as terrain points, and the rest are non-ground points.

[0048] S3. Fit the point cloud data into a local plane by the least squares method according to the neighborhood point set, and obtain the covariance matrix and terrain feature vector of the local plane by principal component analysis. In some embodiments, the neighborhood point set of the point cloud data takes the terrain point P as the target and uses the KD tree structure to find K neighborhood points as the neighborhood point set of the terrain point P. The neighborhood point set of the terrain point P is fitted into a local plane M by the least squares method, and the centroid P of the local plane M is 0 and the normal vector Normal vector satisfy The local plane is analyzed by principal component analysis to obtain the eigenvalues ​​arranged from small to large, and the eigenvector with the smallest eigenvalue is selected as the terrain eigenvector. The expression of the local plane M is as follows: arg min means to find The minimum value, P i is the point numbered i among the K neighboring points, and d is the side length of the cube-shaped voxel grid. 0 The expression can be as follows: Through principal component analysis (PCA), we can find the eigenvalues ​​arranged from small to large (for example, according to λ 1 , 2 , 3 The eigenvectors with the smallest eigenvalue are selected as the terrain eigenvectors (the eigenvector corresponding to the smallest eigenvalue is the normal vector value of the current terrain point P).

[0049] The terrain points whose average angle between the terrain feature vector and the adjacent terrain feature vector is greater than the angle threshold are extracted as potential step feature points. In some embodiments, the potential step feature points are the junctions of the slope surface and the top of the slope, and the slope surface and the bottom of the slope. The method for obtaining the potential step feature points is as follows: select the terrain feature vector of a terrain point, extract the terrain feature vector of its neighborhood, and calculate the average angle between the terrain feature vector of the selected terrain point and the terrain feature vector of the neighborhood (the degree of change in the angle of the normal vector of the point cloud is positively correlated with the deformation of the area where it is located, that is, the greater the change in the angle, the more obvious the deformation of the area). If the average angle is greater than the set angle threshold (in the implementation example of the No. 2 Mine Research Area in the General Gobi Desert, the angle range of the average angle is set to 65° to 70°, and the terrain points corresponding to this angle range are determined as potential step feature points), then the terrain point corresponding to the selected terrain feature vector is determined as a potential step feature point, and all terrain feature vectors are traversed in turn to obtain all potential step feature points; in the implementation example of the No. 2 Mine Research Area in the General Gobi Desert, a certain area is intercepted for initial screening of potential step feature points, such as Figure 7 As shown ( Figure 7 The display includes potential step feature points and point cloud. At this time, the potential step feature points have been obtained).

[0050] S4. Calculate the elevation gradient size of the terrain point in the neighborhood of all potential step feature points of the point cloud data (for open-pit mine point clouds, the greater the elevation gradient change in the neighborhood of a terrain point, the more obvious the edge feature of the point, so the terrain point whose neighborhood elevation gradient change exceeds a specific threshold is defined as a feature point, and the elevation gradient change is also called the elevation gradient size). The elevation gradient size expression is: in Indicates the elevation gradient of terrain point j in the coordinate system, d jx represents the first-order partial derivative of the terrain point j in the x-axis direction in the coordinate system, d jy Represents the first-order partial derivative of the terrain point j in the y-axis direction in the coordinate system. Set the elevation gradient size threshold and filter the potential step feature points that are greater than the elevation gradient size threshold as step feature points.

[0051] In some embodiments, the step feature point screening utilizes approximate differential operation and adopts a four-directional Sobel operator template to perform traversal convolution operation on potential step feature points and point cloud data. The elevation gradient size expression of the potential step feature point is: grad(P) represents the elevation gradient of the terrain point or potential step feature point P, F(P) represents the data after the terrain point or potential step feature point P is expressed by a two-dimensional discrete function, and f n (P) represents the data obtained by the Sobel operator template in four different directions, and the step feature points are screened based on the size of the elevation gradient.

[0052] S5. Use the moving least squares method to fit all the step feature points in the point cloud data in sequence to generate continuous and smooth step lines (in the implementation example of the Jiangjun Gobi Desert No. 2 Mine Research Area, the effect after fitting the step line is as follows Fig. 9 As shown, Fig. 9 The step lines in the point cloud are fitted, and the line segmentation is clear); the present invention performs point-by-point fitting on the slope step feature point set (i.e., all step feature points) in the area to generate continuous and smooth step lines; finally, the fitting of the open-pit mine point cloud slope step lines is completed from local to overall. This embodiment uses the improved moving least squares method (Refined Moving Least Squares, RMLS) to fit the step lines; the moving least squares method is a local weighted least squares fitting method, which is mainly used for smoothing, fitting and interpolation of point cloud data. The core idea is to fit a smooth function or curve through the weighted least squares method of points in the local area, so as to adapt to the local changes of the point cloud and obtain a smooth and continuous fitting result. Step S5 also includes the following method: the step feature points and their neighborhood in the point cloud data are fitted with a quadratic surface using the fitting surface method to obtain the slope angle. The slope angle expression is as follows: S x Indicates the slope of the step feature point in the x direction, S y represents the slope of the step feature point in the y direction, θ represents the slope angle of the step feature point, and the step feature point with a slope angle less than the slope angle threshold (in the implementation example of the Jiangjun Gobi Desert No. 2 Mine Research Area, the slope angle threshold is set to 60°) is excluded; after exclusion (the results after exclusion are shown in Figure 8 ) and then perform step line fitting.

[0053] In some embodiments, the step line generation method includes: expressing the area where the target step feature point x is located by a target function polynomial, f(x)=p T (x)α(x),α(x)=[α 1 (x), α 2 (x), …, α m (x)] T , p T (x) = [p 1 (x), p 2 (x), …, p m (x)], where p m (x) represents the basis function of the m-order polynomial space in the region, α m (x) represents the coefficient vector of the m-order polynomial space in the region, and defines the weighted square error sum J of the fitting function at the target step feature point x in the moving least squares method, and fits to generate a continuous and smooth step line. In some embodiments, the weighted square error sum J is expressed as follows:

[0054]

[0055]

[0056] Where n represents the step feature point x in the area where the target step feature point x is located. i Total,||xx i || represents the target step feature point x and the step feature point x i distance, h represents the smoothing parameter that controls the distance attenuation (when h is small, only the neighboring points close to the target feature point will have an impact), Φ represents the cosine value of the angle between the normal vector of the target step feature point and the normal vector of the neighboring step feature point, θ 平滑 is the smoothing parameter that controls the influence of the normal vector (when θ 平滑 When it is smaller, the difference in the normal vector angle has a more obvious effect on the weight). i Represents the step feature point x i In some embodiments, the present invention can use shape functions And get the fitting function: y represents the total number of node values ​​of the step feature points.

[0057] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for extracting step lines from open-pit mine slopes by integrating multi-parameter point clouds, characterized in that: The methods include: S1. Obtain point cloud data of the research mining area by collecting data with a drone using a set three-dimensional coordinate system, divide the point cloud space of the point cloud data into voxel grids, and simplify the point cloud data by downsampling according to the voxel grids; S2, using point cloud filtering algorithm to filter out non-ground points in point cloud data; S3, fitting the point cloud data into a local plane by the least square method according to the neighborhood point set, obtaining the covariance matrix and terrain feature vector of the local plane by principal component analysis, extracting the terrain points whose mean angle between the terrain feature vector and the adjacent terrain feature vector is greater than the angle threshold as potential step feature points; S4. Calculate the elevation gradient of the terrain points in the neighborhood of all potential step feature points of the point cloud data. The elevation gradient expression is: in Indicates the elevation gradient of terrain point j in the coordinate system, d jx represents the first-order partial derivative of the terrain point j in the x-axis direction in the coordinate system, d jy Represents the first-order partial derivative of terrain point j in the y-axis direction in the coordinate system; Set the elevation gradient threshold and select potential step feature points larger than the elevation gradient threshold as step feature points; S5. Use the moving least squares method to sequentially fit all the step feature points in the point cloud data to generate continuous and smooth step lines.

2. The method for extracting step lines from open-pit mine slopes by fusing multi-parameter point clouds according to claim 1 is characterized in that: The UAV collects point cloud data of the research mining area according to the set three-dimensional coordinate system and flight plan; the voxel grid in the point cloud space of the point cloud data is densely divided, and the down-sampling and simplified point cloud data processing method is as follows: the voxel grid is a three-dimensional grid, the point cloud data in the voxel grid is positionally encoded, the point cloud data is mapped to the voxel grid unit, the grid is down-sampled in the voxel grid and the representative point of the voxel grid is determined, and the representative points of all voxel grids are gathered into simplified point cloud data.

3. The method for extracting step lines from open-pit mine slopes by fusing multi-parameter point clouds according to claim 1 is characterized in that: The point cloud filtering algorithm uses CSF filtering, and the method is as follows: the point cloud data is inverted according to the elevation, the cloth grid is initialized at the highest point position, the cloth particles and the point cloud are projected to the same horizontal plane, and the nearest point cloud is found for each particle as the corresponding point, and the elevation of the corresponding point is defined as the limit height; for all cloth particles, they are first set to the "movable" state, and then their position affected by gravity and internal forces is calculated, and this position is used as the current height; in each iteration, the current height is compared with the limit height. If the former is less than the latter, the particle is moved back to the position of the corresponding point and the particle state is set to "immovable", otherwise it is set to "movable"; when all particles exceed the specified maximum number of iterations, the cloth simulation process is terminated; through cloth simulation, a particle scene that approximates the real terrain point cloud is obtained, and the distance between the original point cloud and the particle is obtained by using the cloud distance calculation method. The distance threshold is set, and the points less than the distance threshold are classified as terrain points, and the rest are non-ground points.

4. The method for extracting step lines from open-pit mine slopes by fusing multi-parameter point clouds according to claim 1 is characterized in that: The neighborhood point set of point cloud data takes terrain point P as the target and uses the KD tree structure to find K neighborhood points as the neighborhood point set of terrain point P. The neighborhood point set of terrain point P is fitted into a local plane M using the least squares method. The centroid P0 and normal vector of the local plane M are The local plane is analyzed by principal component analysis to obtain the eigenvalues ​​arranged from small to large, and the eigenvector with the smallest eigenvalue is selected as the terrain eigenvector.

5. The method for extracting step lines from open-pit mine slopes by fusing multi-parameter point clouds according to claim 1 or 4, characterized in that: The potential step feature points are the junctions between the slope surface and the top of the slope, and between the slope surface and the bottom of the slope. The method for obtaining the potential step feature points is as follows: select a terrain feature vector, extract the terrain feature vectors of its neighborhood, calculate the average value of the angle between the selected terrain feature vector and the terrain feature vectors of the neighborhood, if the average value of the angle is greater than the set angle threshold, then the terrain point corresponding to the selected terrain feature vector is determined to be a potential step feature point, and all terrain feature vectors are traversed in turn to obtain all potential step feature points.

6. The method for extracting step lines from open-pit mine slopes by fusing multi-parameter point clouds according to claim 1 is characterized in that: The step feature point screening uses approximate differential operation and adopts a four-directional Sobel operator template to perform traversal convolution operations on potential step feature points and point cloud data. The elevation gradient size expression of the potential step feature point is: grad(P) represents the elevation gradient of the terrain point or potential step feature point P, F(P) represents the data after the terrain point or potential step feature point P is expressed by a two-dimensional discrete function, and f n (P) represents the data obtained by the Sobel operator template in four different directions, and the step feature points are screened based on the size of the elevation gradient.

7. The method for extracting step lines from open-pit mine slopes by fusing multi-parameter point clouds according to claim 1 is characterized in that: Step S5 also includes the following method: using the fitting surface method to perform quadratic surface fitting on the step feature points and their neighborhoods in the point cloud data to obtain a slope angle, and the slope angle expression is as follows: S x Indicates the slope of the step feature point in the x direction, S y It represents the slope of the step feature point in the y direction, θ represents the slope angle of the step feature point, and the step feature point with a slope angle less than the slope angle threshold is excluded; after exclusion, the step line fitting process is performed.

8. The method for extracting step lines from open-pit mine slopes by fusing multi-parameter point clouds according to claim 1 is characterized in that: The step line generation method includes: expressing the area where the target step feature point x is located by a target function polynomial, f(x)=p T (x)α(x),α(x)=[α1(x),α2(x),…,α m (x)] T , p T (x)=[p1(x),p2(x),…,p m (x)], where p m (x) represents the basis function of the m-order polynomial space in the region, α m (x) represents the coefficient vector of the m-order polynomial space in the region, defines the weighted error sum J of the fitting function at the target step feature point x in the moving least squares method, and fits to generate a continuous and smooth step line.

9. The method for extracting step lines from open-pit mine slopes by fusing multi-parameter point clouds according to claim 8 is characterized in that: The weighted error square sum J expression is as follows: Where n represents the step feature point x in the area where the target step feature point x is located. i Total,||xx i || represents the target step feature point x and the step feature point x i distance, h represents the smoothing parameter that controls the distance attenuation, Φ represents the cosine value of the angle between the normal vector of the target step feature point and the normal vector of the adjacent step feature point, θ 平滑 is the smoothing parameter that controls the influence of the normal vector, y i Represents the step feature point x i The node value of .

Images