A method for extracting a slope based on a three-dimensional oblique photography model of a drone

Abstract

The application discloses a kind of based on unmanned plane three-dimensional oblique photography model's side slope extraction method, to solve the problem of strong artificial dependence in existing side slope measurement, three-dimensional information is missing, parameter extraction efficiency and precision is insufficient.The method comprises: obtaining unmanned plane multi-angle aerial image and generating the three-dimensional real scene model of open pit mine;With digital elevation model as foundation, calculate the ground slope value and curvature value, in combination with edge detection and region growing algorithm, the potential side slope area is segmented, and the complete side slope contour is extracted;Longitudinal profile is generated along the side slope trend, the slope top line and slope bottom line are positioned by the first derivative extreme point of elevation curve, and curve fitting is carried out;Based on the corresponding feature point set on the slope top line and slope bottom line, the side slope geometric parameters are solved by spatial geometric calculation model.The present application realizes the full-process automation of side slope parameter, high-precision extraction, and can provide reliable three-dimensional data support for mine slope stability evaluation.

Description

Technical Field

[0001] This application relates to the field of surveying and mapping technology, and in particular to a slope extraction method based on a UAV three-dimensional oblique photogrammetry model. Background Technology

[0002] The geometric parameters of open-pit mine slopes (including the spatial locations of the slope crest and bottom lines, as well as core indicators such as slope angle, slope height, and slope length) are crucial data sources for mine safety and stability assessment, mining area design optimization, and geological hazard mitigation. Therefore, how to efficiently and accurately obtain these slope parameters has always been an important research direction in the fields of mine surveying and geotechnical engineering monitoring.

[0003] Currently, there are two main technical methods for obtaining the geometric parameters of open-pit mine slopes:

[0004] Firstly, there are traditional ground-contact manual surveying methods. These methods primarily rely on equipment such as total stations and RTK (Real-Time Kinematic) systems, where surveyors delve into the mining area or climb slopes to collect data point-by-point and line-by-line. However, this approach has several inherent limitations: First, it operates on a single-point basis, resulting in extremely low data collection efficiency and making it difficult to complete large-scale, full-area slope coverage surveys within a limited timeframe. Second, the surveying process requires personnel to frequently approach or stand on steep, fragmented slopes, posing a significant risk to personal safety. Third, due to severe limitations imposed by terrain undulations and visibility, blind spots are prevalent, leading to insufficient density and poor uniformity in the spatial distribution of the acquired data, making it difficult to comprehensively depict the overall macroscopic morphology of the slope.

[0005] Secondly, there are two-dimensional interpretation methods based on conventional aerial remote sensing orthophotos. With the widespread use of UAV platforms, some operations have gradually shifted to indoor vectorization interpretation based on orthophotos. While this method has replaced manual climbing operations to some extent and expanded the coverage of a single data acquisition, it is still essentially limited to two-dimensional planar projection information. Specifically, orthophotos lack accurate descriptions of abrupt changes in the actual elevation of the ground surface and the curvature of spatial surfaces. Furthermore, due to projection differences and image distortion, it is often difficult to accurately locate the actual spatial distribution of the slope crest and slope baseline in complex terrain areas. This results in significant errors in the calculated slope angle and slope height, failing to meet the requirements for high-precision slope stability analysis.

[0006] In summary, the common shortcomings of the two existing technical methods are as follows: On the one hand, the data obtained by traditional methods is either too dimensional or severely lacks spatial geometric information, making it impossible to construct a three-dimensional terrain entity with true geometric accuracy and measurable characteristics. As a result, it is difficult to comprehensively and realistically reflect the three-dimensional continuous morphology and spatial topological relationship of the slope. On the other hand, whether it is manual measurement or two-dimensional interpretation, the definition of the slope area, the extraction of the slope top line and slope bottom line, and the subsequent calculation of key parameters such as slope angle and slope height all rely heavily on manual interpretation, manual drawing, and experience-based calculation. The degree of automation is extremely low. This is not only time-consuming and labor-intensive, but also highly subjective and lacks standardization, which seriously restricts the data update frequency and large-scale application. Summary of the Invention

[0007] This application provides a slope extraction method based on a UAV three-dimensional oblique photogrammetry model, aiming to solve the technical problems of heavy reliance on manual labor, lack of three-dimensional information, and insufficient efficiency and accuracy of parameter extraction in existing slope measurement.

[0008] Firstly, a slope extraction method based on a UAV 3D oblique photogrammetry model is provided, including:

[0009] Acquire multi-angle aerial images from drones, and generate a 3D real-scene model of the open-pit mine based on the images. The model includes a dense point cloud, a 3D triangulation model, and a digital elevation model.

[0010] Based on the digital elevation model, the surface slope value and surface curvature value are calculated. Areas with slope values ​​greater than or equal to a preset slope threshold and curvature values ​​that meet the curvature abrupt change characteristics are identified as potential slope areas. The potential slope areas are then segmented based on edge detection algorithms and region growing algorithms to extract complete slope contours.

[0011] Based on the complete slope profile, several longitudinal profiles are generated along the slope direction. The elevation curves of each profile are extracted, the first derivative of the elevation curves is calculated, the extreme points of the first derivatives are located as candidate boundary points of the slope crest line and the slope bottom line, and the discrete candidate boundary points are curve fitted to generate a continuous and smooth slope crest line and slope bottom line.

[0012] Obtain the first set of feature points on the top of the slope and the corresponding second set of feature points on the bottom of the slope, and calculate the slope geometric parameters based on the spatial geometric calculation model.

[0013] Optionally, in the above scheme, the formula for calculating the surface slope is:

[0014] Plane slope angle system:

[0015] ;

[0016] and / or percentages:

[0017] ;

[0018] in, The elevation function of the digital elevation model. These are the partial derivatives of the elevation along the x and y directions, respectively.

[0019] Optionally, in the above scheme, the formula for calculating the surface curvature is:

[0020] ;

[0021] in, The elevation function of the digital elevation model. , , These are the second-order partial derivatives and mixed partial derivatives of elevation, respectively.

[0022] In the above scheme, optionally, the edge detection algorithm uses the Canny edge detection operator to extract the terrain boundary; the region growing algorithm grows seed points based on the slope value and the curvature value to segment the complete slope outline; the curve fitting uses the least squares method to fit the discrete candidate boundary points to correct breakpoints and distortion points.

[0023] In the above scheme, optionally, the slope geometric parameters include the slope vertical height, horizontal projection length, slope length, and slope angle;

[0024] Suppose the first feature point set includes feature points on the slope crest line. ;

[0025] The second feature point set includes the feature points corresponding to the slope line. ,but:

[0026] The vertical height of the slope ;

[0027] The horizontal projection length ;

[0028] The oblique length ;

[0029] The slope angle .

[0030] In the above scheme, optionally, for multi-stage stepped slopes or irregular curved slopes, the slope is divided into n segments, and the slope angle of the i-th segment is calculated for each segment. and slope length The overall slope angle is calculated using a segmented weighted average:

[0031] ;

[0032] Where n is the number of slope segments.

[0033] Optionally, in the above scheme, the acquisition parameters of the multi-angle aerial images taken by the UAV include: flight altitude. For a ground simulation of 30 meters, the directional overlap is... Lateral overlap ;

[0034] The generation process of the 3D real scene model includes: aerial triangulation, dense point cloud generation, triangular mesh construction, and texture mapping.

[0035] Optionally, the above solution may also include system integration and application steps:

[0036] A visual interactive interface is provided, which supports custom settings for the preset slope threshold and / or curvature threshold;

[0037] Supports manual point correction, boundary fine-tuning, model sectioning, and spatial measurement functions;

[0038] Output slope parameter reports, boundary line maps, and 3D thematic maps in batches according to preset specifications.

[0039] In the above scheme, optionally, the slope orientation is determined in the following way:

[0040] Extract all three-dimensional coordinate point clouds of the complete slope contour boundary obtained by edge detection and region growing segmentation, filter out noise discrete points and distortion breakpoints at the contour edge, and retain the continuous and effective boundary point set;

[0041] Projecting the entire boundary 3D point cloud onto the horizontal XY plane, and retaining only the planar coordinates (x, y) for calculation, the point cloud covariance matrix C is constructed:

[0042] ;

[0043] Where N is the total number of boundary point clouds, Let i be the horizontal projection coordinates of the i-th boundary point. These are the mean values ​​of the horizontal coordinates of all boundary points;

[0044] Eigendecomposition is performed on the covariance matrix C to solve for the eigenvalues ​​and corresponding eigenvectors. The eigenvector matching the largest eigenvalue is selected. This vector represents the main extension direction of the spatial distribution of the point cloud on the slope boundary. The direction of this horizontal vector is defined as the main direction of the slope.

[0045] A two-dimensional minimum bounding rectangle is constructed for the complete slope outline. The horizontal extension directions of the two long sides of the rectangle are extracted and compared with the main slope direction. If the angle between the two is less than a preset angle threshold, the direction result is determined to be valid. If the deviation exceeds the threshold, the main slope direction is taken as the final slope direction.

[0046] Furthermore, the longitudinal profile is laid out perpendicular to the slope direction.

[0047] In the above scheme, optionally further, the feature points on the slope crest line Feature points corresponding to the slope line The matching method is:

[0048] Based on the final slope orientation, a set of longitudinal profile cutting planes with equal spacing and strictly perpendicular to the slope orientation are generated. The profile spacing is customized according to the accuracy requirements of the mining area slope.

[0049] Each longitudinal profile cutting plane is spatially intersected with the continuous slope crest line and continuous slope trough line after least square fitting, and each longitudinal profile outputs a unique slope crest intersection point and a unique slope trough intersection point.

[0050] The intersection of the slope crests obtained on the same longitudinal profile is denoted as the first characteristic point. The intersection of the slope and the bottom is recorded as the second characteristic point. It automatically establishes a one-to-one matching relationship, forming a global set of matching feature point pairs;

[0051] Areas where the longitudinal profile intersects with the slope crest line or the bottom line are marked. These intersecting areas include areas where there is no intersection due to local slope folds, fractures, or step gaps. The marked areas are areas to be completed.

[0052] Compared with the prior art, this application has at least the following beneficial effects:

[0053] Based on further analysis and research of existing technical problems, this application recognizes that existing slope measurement technologies suffer from heavy reliance on manual labor, lack of 3D information, and insufficient efficiency and accuracy in parameter extraction. It constructs a high-precision 3D real-scene model using UAV oblique photography and, based on the model's spatial features such as slope, curvature, and elevation abrupt changes, comprehensively utilizes edge detection, region growing, and profile geometric analysis algorithms to achieve automatic identification of slope areas, intelligent extraction of boundary lines, and integrated calculation of core geometric parameters, achieving at least the following beneficial effects:

[0054] Significantly improves data acquisition efficiency and operational safety: replacing traditional total stations and RTK manual slope climbing and point-by-point measurement with drones flying around the terrain for multi-angle aerial photography, a single flight can quickly cover a large area, steep and broken areas, reducing the collection cycle from several days to several hours, and eliminating the need for workers to enter dangerous slopes, thus fundamentally eliminating personal safety hazards.

[0055] Overcoming the limitations of missing information in two-dimensional images, this application achieves accurate extraction in a true three-dimensional environment: Unlike the traditional method of relying on orthophotos for two-dimensional interpretation, the three-dimensional real-scene model constructed in this application has both accurate spatial coordinates and realistic textures. The slope and curvature calculations are based on a three-dimensional digital elevation model, which can objectively reflect the real undulations and abrupt changes of the ground surface, avoiding positioning errors caused by factors such as projection differences and image distortion. The spatial positioning accuracy of the slope top line and slope bottom line can reach the centimeter level.

[0056] The entire process of slope parameter extraction is automated, eliminating subjective human error: from slope area segmentation and boundary line positioning to the calculation of parameters such as slope angle, slope height, and slope length, no manual drawing or empirical calculation is required throughout the process. This overcomes the shortcomings of traditional methods, such as strong subjectivity and low standardization, and ensures the consistency and repeatability of extraction results between different mining areas and different batches of data.

[0057] Supports flexible adaptation and comprehensive analysis under complex working conditions: For multi-stage stepped slopes or irregular curved slopes, this application calculates the comprehensive slope angle by segmented weighted average, which can adapt to slope working conditions of different shapes and scales; at the same time, the key parameters involved in the method, such as slope threshold and curvature threshold, can be customized and adjusted, and can achieve data interoperability with slope stability analysis system and online monitoring and early warning system, providing standardized, high-frequency, and traceable three-dimensional data support for the safety management of mine slopes throughout their entire life cycle. Attached Figure Description

[0058] Figure 1 This is a flowchart illustrating a slope extraction method based on a UAV three-dimensional oblique photography model, provided as an embodiment of this application.

[0059] Figure 2 This is a flowchart illustrating the overall workflow of a slope extraction method based on a UAV 3D oblique photography model, provided as an embodiment of this application.

[0060] Figure 3 This is a functional block diagram of a slope extraction system based on a UAV three-dimensional oblique photography model, provided as an embodiment of this application.

[0061] Figure 4 This is a schematic diagram of slope geometric parameter calculation provided in one embodiment of this application. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0063] In the description of this application: unless otherwise stated, "a plurality of" means two or more. The terms "first," "second," "third," etc., in this application are intended to distinguish the objects referred to and do not have any special meaning in terms of technical connotation (e.g., they should not be construed as an emphasis on importance or order). Expressions such as "including," "comprising," and "having" also mean "not limited to" (certain units, components, materials, steps, etc.).

[0064] This invention aims to provide a slope extraction method based on a UAV 3D oblique photography model. It collects topographic data of the entire mine area through multi-angle UAV photography, constructs a high-precision 3D real-scene model, and automatically and accurately identifies the entire slope outline and locates the spatial positions of the slope crest and bottom line based on the spatial features of the 3D model, including slope, curvature, elevation changes, and normal vector differences. Furthermore, it automatically obtains multiple core parameters such as slope angle, slope height, and slope length through geometric calculations. The entire process requires no manual intervention to achieve integrated, automated, and high-precision extraction of key parameters for various slope types, providing detailed and reliable 3D data support for slope stability assessment and management.

[0065] In one embodiment, reference Figure 1 This paper provides a slope extraction method based on a UAV 3D oblique photogrammetry model, including:

[0066] Acquire multi-angle aerial images from drones, and generate a 3D real-scene model of the open-pit mine based on the images. The model includes a dense point cloud, a 3D triangulation model, and a digital elevation model.

[0067] Based on the digital elevation model, the surface slope value and surface curvature value are calculated. Areas with slope values ​​greater than or equal to a preset slope threshold and curvature values ​​that meet the curvature abrupt change characteristics are identified as potential slope areas. The potential slope areas are then segmented based on edge detection algorithms and region growing algorithms to extract complete slope contours.

[0068] Based on the complete slope profile, several longitudinal profiles are generated along the slope direction. The elevation curves of each profile are extracted, the first derivative of the elevation curves is calculated, the extreme points of the first derivatives are located as candidate boundary points of the slope crest line and the slope bottom line, and the discrete candidate boundary points are curve fitted to generate a continuous and smooth slope crest line and slope bottom line.

[0069] Obtain the first set of feature points on the top of the slope and the corresponding second set of feature points on the bottom of the slope, and calculate the slope geometric parameters based on the spatial geometric calculation model.

[0070] The technical solution of this embodiment will be described in detail below.

[0071] (a) Data acquisition and 3D modeling.

[0072] In this embodiment, data acquisition and the construction of a 3D reality model are performed first. Specifically, a drone equipped with a multi-lens oblique photography camera is used to conduct aerial photography of the open-pit mine by setting an oblique photography flight path. The flight path parameters are set as follows: flight altitude... For a ground simulation of 30 meters, the directional overlap is... Lateral overlap The drone flew in a terrain-following manner according to the above parameters, conducting multi-angle, overlapping aerial photography of the open-pit mine to acquire high-resolution, multi-view images. Among these parameters, the flight altitude... The drone is set to a 30-meter terrain-following altitude, meaning it automatically adjusts its absolute altitude based on terrain undulations during flight, maintaining a relative altitude of approximately 30 meters above the ground. This terrain-following flight strategy ensures consistent image data acquisition across various terrain conditions, avoiding uneven image resolution caused by terrain variations. (Front-course overlap...) and lateral overlap The high overlap rate setting ensures sufficient overlap between adjacent images, providing ample corresponding points for subsequent aerial triangulation and dense matching, thereby improving the accuracy and completeness of 3D reconstruction.

[0073] After acquiring the images, a high-precision 3D oblique photogrammetry model containing a Digital Elevation Model (DEM) is generated using 3D modeling software. The modeling process includes: aerial triangulation, dense point cloud generation, triangular mesh construction, and texture mapping. The final output consists of three types of results: dense point cloud, 3D triangular mesh model, and textured real-world model. The generated models achieve absolute orientation of their 3D spatial coordinates based on image control points or PPK / RTK positioning technology. The single-point planar accuracy and elevation accuracy of the model reach the centimeter level, providing a high-precision 3D data base for the subsequent accurate extraction of slope features. Specifically, in the aerial triangulation stage, corresponding feature points between different images are identified, and the exterior orientation elements (position and attitude) of each image and the 3D coordinates of the feature points are calculated. In the dense point cloud generation stage, based on the results of the aerial triangulation, a high-density 3D point cloud is generated using the Multi-View Stereo (MVS) algorithm. In the triangular mesh construction stage, the dense point cloud is constructed into a continuous irregular triangular mesh (TIN) model using algorithms such as Delaunay triangulation. In the texture mapping stage, the texture information of the original image is mapped onto the surface of the triangular mesh model to generate a textured realistic 3D model. The above process is the standard workflow for oblique photogrammetry 3D modeling and will not be elaborated further here.

[0074] In this embodiment, the three-dimensional reality model uses an elevation function. The function expresses the landform in the form of x and y, where x and y are horizontal coordinates and Z is the elevation value of the corresponding location. This function provides the basis for subsequent calculations of terrain feature parameters such as slope and curvature.

[0075] (II) Slope Area Identification and Contour Extraction (Reference) Figure 2 and Figure 3 ).

[0076] After the three-dimensional reality model is constructed, the surface slope and surface curvature values ​​are calculated based on the digital elevation model to identify potential slope areas.

[0077] The slope is calculated as follows: based on the elevation function of the digital elevation model. The slope components along the x and y directions of the earth's surface are calculated, and then the overall slope value is obtained.

[0078] In one embodiment, the formula for calculating the surface slope (in degrees) is:

[0079] ;

[0080] Slope can also be expressed as a percentage:

[0081] ;

[0082] in, The elevation function of the digital elevation model. These are the partial derivatives of the elevation along the x and y directions, respectively.

[0083] As mentioned earlier, slope is one of the core indicators for identifying slope areas, and its quantitative calculation directly affects the accuracy of slope area determination. In this embodiment, slope calculation is based on the elevation function Z(x,y) of the digital elevation model. The elevation function has been obtained through three-dimensional reconstruction during the modeling stage and stored in the computer system in the form of a regular grid (e.g., a rectangular grid) or an irregular triangular network (TIN).

[0084] In practical implementation, for a regular grid digital elevation model, the partial derivatives Numerical approximation can be performed using the finite difference method. For example, classical methods such as the third-order inverse distance squared weighted difference method or the Sobel operator can be used to calculate the slope components of each grid cell in the x and y directions. After obtaining the slope components in both directions, they can be substituted into the above formula to obtain the comprehensive slope value.

[0085] Slope values ​​can be output in either angle (degrees or radians) or percentage format to suit the needs of different engineering specifications and application scenarios. Angle-based slopes intuitively reflect the angle of inclination of the ground surface, while percentage-based slopes are commonly used in slope design standards for road engineering, railway engineering, and other fields.

[0086] The curvature is calculated as follows: Curvature reflects the characteristics of surface unevenness; at the top and bottom of a slope, there are usually significant abrupt changes in surface curvature. In one embodiment, the formula for calculating surface curvature β is:

[0087] ;

[0088] Where Z(x,y) is the elevation function of the digital elevation model. These are the second-order partial derivatives and mixed partial derivatives of elevation, respectively.

[0089] Curvature reflects the degree of undulation in the landform and is an important topographic feature parameter for distinguishing different landform parts such as the top of a slope, the surface of a slope, and the bottom of a slope.

[0090] In this embodiment, curvature calculation is also based on the elevation function Z(x,y) of the digital elevation model. The first and second partial derivatives in the formula can be calculated from the regular grid digital elevation model using the finite difference method.

[0091] The spatial distribution characteristics of curvature values ​​are of great significance for slope boundary location: at the slope crest line, the surface morphology transitions from the slope crest plateau to the slope surface, and the curvature typically exhibits a positive abrupt change from zero or a small value to a larger value; at the slope bottom line, the surface morphology transitions from the slope surface to the slope bottom plateau, and the curvature typically exhibits a negative abrupt change from a larger value to zero or a smaller value. Therefore, combining the curvature abrupt change characteristics with slope thresholds can significantly improve the accuracy and robustness of slope boundary identification.

[0092] After obtaining the slope and curvature values, a slope threshold is set. When consecutive points within the region satisfy At that time, the area was initially identified as a potential slope area. Based on this, a comprehensive judgment was made by combining the curvature change characteristics: if the curvature value shows a significant change in space (i.e., the first or second derivative of curvature exceeds the preset curvature threshold) under the premise that the slope meets the threshold condition, the determination that the area is a slope boundary is strengthened.

[0093] Subsequently, edge detection and region growing algorithms are used to accurately segment the potential slope area and extract the complete slope contour. In one embodiment, the edge detection algorithm uses the Canny edge detection operator to extract the terrain boundary. The 3D mesh model data needs to be mapped to a 2D terrain feature grayscale image (e.g., using slope or curvature values ​​as pixel grayscale) before being processed using the Canny operator. The region growing algorithm grows seed points based on the slope and curvature values ​​to segment the complete slope contour: first, seed points are selected within the potential slope area (e.g., the point with the highest slope value is selected as the initial seed point), and then region growing is performed according to the similarity criteria of slope and curvature values ​​of adjacent points. Neighboring points that meet the similarity conditions are continuously included in the growing area until further growth is impossible, thereby segmenting the complete slope contour. The curve fitting uses the least squares method to fit the discrete candidate boundary points to correct breakpoints and distortion points.

[0094] Regarding the Canny edge detection operator: Canny edge detection is a classic image edge extraction algorithm with advantages such as high signal-to-noise ratio, high positioning accuracy, and single-edge response. In this embodiment, since the original data is a three-dimensional mesh model, it is necessary to first map the three-dimensional mesh model into a two-dimensional terrain feature grayscale image—for example, using the slope or curvature value of each mesh node as the grayscale value of the corresponding pixel to form a two-dimensional grayscale image—and then use the Canny operator to perform edge detection on this grayscale image to extract the terrain boundaries. The processing flow of the Canny operator includes: Gaussian filtering to smooth the image, calculating the gradient magnitude and direction, non-maximum suppression, double threshold detection, and edge connection.

[0095] Regarding the region growing algorithm: Based on the terrain boundaries extracted by Canny edge detection, a region growing algorithm is used to accurately segment the slope region. The basic idea of ​​region growing is: select one or more seed points within the potential slope region (for example, select the point with the highest slope value and curvature value that meets the abrupt change condition as the initial seed point). Then, starting from the seed point, according to the preset growth criteria (for example, the difference between the slope value of the adjacent point and the slope value of the seed point is less than a preset threshold, and the change in curvature value is within the allowable range), gradually incorporate the neighboring points that meet the conditions into the growing region until no new points meet the growth conditions. After the growing is completed, all the incorporated points constitute the complete slope region, and its outer boundary is the complete slope outline.

[0096] Regarding least squares curve fitting: After extracting discrete candidate boundary points from each longitudinal profile, curve fitting is required to form a continuous and smooth boundary line. The least squares method solves for the optimal curve parameters by minimizing the sum of squared vertical distances from all discrete points to the fitted curve. This effectively suppresses the offset of individual points caused by data noise, local terrain distortion, etc., thereby correcting discontinuities and distortion points, and obtaining smooth and continuous slope crest and slope bottom lines.

[0097] (III) Extraction of the top and bottom lines of the slope.

[0098] After obtaining the complete slope outline, several longitudinal profiles are generated along the slope direction. The elevation curves of each profile are extracted, and then the slope top line and slope bottom line are located.

[0099] Specifically, the slope orientation is first determined. In one embodiment, the slope orientation is determined in the following way:

[0100] Extract all three-dimensional coordinate point clouds of the complete slope contour boundary obtained by edge detection and region growing segmentation, filter out noise discrete points and distortion breakpoints at the contour edge, and retain the continuous and effective boundary point set;

[0101] Projecting the entire boundary 3D point cloud onto the horizontal XY plane, and retaining only the planar coordinates (x, y) for calculation, the point cloud covariance matrix C is constructed:

[0102] ;

[0103] Where N is the total number of boundary point clouds, Let i be the horizontal projection coordinates of the i-th boundary point. These are the mean values ​​of the horizontal coordinates of all boundary points;

[0104] Eigendecomposition is performed on the covariance matrix C to solve for the eigenvalues ​​and corresponding eigenvectors. The eigenvector matching the largest eigenvalue is selected. This vector represents the main extension direction of the spatial distribution of the point cloud on the slope boundary. The direction of this horizontal vector is defined as the main direction of the slope.

[0105] A two-dimensional minimum bounding rectangle is constructed for the complete slope outline. The horizontal extension directions of the two long sides of the rectangle are extracted and compared with the main slope direction. If the angle between the two is less than a preset angle threshold, the direction result is determined to be valid. If the deviation exceeds the threshold, the main slope direction is taken as the final slope direction.

[0106] Furthermore, the longitudinal profile is laid out perpendicular to the slope direction.

[0107] As mentioned earlier, the automatic determination of slope orientation is a key step in the transition from "slope area segmentation" to "longitudinal profile generation". In this embodiment, principal component analysis (PCA) combined with minimum bounding rectangle verification is used to achieve automatic and robust determination of slope orientation.

[0108] Step 1: Boundary Point Cloud Extraction and Preprocessing: After obtaining the complete slope outline through edge detection and region growing algorithms, extract all 3D coordinate point clouds of the outline boundary. Since the original data may contain noise-induced discrete points and distortion breaks due to image matching errors, terrain occlusion, etc., preprocessing is required—for example, using statistical filtering or radius filtering to remove outliers, and using a breakpoint connection algorithm to repair small-scale breaks—to retain a continuous and valid set of boundary points to ensure the accuracy of subsequent orientation determination.

[0109] Step 2: Projection Dimensionality Reduction and Covariance Matrix Construction: Project all boundary 3D point clouds onto the horizontal XY plane, retaining only the planar coordinates. This projection dimensionality reduction operation simplifies the orientation determination problem in three-dimensional space into the principal direction extraction problem in a two-dimensional plane, significantly reducing computational complexity.

[0110] Then, the covariance matrix C of the point cloud is constructed. The covariance matrix describes the spatial distribution characteristics of the boundary point cloud in a two-dimensional plane: diagonal elements and These reflect the degree of discretization of the point cloud in the x and y directions, respectively, and the off-diagonal elements. This reflects the correlation between x and y.

[0111] Step 3: Eigenvalue Decomposition and Principal Direction Extraction: Perform eigenvalue decomposition on the covariance matrix C and solve for the eigenvalues. and its corresponding eigenvectors Largest eigenvalue corresponding feature vector The direction representing the largest variance in the spatial distribution of the point cloud is the principal extension direction of the boundary point cloud. For a long and narrow slope profile, the principal extension direction of its boundary point cloud is the slope's strike direction. Therefore, this horizontal vector direction is defined as the slope's principal strike direction.

[0112] Step 4: Minimum Bounding Rectangle Auxiliary Verification: To enhance the reliability of orientation determination, an auxiliary verification mechanism is introduced. A two-dimensional minimum bounding rectangle is constructed for the complete slope profile—that is, the rectangle with the smallest area that can contain all boundary points. The direction of the rectangle changes with the orientation of the profile. The horizontal extension directions of the two long sides of this rectangle are extracted as reference directions.

[0113] The angle between the main direction obtained by PCA and the direction of the longest side of the minimum bounding rectangle is compared. If the angle is less than a preset angle threshold (preferably...), then... This indicates that the judgment results of the two independent methods are highly consistent, and the current direction result is valid. If the deviation exceeds the threshold, it indicates that there may be special circumstances (such as the special shape of the slope profile causing the PCA result to be unstable, or there is a lot of noise in the profile affecting the PCA calculation). In this case, the PCA main direction is taken as the final slope direction, and a manual review prompt can be issued at the same time.

[0114] After determining the slope orientation, all longitudinal profiles are laid out perpendicular to that orientation to ensure that each profile can effectively traverse from the top to the bottom of the slope and accurately extract the elevation change curve.

[0115] After determining the slope orientation, a series of longitudinal profiles are laid out along the slope orientation, with each profile parallel to the others. For each longitudinal profile, the surface elevation data at that profile location is extracted to form a profile elevation curve. , where l is the horizontal length coordinate of the profile.

[0116] Then, the first derivative of the elevation curve is calculated. The first derivative characterizes the rate of change of elevation along the horizontal direction. At the crest and trough of the slope, the surface elevation typically undergoes significant abrupt changes, reflected on the elevation curve as points of sudden slope changes, corresponding to the extreme points of the first derivative. Therefore, the extreme points of the first derivative are identified as candidate boundary points between the crest and trough of the slope.

[0117] For multiple longitudinal profiles laid out along the slope direction, a pair of candidate boundary points (slope top candidate point and slope bottom candidate point) can be extracted from each profile. By pooling the slope top candidate points from all profiles, discrete slope top boundary points are obtained; by pooling the slope bottom candidate points from all profiles, discrete slope bottom boundary points are obtained.

[0118] Finally, the least squares method is used to perform curve fitting on the discrete slope crest and slope trough boundary points to correct for discontinuities and distortions caused by data noise and local terrain distortion, thereby generating a continuous and smooth slope crest line. With the bottom line of the slope .

[0119] (iv) Calculation of slope geometric parameters (reference) Figure 4 ).

[0120] After obtaining a continuous and smooth slope crest line and slope bottom line, the first feature point set on the slope crest line and the corresponding second feature point set on the slope bottom line are obtained, and the slope geometric parameters are calculated based on the spatial geometric calculation model.

[0121] In one embodiment, feature points on the slope crest line Feature points corresponding to the slope line The matching method is:

[0122] Based on the final slope orientation, a set of longitudinal profile cutting planes with equal spacing and strictly perpendicular to the slope orientation are generated. The profile spacing is customized according to the accuracy requirements of the mining area slope.

[0123] Each longitudinal profile cutting plane is spatially intersected with the continuous slope crest line and continuous slope trough line after least square fitting, and each longitudinal profile outputs a unique slope crest intersection point and a unique slope trough intersection point.

[0124] The intersection of the slope crests obtained on the same longitudinal profile is denoted as the first characteristic point. The intersection of the slope and the bottom is recorded as the second characteristic point. It automatically establishes a one-to-one matching relationship, forming a global set of matching feature point pairs;

[0125] Areas where the longitudinal profile intersects with the slope crest line or the bottom line are marked. These intersecting areas include areas where there is no intersection due to local slope folds, fractures, or step gaps. The marked areas are areas to be completed.

[0126] As mentioned earlier, the slope crest line and slope trough line are two independently extracted spatial curves. When calculating the slope angle and slope height, it is necessary to pair the correct slope crest point with the slope trough point. In this embodiment, the automatic one-to-one matching of feature points is achieved by using the longitudinal profile cutting intersection method.

[0127] Step 1: Generation of longitudinal profile cutting planes: Based on the slope orientation determined in the aforementioned embodiments, a set of equally spaced longitudinal profile cutting planes that are strictly perpendicular to the orientation are generated. These cutting planes are arranged in parallel in space, and each cutting plane is an infinitely extending vertical plane (represented as a straight line on a horizontal plane and as a vertical plane in three-dimensional space).

[0128] The profile spacing directly affects the density of subsequent parameter extraction: a smaller spacing results in more feature point pairs being extracted and more refined parameter calculations, but also increases the computational load; a larger spacing results in higher computational efficiency, but may miss local terrain details. Therefore, the profile spacing can be customized according to the accuracy requirements of the mining area slopes. For example, a 1-meter profile spacing can be set for key monitoring slopes with high accuracy requirements; for general survey slopes, a spacing of 3 meters or 5 meters can be set.

[0129] Step 2: Spatial Intersection Calculation: Perform three-dimensional spatial intersection calculations on each longitudinal profile cutting plane with the fitted continuous slope crest line and continuous slope bottom line. Since the slope crest line and slope bottom line are both three-dimensional spatial curves, and they intersect with the vertical cutting plane in space (rather than being parallel or skew), each cutting plane has at most one intersection point with each curve.

[0130] Specifically, for the j-th longitudinal profile, its intersection with the slope crest line yields the slope crest intersection point. The intersection with the bottom line of the slope yields the intersection point at the bottom of the slope. Since the cutting plane is vertical, and the slope crest line and slope bottom line extend continuously in three-dimensional space, as long as the profile position is within the effective range of the slope, the intersection calculation can usually obtain a unique solution.

[0131] Step 3: One-to-one matching: Match the intersection points of the slope crests obtained from the same longitudinal profile. Intersection with the bottom of the slope These are considered as a set of corresponding points. The physical significance of this matching method is that the surface morphology at the same profile location changes continuously from the top of the slope to the bottom of the slope, and the top and bottom points on this profile naturally form a corresponding relationship of "the same cross section".

[0132] By performing the above operations on each longitudinal profile, a set of globally matching feature point pairs can be generated. , where n is the total number of effective profiles. For each pair of corresponding points, the geometric parameters such as slope angle, slope height, and slope length at that profile can be calculated according to the method described in the following embodiments.

[0133] Step 4: Handling Intersecting Failure Areas: In actual working conditions, due to the possibility of local folds, fracture structures, step gaps, or missing data on the slope, some longitudinal profile cutting planes may not intersect with the slope crest line or slope bottom line (for example, the profile location happens to be in a step gap area, where the slope bottom line is discontinuous). For such intersecting failure areas, the system automatically marks and records the location of the failed profile and the cause of failure (whether it is a lack of intersection with the slope crest line, a lack of intersection with the slope bottom line, or a lack of intersection with both).

[0134] The marked area serves as the region to be completed. Subsequent completion can be achieved through interpolation using parameter values ​​from adjacent valid profiles—for example, using linear interpolation or spline interpolation methods, and calculating parameter values ​​for the failed area using parameters such as slope angle and slope height from adjacent profiles—to ensure the continuity and integrity of parameters along the slope direction. The specific interpolation method for the region to be completed can be flexibly selected according to actual accuracy requirements; this invention does not limit this method.

[0135] In one embodiment, let the matched pair of corresponding points be: the first feature point set includes feature points on the slope crest line. The second feature point set includes the corresponding feature points on the slope line. ,

[0136] Based on the corresponding point pairs mentioned above, the following slope geometric parameters are calculated:

[0137] The vertical height of the slope ;

[0138] The horizontal projection length ;

[0139] The oblique length ;

[0140] The slope angle .

[0141] After obtaining the corresponding feature point pairs on the top and bottom slopes, the slope geometric parameters at each location can be calculated one by one based on the three-dimensional spatial coordinates of each pair of corresponding points using spatial geometry principles.

[0142] Vertical height H of a slope: This is the elevation difference between the top and bottom of the slope, directly reflecting the vertical scale of the slope and is one of the most fundamental parameters in slope stability assessment. Its calculation involves only the elevation coordinates of two points. and Its physical meaning is clear.

[0143] Horizontal projection length This refers to the straight-line distance between the top and bottom of the slope on a horizontal plane, reflecting the horizontal extent of the slope. Its calculation involves only the plane coordinates of two points. and .

[0144] Slope length L: This is the straight-line distance between the top and bottom of the slope, reflecting the actual length of the slope surface. Its calculation combines horizontal distance and elevation difference, and is an intermediate parameter for calculating the slope angle.

[0145] slope angle The slope angle, or angle between the slope surface and the horizontal plane, is a core parameter for measuring the steepness of a slope. The slope angle can be calculated using two equivalent methods: the arcsine function (based on the ratio of vertical height to slope length) or the arctangent function (based on the ratio of vertical height to horizontal projected length). These two methods are mathematically equivalent, and either can be chosen based on data availability in practical applications.

[0146] In actual calculations, since both the top and bottom slope lines are continuous curves, multiple pairs of corresponding points can be sampled evenly along the slope direction, and the geometric parameter values ​​at each sampling location can be calculated to form a distribution curve of parameters along the slope direction, thereby comprehensively reflecting the spatial variation law of the slope's geometric characteristics.

[0147] The above calculations are performed for each pair of corresponding points, yielding geometric parameter values ​​at multiple locations along the slope's direction. In one embodiment, for a multi-step slope or an irregular curved slope, the slope is divided into n segments, and the slope angle of the i-th segment is calculated for each segment. and slope length The overall slope angle is calculated using a segmented weighted average. :

[0148] ;

[0149] Where n is the number of slope segments.

[0150] Through the above steps, this embodiment realizes fully automated extraction of slope geometric parameters from UAV image acquisition to slope geometric parameter output.

[0151] In actual mine slopes, the slope morphology is often not a single slope surface, but a complex combination of multiple steps, platforms, and slope segments. For such slopes, a single slope angle value cannot fully reflect the overall steepness of the slope, and a comprehensive evaluation index is required.

[0152] In this embodiment, the entire slope is first divided into n segments along its direction or along its surface. The division can be based on the location of steps, inflection points of slope changes, or equal-spaced segments. For each slope segment, the slope angle is calculated. and slope length .

[0153] Then, based on the length of each slope segment As a weight, the slope angle of each segment The overall slope angle is obtained by performing a weighted average. The physical meaning of weighted averaging is that longer slope segments contribute more to the overall macroscopic shape of the slope, and therefore should be given higher weight in the comprehensive evaluation. This segmented weighted averaging method can objectively and accurately reflect the overall steepness of complex slopes, providing a more scientific parameter basis for slope stability analysis.

[0154] In one embodiment, the system integration and application steps are also included:

[0155] A visual interactive interface is provided, which supports custom settings for the preset slope threshold and / or curvature threshold;

[0156] Supports manual point correction, boundary fine-tuning, model sectioning, and spatial measurement functions;

[0157] Output slope parameter reports, boundary line maps, and 3D thematic maps in batches according to preset specifications.

[0158] In this embodiment, regarding the visual interactive interface: the system provides a graphical user interface, through which users can intuitively view the 3D real-world model, slope outline, slope crest line, slope baseline, and calculated geometric parameters. The interface supports key algorithm parameters (such as slope thresholds)... The system adjusts parameters (such as curvature threshold) in real time, and automatically recalculates and updates the displayed results after adjustment, making it easy for users to optimize parameters according to the actual terrain characteristics of the mining area.

[0159] Regarding manual intervention and verification functions: The system supports manual point correction (e.g., manually dragging and adjusting individual deviation points extracted automatically), boundary fine-tuning (e.g., smoothing or correcting local sections of the slope crest or bottom line), model sectioning (e.g., generating cross-sectional or longitudinal section views at any location), and spatial measurement (e.g., directly measuring distances, areas, elevations, etc. on the 3D model). These functions provide means for verifying and fine-tuning the automatically extracted results.

[0160] Regarding output: The system supports batch output of various output files according to industry standards, including but not limited to: Excel-formatted slope parameter reports (containing detailed lists of parameters such as slope angle, slope height, and slope length for each section or segment), CAD-formatted boundary line maps (containing vector data such as slope top line and slope bottom line), and 3D thematic maps (displaying the distribution of various parameters in chromatographic or labeled form on a 3D model). These outputs can be directly used in engineering documents such as slope stability analysis reports and treatment design schemes.

[0161] Compared with the prior art, the present invention has at least the following advantages and positive effects:

[0162] Improving data acquisition efficiency and security: Drone operations replace manual field surveys, quickly covering large areas and high-risk slope areas, greatly shortening the data collection cycle, and completely avoiding the risk of personnel entering dangerous environments.

[0163] Achieve automated and high-precision extraction of slope parameters: Based on a high-resolution 3D real-scene model, the integrated algorithm automatically completes feature recognition and parameter calculation, overcoming the subjective error and efficiency bottleneck of traditional methods. The extraction accuracy can reach the centimeter level, and it supports continuous slope analysis under complex terrain conditions.

[0164] Provides a fully three-dimensional, measurable data base: The generated three-dimensional oblique photogrammetry model has both geometric accuracy and realistic texture, supports arbitrary cutting, measurement and visualization analysis, and provides richer and more intuitive data support for slope stability assessment, deformation monitoring and treatment design.

[0165] Promoting the digital and intelligent transformation of slope management: This method automates the entire process from data collection and processing to parameter extraction, significantly improving the standardization and update frequency of slope data production, and laying a solid technical foundation for the full life-cycle safety management of mine slopes.

[0166] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A slope extraction method based on a UAV 3D oblique photogrammetry model, characterized in that, include: Acquire multi-angle aerial images from drones, and generate a 3D real-scene model of the open-pit mine based on the images. The model includes a dense point cloud, a 3D triangulation model, and a digital elevation model. Based on the digital elevation model, the surface slope value and surface curvature value are calculated. Areas with slope values ​​greater than or equal to a preset slope threshold and curvature values ​​that meet the curvature abrupt change characteristics are identified as potential slope areas. The potential slope areas are then segmented based on edge detection algorithms and region growing algorithms to extract complete slope contours. Based on the complete slope profile, several longitudinal profiles are generated along the slope direction. The elevation curves of each profile are extracted, the first derivative of the elevation curves is calculated, the extreme points of the first derivatives are located as candidate boundary points of the slope crest line and the slope bottom line, and the discrete candidate boundary points are curve fitted to generate a continuous and smooth slope crest line and slope bottom line. Obtain the first set of feature points on the top of the slope and the corresponding second set of feature points on the bottom of the slope, and calculate the slope geometric parameters based on the spatial geometric calculation model.

2. The method according to claim 1, characterized in that, The formula for calculating the surface slope is: Plane slope angle system: ； and / or percentages: ； in, The elevation function of the digital elevation model. These are the partial derivatives of the elevation along the x and y directions, respectively.

3. The method according to claim 1, characterized in that, The formula for calculating the surface curvature is: ； in, The elevation function of the digital elevation model. , , These are the second-order partial derivatives and mixed partial derivatives of elevation, respectively.

4. The method according to claim 1, characterized in that, The edge detection algorithm uses the Canny edge detection operator to extract the terrain boundary; the region growing algorithm grows seed points based on the slope value and the curvature value to segment the complete slope outline; the curve fitting uses the least squares method to fit the discrete candidate boundary points to correct breakpoints and distortion points.

5. The method according to claim 1, characterized in that, The slope geometric parameters include the slope's vertical height, horizontal projected length, slope length, and slope angle; Suppose the first feature point set includes feature points on the slope crest line. ; The second feature point set includes the feature points corresponding to the slope line. ,but: The vertical height of the slope ; The horizontal projection length ; The oblique length ; The slope angle .

6. The method according to claim 5, characterized in that, For multi-stage stepped slopes or irregular curved slopes, the slope is divided into n segments, and the slope angle of the i-th segment is calculated for each segment. and slope length The overall slope angle is calculated using a segmented weighted average: ； Where n is the number of slope segments.

7. The method according to claim 1, characterized in that, The parameters for acquiring the multi-angle aerial images from the UAV include: flight altitude. For a ground simulation of 30 meters, the directional overlap is... Lateral overlap ; The generation process of the 3D real scene model includes: aerial triangulation, dense point cloud generation, triangular mesh construction, and texture mapping.

8. The method according to claim 1, characterized in that, It also includes system integration and application steps: A visual interactive interface is provided, which supports custom settings for the preset slope threshold and / or curvature threshold; Supports manual point correction, boundary fine-tuning, model sectioning, and spatial measurement functions; Output slope parameter reports, boundary line maps, and 3D thematic maps in batches according to preset specifications.

9. The method according to claim 1, characterized in that, The slope orientation is determined in the following way: Extract all three-dimensional coordinate point clouds of the complete slope contour boundary obtained by edge detection and region growing segmentation, filter out noise discrete points and distortion breakpoints at the contour edge, and retain the continuous and effective boundary point set; Projecting the entire boundary 3D point cloud onto the horizontal XY plane, and retaining only the planar coordinates (x, y) for calculation, the point cloud covariance matrix C is constructed: ； Where N is the total number of boundary point clouds, Let i be the horizontal projection coordinates of the i-th boundary point. These are the mean values ​​of the horizontal coordinates of all boundary points; Eigendecomposition is performed on the covariance matrix C to solve for the eigenvalues ​​and corresponding eigenvectors. The eigenvector matching the largest eigenvalue is selected. This vector represents the main extension direction of the spatial distribution of the point cloud on the slope boundary. The direction of this horizontal vector is defined as the main direction of the slope. A two-dimensional minimum bounding rectangle is constructed for the complete slope outline. The horizontal extension directions of the two long sides of the rectangle are extracted and compared with the main slope direction. If the angle between the two is less than a preset angle threshold, the direction result is determined to be valid. If the deviation exceeds the threshold, the main slope direction is taken as the final slope direction. Furthermore, the longitudinal profile is laid out perpendicular to the slope direction.

10. The method according to claim 9, characterized in that, Feature points on the slope crest line Feature points corresponding to the slope line The matching method is: Based on the final slope orientation, a set of longitudinal profile cutting planes with equal spacing and strictly perpendicular to the slope orientation are generated. The profile spacing is customized according to the accuracy requirements of the mining area slope. Each longitudinal profile cutting plane is spatially intersected with the continuous slope crest line and continuous slope trough line after least square fitting, and each longitudinal profile outputs a unique slope crest intersection point and a unique slope trough intersection point. The intersection of the slope crests obtained on the same longitudinal profile is denoted as the first characteristic point. The intersection of the slope and the bottom is recorded as the second characteristic point. It automatically establishes a one-to-one matching relationship, forming a global set of matching feature point pairs; Areas where the longitudinal profile intersects with the slope crest line or the bottom line are marked. These intersecting areas include areas where there is no intersection due to local slope folds, fractures, or step gaps. The marked areas are areas to be completed.

Images