A regularity-constrained method for extracting bench lines in open-pit mines

Abstract

The present invention relates to the field of industrial information and data processing technology, and discloses a regularity-constrained open-pit mine step line extraction algorithm, comprising: step 1, identifying geometric feature surfaces of original point cloud data, and extracting feature surface data based on the feature surfaces; step 2, performing supervoxel segmentation on the feature surface data based on boundary features of the feature surfaces to form supervoxel units; plane-classifying the supervoxel units according to resolution to form supervoxel sets; step 3, establishing a global graph model based on the adjacency relationship of the supervoxel units; performing surface segmentation on the global graph model based on the energy optimization principle to obtain a face set; selecting faces with verticality exceeding a threshold as a slope candidate set; and step 4, extracting slope boundaries from the slope candidate set, using the slope boundary point cloud as the feature points of the step line, and generating the step line. The present invention extracts step line feature points by identifying geometric feature surfaces rather than feature points, thereby solving the problem of discontinuity of step lines caused by insufficient feature point extraction.

Description

Technical Field

[0001] The invention relates to the technical field of industrial information and data processing, and in particular to a regularity-constrained open-pit mine step line extraction method. Background Art

[0002] Step line extraction is crucial in geometric model construction and production management in open-pit mines. First, step lines intuitively reflect the mining progress and geometric form of the mine, providing key reference for managers. Second, accurate step line information helps identify potential safety hazards. By monitoring and adjusting mining plans, geological disasters such as landslides can be prevented, improving production safety. At the same time, regularly extracting and storing step line data can track mine evolution trends and optimize future mining plans. More importantly, accurate step line data can help assess the impact of mining on the environment, develop scientific restoration and management plans, and ensure sustainable ecological and environmental development. However, due to the complex terrain of open-pit mines, step line data extraction is cumbersome, resulting in low data extraction efficiency and difficulty in ensuring accuracy. Therefore, automated step line extraction is a key step in mine automation, thereby improving extraction efficiency and accuracy.

[0003] To achieve the above goals, data extraction is often carried out by combining drone point cloud data with automated algorithms to realize automated extraction of step lines. This method not only improves the extraction efficiency and accuracy, but also reduces human intervention to a certain extent, providing strong support for the modernization and transformation of mines.

[0004] However, after step line data extraction, step line drawing is still required. Currently, step line drawing mainly relies on visual interpretation by the draftsman. This method is cumbersome, has low precision, low automation, and is prone to uneven sampling of point cloud data. In addition, when encountering complex terrain, local features are easily lost, and continuous step lines cannot be fully extracted, resulting in discontinuities in feature lines. At the same time, the lack of feature lines can easily lead to low accuracy of extracted features, which requires more time to perfect. According to specific task requirements, when a model needs to be built quickly in a short period of time, it is impossible to quickly obtain effective feature points, resulting in large errors in the model, making it difficult to achieve timely and accurate construction requirements, resulting in long construction times and difficulty meeting the needs of modern mines. Summary of the Invention

[0005] The present invention aims to provide a regularity-constrained open-pit mine step line extraction method to solve the problem that the existing open-pit mine step line feature point extraction is not timely enough and it is difficult to ensure accuracy.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a regularization-constrained open-pit mine step line extraction method. This method extracts step line feature points by identifying geometric feature surfaces rather than feature points. This method analyzes environmental factors for open-pit mines in different environments to quickly obtain effective and accurate ground points, enabling timely and accurate feature point extraction. Specifically, the method includes the following steps:

[0007] Step 1: perform geometric feature surface recognition on the original point cloud data and extract feature surface data based on the feature surface;

[0008] Step 2: Perform multi-scale supervoxel segmentation on the feature surface data according to the boundary features of the feature surface to form supervoxel units; perform plane classification on the supervoxel units according to the resolution to form a supervoxel set;

[0009] Step 3: Establish a global graph model based on the adjacency relationship of the supervoxel units; perform regularization-constrained surface segmentation on the global graph model according to the energy optimization principle to obtain a face set; select faces with verticality exceeding a threshold as the slope face candidate set;

[0010] Step 4: extract the slope boundary from the slope candidate set, use the extracted slope boundary point cloud as the feature point of the step line, and generate the step line.

[0011] Furthermore, in step 1, it also includes extracting contour feature points from the feature surface data according to environmental factors, and performing point-by-point geometric feature calculations, wherein the geometric features include normal vectors, curvatures, and verticality; if each point is represented in space as

[0012] ;

[0013] Each point and its geometric features are described by an octet:

[0014] ;

[0015] in, is the normal vector, is the verticality, is the curvature.

[0016] Furthermore, in step 2, the multi-scale super-voxel segmentation is to perform voxel segmentation on the feature surface data in a manner of preserving boundary features, which is specifically expressed as follows:

[0017] ;

[0018] Where, is the distance between point cloud data, is the elevation continuity between points, is the spatial similarity between points in the coordinate space; is the normal vector direction similarity; and They are and The normalized weight of .

[0019] Furthermore, in step 2, the plane classification is to divide the supervoxel units into a planar supervoxel set and a non-planar supervoxel set according to the resolution of the initial supervoxel; and within the non-planar supervoxel set, after lowering the resolution of the supervoxel, further divide it into a planar supervoxel set and a non-planar supervoxel set; until all are divided into planar supervoxel sets or the resolution is less than the set threshold resolution.

[0020] Furthermore, in step 2, it is also included to calculate the supervoxel unit The covariance matrix of Get the eigenvalues:

[0021] ;

[0022] Where, Supervoxel unit The center of mass, It belongs to the supervoxel unit All point clouds, Represents supervoxel unit The size of the covariance matrix The eigenvalue of .

[0023] Furthermore, it also includes obtaining a supervoxel unit according to the eigenvalue The significant features include linearity , flatness , divergence :

[0024] , where C is the covariance matrix of the supervoxel unit.

[0025] Furthermore, for the average spacing Point cloud, set the initial resolution of supervoxel segmentation for , the lowered resolution is ’ = , where r des for ; The threshold resolution for .

[0026] Furthermore, it also includes limiting the supervoxel unit by normal vector constraint for non-planar supervoxel set The resolution of , small-scale planar structure is obtained:

[0027] ;

[0028] Where, Supervoxel unit Internal Points The normal vector of Supervoxel unit The normal vector of Represents supervoxel unit The size of Supervoxel unit The vertical difference.

[0029] Furthermore, the standard for plane classification of supervoxel sets is:

[0030] ;

[0031] Where, Indicates the vertical difference of the plane; Expressed as angle threshold; Expressed as the average distance from the point to the fitted plane; It is expressed as the distance threshold.

[0032] Furthermore, in step 3, the energy optimization principle is to minimize the energy function by using the gradient descent method, perform regularity-constrained surface segmentation, obtain the point cloud segmentation result, and form a face set. The energy function is expressed as:

[0033] ;

[0034] Where, is the set of reconstructed normal vectors; It is measured and Energy of matching degree; For points; is the adjacent point of the i-th point; i is the index of the i-th point V.

[0035] The principles and advantages of this solution are:

[0036] Point cloud data in natural terrain scenes is often sampled unevenly. The sparsity and noise of point cloud data can affect feature extraction, resulting in discontinuous step lines. Furthermore, due to the different locations of open-pit mines, they are affected by different environmental factors, which can interfere with the extracted data and affect the validity and efficiency of the data. In complex terrain, the impact of environmental factors can lead to failure or loss of local feature extraction, making it difficult to quickly obtain accurate feature data. This results in large errors and low efficiency in the feature data obtained, making it difficult to meet the demand for rapid data acquisition.

[0037] This solution shifts the focus from identifying step line feature points to identifying geometric feature surfaces, ensuring the integrity and continuity of feature point extraction from the surface structure, and segmenting the contour boundaries of the surface to reduce the amount of calculation. At the same time, it can quickly locate and obtain step line feature points through the contour boundaries, further ensuring the rapid extraction and accuracy of feature points, while ensuring the uniformity of sampling, and solving the problem of step line discontinuity caused by low efficiency and low accuracy in feature point extraction. At the same time, stable and effective feature points are selected according to environmental factors as feature surface data to avoid feature points that are difficult to identify or fail due to factors such as signal occlusion, thereby achieving accurate and rapid data extraction, reducing data volume while ensuring extraction efficiency. The use of multi-scale supervoxels to pre-process the point cloud improves the speed of slope segmentation and extraction and enhances the algorithm's robustness to noise; the global energy optimization model is used to segment the entire open-pit mine point cloud. Compared with the traditional threshold segmentation method, the extraction accuracy is significantly improved, and the slope surface can be segmented more accurately and quickly. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of a regularity-constrained open-pit mine step line extraction method of the present invention;

[0039] Figure 2 The present invention provides a schematic diagram of the step line generation process of a regularity-constrained open-pit mine step line extraction method. DETAILED DESCRIPTION

[0040] The following is further described in detail through specific implementation methods:

[0041] Example 1

[0042] As attached Figure 1As shown, in this embodiment, a regular-constrained open-pit mine step line extraction method is used. The method first analyzes the environmental factors of the open-pit mine environment, resamples the open-pit mine point cloud according to the environmental factors, calculates the normal vector, verticality and curvature of each point, and uses a multi-scale super-voxel segmentation technology that retains boundary features to over-segment the open-pit mine point cloud into super-voxel units. The center point of each voxel is used to inherit all the local features of the voxel and represent the entire voxel block to participate in point cloud segmentation. The voxel adjacency relationship is used to establish the global graph structure of the voxel, and the extended global-l0 algorithm is used to perform energy optimization segmentation on the voxel global graph, thereby obtaining a face set after the surface segmentation of the open-pit mine point cloud. Finally, the slope surface is extracted based on the geometric characteristic that the angle between the point cloud normal vector of the open-pit mine step plate and the vertical plane is significantly different. The concave hull algorithm is used to extract the contour boundary of the segmented slope point cloud to obtain the step line feature points, and the feature points are used to generate the open-pit mine step line. Specifically, it includes the following steps:

[0043] S1, constructing point cloud octet data: extracting ground points from the input original point cloud data to obtain ground point data, performing feature calculation on the ground points point by point, identifying geometric feature surfaces, and extracting feature surface data based on the feature surfaces. The process includes the following sub-steps:

[0044] S1.1, ground filtering

[0045] In this embodiment, the open-pit mine point cloud is first regularly resampled, and the ground points of the open-pit mine point cloud are extracted using the CSF ground filtering algorithm. When extracting the ground points, the surrounding environmental factors are analyzed according to the location of the open-pit mine. In this embodiment, the environmental factors include obstructions, signal obstruction, refractive index, reflectivity and other factors. Based on the environmental factors, the ground points with the least impact and the most complete information acquisition are selected as valid ground points, and the point cloud data of mining buildings and vehicles are eliminated at the same time to improve the efficiency of ground point advance and ensure the accuracy and effectiveness of the data. At the same time, the point cloud is voxel filtered to reduce the point cloud density.

[0046] S1.2, calculate the features of ground points point by point

[0047] In this embodiment, according to the recognition requirements of geometric feature surfaces, the features are mainly geometric features, including normal vectors, curvature and verticality. Specifically, the PCA method is used to calculate the geometric features of each individual point. In this embodiment, each point is represented in spatial coordinates as , then the geometric feature calculation process of each point is as follows:

[0048] Point of Use and The covariance matrix of neighboring points is calculated according to formula (1) ,

[0049] (1). PCA performs eigenvalue decomposition of the covariance of point cloud coordinates. This yields eigenvalues and corresponding eigenvectors. Eigenvalues represent the variance of the data in the direction of the eigenvectors. The calculated eigenvalues are used to calculate various geometric features.

[0050] Each point And its geometric features are described by an octet data, expressed as formula (2), and will be used for the following data processing,

[0051] (2), In the formula is the normal vector, is the verticality, is the curvature, for The main direction of the axis is used to represent the identified feature surface data, facilitating the rapid acquisition of feature point data and ensuring data accuracy and validity. At the same time, the point cloud octet data usually contains a variety of information such as point position, color, and normal, which is used for precise recognition and segmentation, improving segmentation accuracy and robustness.

[0052] S2, generate supervoxel units: perform multi-scale supervoxel segmentation on the point cloud octet data to form supervoxel units.

[0053] Segmenting planes from raw point cloud data of open-pit mine scenes is extremely time-consuming and can contain numerous outliers and noise. Direct segmentation using these points can result in numerous erroneous planes. Therefore, to rapidly obtain accurate plane models from raw point cloud data, this solution first employs a voxel-based segmentation method that preserves boundary features. This approach reduces the number of cells required for computation and mitigates the impact of point cloud density variations and outliers. Consequently, this solution segments the raw point cloud data using multi-scale supervoxels generated by the extended segmentation supervoxel method (NSSS). This over-segmentation of the open-pit mine point cloud into supervoxel units is achieved, with the center point of each voxel inheriting all local features of the voxel and representing the entire voxel block in point cloud segmentation. Specifically, normalized spatial metric segmentation is employed to form supervoxel units, which in turn form a supervoxel set. This ensures accurate slope segmentation, meeting segmentation requirements in open-pit mines of varying mining geology and shapes, and accurately and completely capturing supervoxel units.

[0054] In this embodiment, the normalized spatial metric is used to segment the point cloud octet data. and Used to evaluate the distance between points , vertical angle distance is used to describe the elevation continuity between points , Euclidean distance is used to represent the spatial similarity between points in coordinate space , is the normal vector direction similarity; specifically expressed as:

[0055] (3),

[0056] Where, and They are and Each distance factor has a different value range, so in order to map the distance factors of different distance ranges to the same range, according to and The value range of is scaled to a fixed range using normalized weights to eliminate scale differences and improve segmentation accuracy and generalization.

[0057] Specifically, elevation continuity The spatial similarity is calculated by the following formula (4): The normal vector direction similarity is obtained by calculation using formula (5): Obtained by formula (6).

[0058] (4); (5); (6);

[0059] Where, and is the verticality, and is a space vector, and yes This ensures accurate analysis of subtle features and changes in complex terrain, and ensures the integrity of surface boundaries after supervoxel segmentation.

[0060] At the same time, in this embodiment, in order to map the distance factors of different distance ranges to the same range, according to and The value range of and Normalize; the specific weight setting method is as follows (7), where is the resolution of the current segmentation supervoxel.

[0061] (7);

[0062] Minimum usage point Initialize with the median value of , thereby doubling the fast iteration to improve the segmentation efficiency and ensure the timeliness and effectiveness of feature point acquisition.

[0063] S3, planar classification of the supervoxel units according to the resolution to form supervoxel sets: according to the resolution of the initial supervoxel segmentation, the supervoxel units are divided into planar supervoxel sets and non-planar supervoxel sets; and within the non-planar supervoxel sets, after lowering the supervoxel resolution, they are further divided into planar supervoxel sets and non-planar supervoxel sets; until all are divided into planar supervoxel sets or the resolution is less than the set threshold resolution.

[0064] The multi-scale supervoxel unit is generated by the supervoxel segmentation method of normalized spatial metric. In this embodiment, the supervoxel with adaptive resolution is selected to ensure that the segmented supervoxel can cover most of the plane area. The original point cloud is first segmented into points with a resolution of The supervoxel set is then calculated at the initial resolution It is divided into two types: planar supervoxel sets and non-planar supervoxel sets. It improves the speed of slope segmentation and extraction, and enhances the algorithm's robustness to noise, making slope extraction more accurate.

[0065] Then the supervoxels in the non-planar supervoxel set are further divided into smaller scales individually, and the supervoxel resolution is reduced to ’ = , where r des for . According to the reduced resolution ’ From these non-planar supervoxel sets, further classify planar supervoxel sets and non-planar supervoxel sets, and repeat the above steps until all points are assigned to planar supervoxel sets or the current resolution. Less than threshold resolution In this embodiment, for , for All the divided supervoxel units are grouped into supervoxel sets Specific multi-scale supervoxel set The calculation process is as follows:

[0066] The supervoxel unit is calculated by formula (8) The covariance matrix of :

[0067] (8), Where, Supervoxel unit The center of mass, It belongs to the supervoxel unit All point clouds, Represents supervoxel unit The size of . Thus, the covariance matrix can be obtained The eigenvalue of In this embodiment, the supervoxel unit The centroid is the average of all coordinates. This point may not exist in the point cloud or may even be significantly offset. It is used for feature calculation. The center is the point closest to the centroid in the point cloud and is used for visualization. The centroid can also be used instead of the center. This allows for rapid determination of correlations between supervoxel units, accurately screening for valid features, and reducing computational effort.

[0068] Unlike scenes like urban buildings and interior structures, which are primarily plane-based and have strong directional constraints, the Manhattan model in open-pit mines is characterized by irregular curved surfaces. To segment the open-pit mine point cloud into a larger number of planar voxels, this embodiment employs multiple planar metrics, forming a more stringent segmentation standard. This approach maximizes the over-segmentation of curved surface structures into smaller planes. This achieves precise segmentation of irregular curved surfaces while also prioritizing planes, achieving a precise planar segmentation effect and precise planar segmentation of curved surfaces.

[0069] According to the obtained eigenvalues, the supervoxel units can be calculated by the following formula (9): The significant characteristics of , flatness , divergence ,

[0070] (9), Where C represents the covariance matrix of a single supervoxel unit.

[0071] At the same time, in this embodiment, for the supervoxel unit containing some non-planar points, the supervoxel unit is restricted by applying a normal vector constraint. The resolution is limited by the characteristic that the points in such supervoxel units have different directions. The standard deviation of the angle between the normal of the supervoxel unit and the normal of the points inside the supervoxel unit is calculated using formula (10) to obtain the small-scale planar structure.

[0072] (10),

[0073] Where, Supervoxel unit Internal Points The normal vector of Supervoxel unit The normal vector of Represents supervoxel unit The size of Supervoxel unit By further constraining the resolution, we can improve the accuracy and quality of classification, ensuring that the supervoxel units are divided into planar supervoxel sets as much as possible, achieving accurate segmentation of the surface and ensuring sampling uniformity.

[0074] Since a plane model can be described by using the center of the supervoxel and the normal vector, the residuals of all points can be calculated when fitting the plane model, that is, the inconsistency between the points and the fitting plane can be considered. Distance to the fitting plane , and calculate the residual by averaging the distances of all points , combined with planarity, directional difference and distance residual, the standard for supervoxel plane classification is obtained as follows (11):

[0075] (11), Where, Indicates the vertical difference of the plane; Expressed as angle threshold; Expressed as the average distance from the point to the fitted plane; It is expressed as a distance threshold. Planar means plane and nonplanar means non-planar. The supervoxel unit is judged and classified according to the set conditions, where the judgment condition is flatness. Greater than linearity And flatness Greater than divergence And the vertical difference of the plane The average distance from the point to the fitted plane that is less than the angle threshold Less than the distance threshold If the condition is satisfied, it is classified as a planar supervoxel set; if it is not satisfied, it is classified as a non-planar supervoxel set. In this way, an accurate planar supervoxel set is constructed and an accurate planar model is quickly obtained.

[0076] S4, constructing a set of slope candidate faces: The supervoxel set is connected based on the adjacency of the supervoxel units to establish a global graph model. Based on the energy optimization principle, the global graph model is segmented with regularized constraints to obtain a set of faces. Faces with verticality exceeding a threshold are selected as the set of slope candidate faces. The explicit relationships between intersecting regions and the implicit relationships between non-intersecting regions are enhanced to ensure the merging of major planes and the accurate segmentation of local surface boundaries. The range of slope face extraction is also narrowed to ensure both accuracy and efficiency.

[0077] Each supervoxel unit of the segmentation is a point cloud set with similar local primitive features. In order to map the supervoxel unit to the optimal surface regularization constraint model, this solution transforms the surface segmentation problem into the construction and division problem of the global graph model. By expanding the global-l0 algorithm, considering the L0 regularity between local graph models, and introducing global regularization by using a constructed global regularizer, the explicit relationship between intersecting areas and the implicit relationship between non-intersecting areas are enhanced, thereby improving the merging of major planes and the segmentation accuracy of local surface boundaries, merging over-segmented point cloud voxels into a complete set of faces, which is more convenient for the subsequent rapid identification and extraction of feature points. Specifically, it includes the following sub-steps:

[0078] S4.1. Voxel global graph model construction

[0079] Before segmentation, a graph representing the voxel adjacency structure is first created. There are many existing methods for constructing graphs. Common methods are to derive graphs from neighborhood relationships, such as nearest neighbor graphs, cylindrical neighborhoods, or adaptive neighborhoods, and to use Delaunay triangulation to build graphs. In this embodiment, the method based on neighborhood relationships is used. Unlike neighbor mapping and triangulated network mapping, this solution uses the adjacency of supervoxel units to establish a topological relationship between two supervoxel units, construct a voxel adjacency graph, and thus build a global graph model. This ensures that local voxels can fully establish a direct relationship with candidate merging objects and inhibits cross-region fusion phenomena to ensure accuracy.

[0080] Defined by a set of nodes and a set of edges Basic diagram of composition On each node Represents a supervoxel unit, each edge Represents the spatial relationship between two adjacent nodes i and j, Represents the edges between adjacent nodes This ensures that local voxels can fully establish direct relationships with candidate merging objects and inhibit cross-region fusion phenomena, ensuring rapid acquisition of precise spatial relationships and establishment of accurate topological relationships.

[0081] Specifically, first traverse all voxel points, use the concave hull algorithm to extract voxel edge points, and retrieve the 5 points closest to the edge points to ensure comprehensive and accurate retrieval while reducing the amount of data. If there is a point where the voxel is extracted among the retrieval points, then the voxel is established with the retrieved voxel in an adjacent relationship, and a weight relationship between the voxels is established: .

[0082] S4.2, Surface Segmentation with Regularity Constraints

[0083] The constructed global graph model is segmented. In this embodiment, the extended global-l0 algorithm is used to perform energy optimization segmentation on the voxel global graph model. Specifically, the gradient descent method is used to minimize the energy function shown in formula (12), thereby obtaining the best point cloud segmentation result and obtaining the face set after the surface segmentation of the open-pit mine point cloud.

[0084] (12),

[0085] Where, is the set of reconstructed normal vectors; It is measured and Energy of matching degree; For points; is the adjacent point of the i-th point; i is the index of the i-th point V. At the same time, in formula (12), Used to guide the output normal vector to be as close as possible to the input. Used to constrain sparsity in a local sense, it guides the normal vector of each point to be close to the normals of its neighbors. Used to constrain the connectivity of the surface transition area, parameters , Used to balance the effects of two items. It is used to constrain global sparsity, thereby obtaining the best point cloud segmentation results, ensuring the effectiveness of segmentation, and quickly completing contour feature point extraction.

[0086] S4.3, select the slope candidate set

[0087] The verticality of the surface after the open-pit mine point cloud is segmented is calculated according to formula (13). The slope surface is extracted based on the geometric characteristic that the angle between the point cloud normal vector and the vertical plane of the open-pit mine step plate and slope surface is significantly different. When the verticality of a surface exceeds a given threshold, the slope surface is extracted. When , it is added to the candidate set of the slope to form the candidate set of the slope.

[0088] (13). Threshold The value is based on the design angle of the open pit mine step. Usually the slope angle of the step design is In this embodiment, the threshold Set to .

[0089] S5, generating step lines: extracting slope boundaries from the slope candidate set, using the extracted slope boundary point cloud as the feature points of the step line, and generating the step line.

[0090] In order to ensure the integrity of the extracted slope surface, in this embodiment, the faces with too few point clouds in the candidate set of slope surfaces are eliminated. Due to the irregular curved surface structure of the extracted slope surface itself, the existing boundary extraction algorithm will produce too many erroneous points. Therefore, the slope surface is extracted based on the geometric characteristics that the angle between the point cloud normal vector of the open-pit mine step plate and the vertical plane is significantly different. The extended three-dimensional concave hull algorithm is used to extract the slope surface boundary contour to improve the accuracy of the data. Figure 2 As shown, the specific process is as follows:

[0091] First, any point on the final slope is extracted Perform a spatial k-nearest neighbor search and construct a vector unit with the retrieved neighbor points according to formula (14) ,

[0092] (14);

[0093] Secondly, the ratio of all vectors constructed by superimposing k nearest neighbor points to K It is called the edge coefficient. In order to enhance the adaptability of the algorithm, the experimentally obtained As a judgment indicator of boundary candidate points, it can remove most of the internal points while retaining the edge points, thereby improving the efficiency of edge detection. for The mean of for The variance of It is calculated by the following formula (15):

[0094] (15). When it is less than the judgment index, it is a candidate boundary point, otherwise it is a non-edge point, so as to improve the accuracy of slope boundary contour extraction.

[0095] Again, for the boundary candidate points The specified neighborhood radius of Point cloud Perform RANSAC plane fitting and Project to the fitting plane and use the concave hull algorithm to extract the boundary of the projected point cloud. If the obtained concave hull boundary point cloud set contains ,but is the final boundary point.

[0096] Finally, the extracted slope boundary point cloud is used as the feature points of the step line, and the feature points are used to generate the open-pit mine step line.

[0097] In this embodiment, since open-pit mines often have irregular curved structures, to ensure data validity and simplicity, feature points are identified and extracted on a point-by-point basis when extracting step lines. This overcomes the difficulty of identifying irregular curved surfaces. Surface-by-surface identification is not considered, as this is more difficult for non-planar surfaces and makes it difficult to ensure data accuracy. This solution overcomes technical biases by extracting step line feature points by identifying geometric feature surfaces rather than feature points. The curved surface is then segmented and converted into a planar model. Boundary contour feature points are extracted as step line feature points, effectively resolving the discontinuity of step lines caused by insufficient feature point extraction. By combining the surface segmentation algorithm with the concave hull algorithm, the contours of the characteristic surface are extracted to obtain the step lines; multi-scale supervoxels are used to preprocess the point cloud, which improves the speed of slope segmentation and extraction and enhances the algorithm's robustness to noise; the global energy optimization model is used to segment the entire open-pit mine point cloud. Compared with the traditional threshold segmentation method, the extraction accuracy is significantly improved and the slope surface can be segmented more accurately; this method shows good adaptability in open-pit coal mines of different mining geological conditions and different shapes, and can accurately and completely extract open-pit mine step lines.

[0098] In this embodiment, the problem of discontinuous feature points and low accuracy is overcome by first identifying feature surfaces rather than feature point identification. Generally, for irregular planes, points rather than surface structures are used for identification and extraction. This is because irregular curved surfaces have complex structures, are difficult to identify, and have many data features, making them difficult to obtain accurately and quickly. Compared with point recognition and extraction, surface structures are more commonly used for planes, rather than complex curved surface structures. This solution overcomes conventional technical biases, uses surface feature recognition to obtain feature points, and through a set plane segmentation method, divides irregular curved surfaces into plane models that can be extracted and identified as much as possible, thereby meeting the requirements of surface feature processing and ensuring that the extracted feature points can be accurate and fast, ensuring the continuity of feature points, and avoiding loss problems.

[0099] Example 2

[0100] In this embodiment, the environmental factors of the point cloud data are obtained by feature analysis of the point cloud data, and the validity and stability of the point cloud data are judged based on the environmental factors, thereby resampling the point cloud data. Specifically, in this embodiment, the environmental factors include height, inclination, illumination, refraction angle, reflectivity, etc. The stability of the current point cloud data is judged based on the environmental factors, and valid and stable point cloud data are screened out as contour feature points for the extraction of feature surface data to avoid the point cloud data being unable to fully reflect the characteristics of the current open-pit mine area due to interference from environmental factors, thereby causing feature loss. Ensure that the acquired point cloud data can accurately and stably reflect the step line characteristics to improve the efficiency and accuracy of the extraction.

[0101] In this embodiment, the collected original point cloud data are collected according to the set indicators, which will generate a large amount of point cloud data that is considered valid at the time of collection. However, due to the different locations of open-pit mines, they will be affected by changes in the surrounding environment and the geographical environment, which will cause unstable changes in the point cloud data, causing the point cloud data to differ from the actual situation, making it difficult to accurately reflect the current step line characteristics. For example, when the light refractive index of the current point is low during collection, the point is easy to be identified and can be extracted as a feature point. However, according to environmental analysis, the point may be difficult to be identified or extracted when the refractive index is high. Therefore, it is easy to be missing when it needs to be identified again, resulting in unstable feature points. Therefore, analyzing the stability of point cloud data based on environmental factors can ensure that the identified point cloud data can be accurately and quickly extracted, avoid repeated verification of the validity of the data, improve recognition efficiency, and meet the extraction efficiency when it is necessary to quickly obtain feature points.

[0102] The above is only an embodiment of the present invention, and the common knowledge such as the specific technical solutions and / or characteristics in the solution are not described in detail here. It should be pointed out that for those skilled in the art, without departing from the technical solution of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the description can be used to interpret the content of the claims.

Claims

1. A regularity-constrained open-pit mine step line extraction method, characterized in that: The following steps are included: Step 1: perform geometric feature surface recognition on the original point cloud data and extract feature surface data based on the feature surface; Step 2: Perform multi-scale supervoxel segmentation on the feature surface data according to the boundary features of the feature surface to form supervoxel units; perform plane classification on the supervoxel units according to the resolution to form a supervoxel set; Step 3: Establish an adjacency relationship between the supervoxel set according to the adjacency of the supervoxel units to build a global graph model; perform regular constrained surface segmentation on the global graph model according to the energy optimization principle to obtain a face set; Select the surface with verticality exceeding the threshold as the slope candidate set; The energy optimization principle is to minimize the energy function by using the gradient descent method, perform regularity-constrained surface segmentation, obtain the point cloud segmentation result, and form a face set. The energy function is expressed as: ； Where, is the set of reconstructed normal vectors; It is measured and Energy of matching degree; For points; is the adjacent point of the i-th point; i is the i-th point, which is the index of V; Used to guide the output normal vector close to the input; Used to constrain sparsity in a local sense, it guides the normal vector of each point to be close to the normals of its neighbors; Used to constrain the connectivity of the surface transition area, parameters , Used to balance the effects of two items; Used to constrain global sparsity; Step 4: extract the slope boundary from the slope candidate set, use the extracted slope boundary point cloud as the feature point of the step line, and generate the step line.

2. The open-pit mine step line extraction method with regularity constraints according to claim 1, characterized in that: In step 1, it also includes extracting contour feature points from the feature surface data according to environmental factors and performing point-by-point geometric feature calculations, wherein the geometric features include normal vectors, curvatures, and verticality; if each point is represented in space as ； Each point and its geometric features are described by an octet: ； in, is the normal vector, is the verticality, is the curvature.

3. The open-pit mine step line extraction method with regularity constraints according to claim 1, characterized in that: In step 2, the multi-scale super-voxel segmentation is to perform voxel segmentation on the feature surface data in a manner of preserving boundary features, which is specifically expressed as follows: ； Where, is the distance between point cloud data, is the elevation continuity between points, is the spatial similarity between points in the coordinate space; is the normal vector direction similarity; and They are and The normalized weight of .

4. The open-pit mine step line extraction method with regularity constraints according to claim 1, characterized in that: In step 2, the plane classification is to divide the supervoxel units into planar supervoxel sets and non-planar supervoxel sets according to the resolution of the initial supervoxel; and in the non-planar supervoxel set, after lowering the resolution of the supervoxel, further divide it into planar supervoxel sets and non-planar supervoxel sets; until all are divided into planar supervoxel sets or the resolution is less than the set threshold resolution.

5. The open-pit mine step line extraction method with regularity constraints according to claim 4 is characterized in that: In step 2, it also includes calculating the supervoxel unit The covariance matrix of Get the eigenvalues: ； Where, Supervoxel unit The center of mass, It belongs to the supervoxel unit All point clouds, Represents supervoxel unit The size of the covariance matrix The eigenvalue of .

6. The open-pit mine step line extraction method with regularity constraints according to claim 5, characterized in that: It also includes obtaining supervoxel units based on eigenvalues The significant features include linearity , flatness , divergence : , where C represents the covariance matrix of the supervoxel unit.

7. The open-pit mine step line extraction method with regularity constraints according to claim 4, characterized in that: For the average spacing Point cloud, set the initial resolution of supervoxel segmentation for , the lowered resolution is , where for ; The threshold resolution for .

8. The open-pit mine step line extraction method with regularity constraints according to claim 7, characterized in that: Also includes the ability to constrain supervoxel units by using normal vector constraints for non-planar supervoxel sets. The resolution of , small-scale planar structure is obtained: ； Where, Supervoxel unit Internal Points The normal vector of Supervoxel unit The normal vector of Represents supervoxel unit The size of Supervoxel unit The vertical difference.

9. The open-pit mine step line extraction method with regularity constraints according to claim 6, characterized in that: The criteria for plane classification of supervoxel units are: ； Where, Indicates the vertical difference of the plane; Expressed as angle threshold; Expressed as the average distance from the point to the fitted plane; It is expressed as the distance threshold.

Images