Rock wall linear structure information extraction from point cloud data and linear structure measurement method and system

Abstract

The present application discloses a method and system for extracting and measuring linear structure information of rock wall from point cloud data. In view of the problem that the prior art does not solve the whole process of extracting linear structure information of rock wall from point cloud data, the present application provides a method and system for extracting and measuring linear structure information of rock wall from point cloud data. The technical scheme fully utilizes three-dimensional color point cloud information of rock wall, comprehensively uses DBSCAN clustering, improved region growing algorithm, graph model, minimum spanning tree and other methods, establishes an automatic processing scheme of rock wall point cloud based on collaborative analysis of geometric features and color information, accurately identifies linear structure in the point cloud, and directly obtains measurement data through quantization, which is a point cloud automatic and high-precision processing scheme from field original collection to measurement data output. The present application provides a new idea of understanding rock wall structure from color information. The present application can be applied to engineering design and construction safety assessment, and can improve the response speed and accuracy of geological disaster warning.

Description

Technical Field

[0001] This invention relates to a method and system for extracting linear structural information of rock walls and measuring linear structures from point cloud data. In particular, it relates to a technology for directly extracting linear structural information of rock wall surfaces from three-dimensional point cloud data and completing linear structural measurements. This technology belongs to the fields of measurement, image and video recognition or understanding of irregular surfaces or contours. Background Technology

[0002] Linear structures refer to all linear structural elements within a geological body, serving as crucial evidence for studying rock deformation history and tectonic environments. Interpreting linear structures reveals important information such as rock deformation mechanisms, tectonic stress fields, and tectonic evolution history. While providing information about geological strata, linear structures also represent potential geological hazards and important reservoir spaces for oil and gas. Therefore, investigating linear rock structures is a fundamental and vital component of geological environmental surveys, resource exploration, and hazard monitoring.

[0003] The technology of combining a LiDAR sensor with an RGB camera with a drone (UVA) can acquire high-precision, high-resolution three-dimensional color point cloud data of the survey area. This greatly overcomes the environmental obstacles of collecting detailed information on large-area rock wall surfaces. It can acquire detailed information on a specified rock wall surface, including geometric shape and color information, under conditions that are almost unrestricted by terrain. This provides data possibilities for the identification and interpretation of linear structures on rock wall surfaces, but corresponding data processing technology is still needed.

[0004] The existing technology "Digital Identification of Rock Mass Structures and Characterization of Mechanical Parameter Variability and Its Engineering Application" (Yi Xiaoyu, Chengdu University of Technology, 2022) discloses a method for identifying rock mass structural surfaces from point cloud data using a multi-rule region growing algorithm based on colored point cloud data of rock masses. The method uses the Kd-tree algorithm to organize the point cloud data, principal component analysis to calculate the point cloud normal vector and curvature, and a multi-rule region growing algorithm to extract structural surfaces and output their attitude. However, because this study believes that "quantitative analysis of the gap width of rock mass structural surfaces using images is more convenient than quantitative analysis using point cloud data, it uses rock mass outcrop image data to quantify the gap width of rock mass structural surfaces." This method represents the concept of existing technical solutions for interpreting rock mass geological structures using point cloud data; that is, it only solves the technical problem of identifying and extracting structural surfaces and their attitude information from point cloud data, but does not solve the technical problem of identifying and extracting fracture structures from point cloud data, the latter of which is solved using two-dimensional image recognition methods.

[0005] Chinese patent application 2024116249450 discloses a data risk analysis method based on the Internet of Things (IoT), which solves the technical problem of obtaining slope crack information of a target mining area based on a three-dimensional model of the mining area and using point cloud computing. The method consists of three parts: first, identifying the initial crack point cloud region; second, using a growth algorithm to identify and filter out non-crack point clouds in the initial crack point cloud region to obtain the target crack point cloud region; and finally, calculating the slope crack information of the target mining area based on the two-dimensional coordinates of the target crack point cloud. To identify target crack point cloud regions, the specific implementation steps are as follows: First, the point cloud data is fitted to a surface to obtain the slope fitting surface corresponding to the slope region; second, the residual value between the slope point cloud and the slope fitting surface is calculated, and the slope point clouds with residual values ​​greater than a preset first threshold are integrated into suspected crack point cloud regions; third, for each suspected crack point cloud region, a combination of a normal vector angle threshold and a preset second residual threshold is used as a screening condition to confirm the region edge point cloud; fourth, the average curvature of all region edge point clouds is calculated, and if the average curvature is positive, the suspected crack point cloud region is taken as the initial crack point cloud region; finally, a region growing algorithm is used to filter out non-crack point clouds in the initial crack point cloud region to obtain the target crack point cloud region. The main drawbacks of this method are: 1. Repeatedly calculating the point cloud residual value and curvature and comparing them with the corresponding preset thresholds to gradually screen out slope crack point clouds; the method has a long process and many steps. Second, to identify and screen fracture point clouds, after acquiring the point cloud data of the target mining area, it is first necessary to perform a large number of surface fitting calculations using the point cloud data to construct a three-dimensional model of the mining area. Then, from the three-dimensional model, the slope fitting surface, suspected fracture point cloud area, initial fracture point cloud area, and target fracture point cloud area are determined layer by layer, resulting in a large amount of computation. Third, although the original point cloud is a colored point cloud, the identification calculation process only utilizes the geometric information of the point cloud, wasting the color information in the point cloud data. Fourth, when using the region growing algorithm to screen non-fracture point clouds, the traditional region growing algorithm is used, which relies on the angle between the normal vectors and the curvature threshold for judgment. This method is prone to overgrowing or undergrowing problems in scenarios with complex rock surface textures and uneven lighting. Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a technical solution for directly extracting linear structural information of rock walls from rock wall point cloud data, and a technical solution for using this method to perform engineering quantitative analysis of linear structures of rock walls to obtain linear structural measurement data.

[0007] To achieve the above objectives, the present invention first provides a method for extracting linear structural information of rock walls from point cloud data.

[0008] A method for extracting linear structural information of rock walls from point cloud data. Step S100: Use a drone to acquire the original three-dimensional point cloud of the rock wall, and obtain the preserved three-dimensional point cloud after preprocessing. The preprocessing includes voxel downsampling. Step S200: Based on the preserved 3D point cloud, perform cluster analysis using the DBSCAN algorithm in the RGB color space to extract outliers during the clustering process and obtain the initial linear structure region. R initial ; Step S300, Initial identification of the linear structure region R initial Extract the geometric properties of the rock wall and identify the initial seed point. P ps Obtain the initial linear construction point set F 0, including, Step S310: Through local curvature feature analysis, the initial identification of the linear structure region is achieved. R initial Filtering out those with curvature less than the curvature threshold k 1 The points are used as feature points of the concave region to identify and extract the concave region; Curvature threshold k 1 The determination method is as follows: calculate the initial identification region of the linear structure. R initial The curvature of all points within the area is calculated and a frequency histogram is plotted. Local minima are then used to determine the local minimum values. k 1; Step S320: Based on the point cloud data of the concave region, search for the nearest neighbors of any point and filter out the angle between the normal vectors of any point and all its neighbors. i All are greater than the included angle threshold i Points with a value of 1 are identified as initial seed points. P ps 30º≤ i 1≤50º; Step S330: Construct an initial linear construction point set. F 0, set the initial seed point P ps deposit F 0; Step S400, Initial identification of the linear structure region R initial Extracting linear structural regions R linear ,include, Step S410, Create a seed point queue Q ,Will F 0 Midpoint Push In Q , Q It is a first-in, first-out dynamic queue; Step S420: Read the seed point queueQ Points in P s Determined by the system of equations 1 P s Local linear structure growth direction , Equation 1-1 Formula 1-2 In the formula, Seed point P s The normal vector, - Global reference vector; Step S430: Construct seed points according to the system of equations 2. P s Introduction to Linear Structures R initial Candidate point directional search area R search , Equation 2-1 Equation 2-2 Equation 2-3 In the formula, R search - Candidate point targeted search area, P - Introduction to Linear Structures R initial any point in, w - R search Width parameter, q - Coefficient, with a value of 1.2 ≤ q ≤1.5, w 0 – Expected width of the linear structure. k - Coverage factor, 2≤ k ≤5, d voxel - Voxel downsampling size during the preprocessing stage of raw 3D point cloud data; Step S440: Create an empty linear construction point set. F ; In the candidate point directional search area R search Inside, search for seed points P s nearest neighbor Pj ,like P j Meeting the growth conditions and not having been pushed into the seed point queue Q Identified as a seed point P s ,join in F And push in Q Otherwise ignore this P j Continue searching and identifying the next one. P j The growth conditions are expressed according to Formula 3. Formula 3 In the formula, k j - Neighboring points P j curvature, k grow - Curvature growth threshold, k 1< k grow <0.06, i sj Seed point P s and neighboring points P j The difference in the angle of the normal vector i grow - Normal vector angle difference growth threshold, 30º < i grow < i 1; Step S450: Repeat steps S420 to S440 and perform iterative calculations until the growth termination condition is met. The growth termination condition is that the algorithm exhausts all possibilities starting from the seed point. P s Starting from a candidate region that is spatially connected and satisfies geometric constraints, the growth front achieves complete closure; After the growth operation terminates, the linear structure point set F This refers to the extracted linear structural region. R linear .

[0009] The main purpose of the above-mentioned method for extracting linear structural information from point cloud data is to construct an automated and high-precision processing flow from the original point cloud of the rock face to the extraction of linear structural information on the rock face. By fully utilizing the multi-dimensional information such as color and coordinates in the color point cloud information, it gradually solves the problem of initially identifying linear structural areas by clustering in the color space. R initialFrom using multi-dimensional geometric features to identify initial seed points P ps Then, the final linear structural region is extracted using an improved region growing algorithm. R linear Technical issues.

[0010] This invention also provides the following optimized solution for the method of extracting linear structural information of rock walls from the aforementioned point cloud data. The optimized concept improves the technical solution from multiple technical perspectives, and can be implemented individually while maintaining technical independence, or simultaneously without creating technical logical conflicts.

[0011] In optimization 1, step S200, the DBSCAN clustering process is used to perform "reverse filtering" on the 3D point cloud of the rock wall in the color space to identify and extract outliers in the clustering process, and to extract the initial identification area of ​​the linear structure of the rock wall. R initial .

[0012] Previous research in this invention revealed that linear structures on rock face surfaces (such as joints and cracks) often exhibit significant color differences from the surrounding intact rock face, which can be reflected in the color information of point clouds. Therefore, the color information in point clouds can provide accurate differentiation beyond coordinate information for different rock face structures, becoming a better tool for interpreting rock face surface structural features than three-dimensional coordinate information. Furthermore, density-based spatial clustering algorithms (DBSCAN clustering algorithm) can enhance the discriminative power of color information in rock face point clouds, making full use of it. Based on this observation, this invention introduces the DBSCAN clustering algorithm into the color space of the point cloud, dividing the point cloud based on color similarity and spatial density during the DBSCAN clustering process. Further analysis of experimental data revealed that during DBSCAN clustering, point clouds in intact rock face regions typically exhibit uniform color and dense point cloud distribution, forming stable core clusters. In contrast, point clouds in areas with linear structures are identified as outliers (noise) by the DBSCAN algorithm due to their significant color differences and relatively sparse distribution in the color space. The entire process is concise and efficient. Based on these findings, this invention utilizes the DBSCAN clustering process to perform "reverse filtering" on the preserved 3D point cloud of the rock face in the color space, identifying and extracting outliers during the clustering process, thus solving the problem of extracting the initial identification area of ​​linear structures on the rock face. R initial Technical issues related to point clouds.

[0013] Specifically: First, the RGB value of each point is normalized to the [0,1] interval to obtain a three-dimensional color feature vector; subsequent clustering is based solely on the color space, and the distance between color components is measured using three-dimensional Euclidean distance. Second, an adaptive neighborhood radius is set.r M points are randomly selected from the preserved 3D point cloud (M is determined based on the total size of the preserved point cloud and the system's computing power, typically 5000-10000 points to effectively control the complexity of distance calculation while ensuring the representativeness of color statistical features) to form a sample set. The distances between color components of all point pairs within the sample set are calculated to obtain a distance set. A frequency histogram H(d) of the distance set is plotted using the group interval Δd (Δd can be set empirically to a fixed step size such as 0.02-0.05, or dynamically determined by dividing the maximum color distance by the preset total number of histogram intervals such as 50-100). The main peak of the histogram is located, and the first local minimum point is searched to the right of the main peak. d val (i.e. H( d val (less than the values ​​of the left and right adjacent intervals), take r = d val Use the global neighborhood radius as the reference; set the minimum number of neighborhood points, MinPts. Finally, use the above parameters. r Compared with MinPts, DBSCAN clustering is performed on all points in the color space, and points labeled as noise (label-1) are extracted as linear initial recognition regions. R initial Dotted clouds.

[0014] Optimization 2: Provides a curvature threshold. k 1. Calculation scheme.

[0015] In step S310, in the local curvature feature analysis method, the curvature threshold is used as a control condition. k 1. Generally, the value can be determined by combining empirical values ​​with trial calculations using a small sample size. This invention utilizes a frequency histogram to determine local minima. d val take it as k The value is 1, specifically constructing the quantization function model of this method, expressed according to the system of equations 4. In equation 4, H ( k - The frequency histogram function of the point cloud curvature distribution. k speak - The curvature value corresponding to the maximum frequency in the frequency histogram. k i - Discrete curvature values ​​of the frequency histogram interval.

[0016] Equation 4-1 Equation 4-2 Optimization 3 provides k grow , i grow Experience-based constraints.

[0017] In step S440, k grow , i grow They satisfy equations 5 and 6 respectively.

[0018] Formula 5 Formula 6 Optimization 4, the method for extracting linear structural information of rock walls from point cloud data also includes step 500, which solves the problem of extracting information from linear structural areas. R linear Extracting linear construction paths P linear Technical issues.

[0019] Step S500, from the linear structural region R linear Extracting linear construction paths P linear ,include, Step S510: Based on the linear construction point set obtained in step S450 F The DBSCAN clustering algorithm was used to construct the linear region. R linear The Euclidean distance between each point in the interior will R linear Point cloud clustering and grouping yields segmented, independent linear structural clusters. C i ; Step S520, for any linear structural cluster C i Convert it into a weighted graph model G i ( V i , E i ),Depend on C i Vertices are directly generated from points within the vertices. V i Established through Delaunay triangulation C i Topological connections between interior points define edge sets. E i Then, the Euclidean distance between points is used as the weight of the corresponding edges in the graph. w ij ; Step S530, from the weighted graph model G i Extract the minimum spanning tree and the longest path as the initial path for linear construction. Pinitial ; Step S540: Initial identification of the linear structure path P initial Smoothing process is performed to obtain linear structural paths. P linear .

[0020] Optimization 5: In step S540, the initial path of the linear structure is identified. P initial During the smoothing process, when there are unnatural jitters or sharp corners in the path due to noise, the following two correction and optimization mechanisms can be added, either one or both, to further improve the quality of the linear path.

[0021] Curvature constraint optimization mechanism. Specifically, the curvature of each point on the path is calculated. If the curvature exceeds a certain level (e.g., exceeding a preset curvature threshold, or exceeding three times the average level of the surrounding area), the system identifies that point as a curvature abrupt change point. For curvature abrupt changes, a local position fine-tuning method is used for smoothing, that is, the coordinates of the curvature abrupt change point are appropriately moved closer to the middle position of its adjacent points, straightening out minor jitters and avoiding unnatural excessive distortion of the path.

[0022] A redundancy detection and optimization mechanism is implemented. Specifically, redundant points in the path that do not contribute to describing the path's shape are detected and deleted, provided the path shape remains unchanged. Redundancy point identification can employ a straight-line deviation method. Specifically, the midpoint of three adjacent points on the path is examined, and its positional relationship with the line connecting the midpoint and the adjacent points is determined. If the distance from the midpoint to the line connecting the midpoint and the adjacent points is less than a preset small distance (this small distance can be determined based on the specific specifications of the linear structure, such as a preset 0.005m or 0.01m), it indicates that the midpoint and the adjacent points are actually on a straight line. Deleting it will not affect the path's shape; therefore, it is marked as a redundant point and removed. This redundancy detection and optimization mechanism preserves key inflection points and shape features of the path, avoids jitter or sharp corners without practical technical significance, and effectively reduces unnecessary data volume, saving computational resources.

[0023] In technological research and development and engineering practice, one of the ultimate goals of extracting linear structural information from rock face using point cloud data is to conduct quantitative engineering analysis of the linear structural information on the rock face, output measurement data of the linear structural information, and then further analyze and utilize it. Therefore, this invention also provides the following technical solution for measuring linear structural information on rock faces, solving the technical problem of directly measuring linear structural information on rock faces using point cloud data.

[0024] A method for measuring linear rock wall structures, implemented using the aforementioned method for extracting linear rock wall structure information from point cloud data, includes: First, determine the rock wall surface to be measured; Secondly, the method for extracting linear structural information of rock walls from the point cloud data is implemented, and the linear structural area is output at the end of step S400. R linear Data and / or output linear construction path at the end of step S500 P linear data; Finally, regarding the linear structural region R linear and / or linear structural paths P linear Conduct engineering quantitative analysis to obtain corresponding measurement data.

[0025] The present invention also provides the following solution for measuring linear rock wall structures.

[0026] A rock wall linear structure measurement system includes a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the above-described rock wall linear structure measurement method.

[0027] Compared with the prior art, the beneficial effects of the present invention are: (1) In order to realize the direct extraction of linear structural information of rock wall surface using rock wall point cloud data, the present invention provides a new method of "reverse filtering" in point cloud color space. This method performs DBSCAN clustering operation on the three-dimensional color point cloud of rock wall in color space, and uses the noise points in the clustering process as the basis for subsequent point cloud analysis. "Reverse filtering" regards the color information of color point cloud as the key distinguishing tool for understanding the geological features of rock wall surface, and establishes a point cloud interpretation framework that coordinates color similarity and spatial density. It is a brand-new technical concept for understanding rock wall structure with the help of point cloud information. Based on this, the technical solution further provides an improved region growth algorithm. By simulating the natural extension characteristics of linear structures in three-dimensional space, the local linear structure growth direction of the seed point is first calculated, and then the linear growth space along the growth direction is calculated to determine the candidate point orientation search area. The spatial conditions (belonging to the nearest neighbor) and geometric conditions (curvature conditions and normal vector angle conditions) are coupled and utilized to construct the growth conditions of the dual verification mechanism. The discrete candidate points are connected into a complete linear structure region, realizing the intelligent connection and expansion of the linear structure region. This effectively overcomes the defect of the traditional region growth algorithm in maintaining the continuity of linear structures and solves the technical problem of over-growth or under-growth in scenarios with complex rock surface texture and uneven lighting. (2) In the optimization scheme, the present invention provides a precise quantization processing scheme of "reverse filtering" in the color space based on a mathematical model. (3) In the linear structure path extraction scheme based on point cloud, a scheme combining DBSCAN clustering, graph model and minimum spanning tree is provided so that the extracted linear structure path can maintain the complete topological structure of the linear structure path network, highly restore the natural form, and is beneficial to the processing of complex linear structures. (4) This invention also provides a measurement system using a rock wall linear structure measurement method. (5) Through optimization, this invention provides a technical solution for automated analysis and processing of rock wall linear structures based on three-dimensional color point clouds of rock walls and fully interpreting the geometric and color features in the point cloud information. This solution can accurately identify and extract linear structure information from the color point cloud and connect with automated analysis tools for engineering quantification to obtain key indicator measurement data of the linear structure area and path of the rock wall. This achieves a fully automated and high-precision complete processing solution within the point cloud computing framework, from raw field acquisition to measurement data output. The technical solution expands the integrated tool framework for understanding color point clouds of rock walls, overcoming the technical problems of existing technologies that can only use image processing to extract linear structures from rock wall surfaces, which wastes point cloud information, has detection bottlenecks in complex curved surfaces and noisy environments, and increases computational complexity. This solution can provide important support for engineering design, construction safety assessment, and other aspects, and can further improve the response speed and information accuracy of geological disaster early warning systems. Attached Figure Description

[0028] Figure 1 This is a schematic diagram illustrating the main steps of extracting and measuring linear structural information of rock walls from point cloud data.

[0029] Figure 2 It is a colored dot cloud of the target rock area.

[0030] Figure 3 This is a preliminary map of the linear structure.

[0031] Figure 4 It is a map of a linear structural region.

[0032] Figure 5 It is a smooth linear path diagram.

[0033] Figure 6 This is a comparison chart of the quality of the original data.

[0034] Figure 7 It is a line structure diagram extracted by a two-dimensional image recognition method.

[0035] Figure 8 This is a comparison chart showing how lighting conditions affect the recognition results. Detailed Implementation

[0036] The preferred embodiments of the present invention will now be further described with reference to the accompanying drawings.

[0037] Figure 1 This is a schematic diagram illustrating the main steps of extracting and measuring linear structural information of rock walls from point cloud data.

[0038] Example 1

[0039] like Figure 1~Figure 4 As shown, the method of this invention is used to extract linear structural regions from point cloud data of rock face linear structures. R linear .

[0040] 1. Acquisition and preprocessing of 3D point cloud data A drone equipped with a LiDAR sensor and a high-resolution RGB camera flies over the target rock wall area to simultaneously acquire color point cloud data containing three-dimensional coordinates and color information, and stores it as the original three-dimensional point cloud.

[0041] Preprocessing the original 3D point cloud improves the quality of the point cloud data, thereby enhancing the efficiency of subsequent processing steps. In this example, the preprocessing method employs voxel downsampling of the point cloud and uses voxel grid filtering for uniform downsampling to reduce redundant data and achieve a controllable balance between computational efficiency and feature preservation.

[0042] The original 3D point cloud is preprocessed to obtain a preserved 3D point cloud.

[0043] In this example, the original 3D point cloud collected by the UAV contained approximately 300,000 points, corresponding to a file size of about 22MB and an average point density of about 1,000 pts / m². To balance the preservation of spatial geometric features and algorithm processing efficiency in subsequent calculations, the preprocessing downsampling voxel size was set to 0.05m, and the centroid of each voxel was selected as the output during voxel mesh filtering, with its color and intensity attributes taken as the average value within the voxel. After preprocessing, approximately 85,000 3D point clouds were retained, the average point density was reduced to about 280 pts / m², and the data volume was compressed by about 71.6%.

[0044] Figure 2 It is a colored dot cloud of the target rock area. Figure 2 (a) Original 3D point cloud, Figure 2 (b) preserves the three-dimensional point cloud.

[0045] 2. Extracting the initial identification area of ​​linear structures R initial Based on the previous research findings of this invention regarding the specific technical significance of point cloud color information in rock wall structure, this invention focuses on preserving point cloud data, using the DBSCAN algorithm to extract color outliers in the RGB color space, and implementing an optimized scheme of inverse filtering within the color space.

[0046] First, the RGB values ​​of each point are normalized to the [0,1] interval to obtain a three-dimensional color feature vector; subsequent clustering is based solely on the color space, and the distance between color components is measured using three-dimensional Euclidean distance. Second, an adaptive neighborhood radius is set. r M = 5000 points are randomly selected from the preserved point cloud to form a sample set. The distances between the color components of all point pairs in the sample set are calculated to obtain the distance set. A frequency histogram H(d) of the distance set is plotted with a class interval Δd = 0.025 (dynamically determined by dividing the maximum color distance in the sample set by the preset total number of histogram intervals of 50) to locate the main peak. d speak Search for the first local minimum point on the right side of the main peak. d val ,Pick r = d val As the global neighborhood radius; set the minimum number of neighborhood points MinPts (using the D+1 rule, preferably 2D as a robust heuristic threshold). Finally, use the parameters... r Compared to MinPts, DBSCAN clustering is performed on all points in the color space, and points labeled as noise (label-1) are output as linear initial recognition regions. R initial Point cloud. To automate the algorithm's processing, a system of 7 equations is constructed as follows.

[0047] 3D color feature vector D i The function is as shown in equation 7-1, where, r i , g i , b i These are the normalized red, green, and blue channel values, respectively.

[0048] Equation 7-1 Frequency histogram main peak d speak The positioning function is shown in Equation 7-2.

[0049] Equation 7-2 First local minimum point d val The calculation function is shown in Equation 7-3 below.

[0050] Formula 7-3 any point P i exist r Neighborhood point set within the radius N r ( P i density scale As shown in equation 7-4, where, This is an indicator function; it takes the value 1 if the condition is true, and 0 otherwise.

[0051] Equation 7-4 The core function is shown in Equation 7-5.

[0052] Formula 7-5 The outlier function is shown in Equation 7-6.

[0053] Formula 7-6 Introduction to Linear Structures R initial Calculation examples are shown in Table 1 below.

[0054] Table 1 R initial Relevant data (partial)

[0055] Figure 3This is a preliminary map of linear structural areas. Red dots in the map represent points where linear structural areas have been identified, while the remaining unmarked points represent intact rock face areas.

[0056] The original 3D point cloud was downsampled using voxels and then clustered using DBSCAN to initially identify regions from a linear structure. R initial The target features were effectively separated and extracted from the original 3D point cloud, achieving 85.3% of the target features.

[0057] 3. Extracting the geometric characteristics of linear structures on the rock wall Initial understanding of linear structures R initial Multidimensional features in point clouds serve as key geometric features of linear structures, providing a quantitative basis for accurate identification of subsequent linear structures.

[0058] 3.1 Filtering Feature Points in Depression Areas Using local curvature as the first-dimensional feature, the depression region containing the linear structure is identified. The local curvature feature analysis method is employed to initially identify the region of the linear structure. R initial Filtering points from the point cloud with curvature less than the curvature threshold k Points with a value of 1 are used as feature points of the concave region. These feature points are key characteristics for quantitatively identifying concave regions containing linear structures. All identified feature points of the concave region constitute the concave region.

[0059] This example introduces the linear structure. R initial In the point cloud, using the model in Equation 4, we obtain... k 1 ( k The minimum value is 1 = 0.04. k i The value, i.e., the breakpoint in the distribution of curvature value data. (Based on...) k A 1=0.04 ratio was used to filter out the recessed areas, achieving a recall rate of 80.3%. Relevant calculation data are shown in Table 2 below.

[0060] Table 2 k 1. Calculate relevant data

[0061] 3.2 Selecting initial seed points P ps Based on point cloud data of the depression region, the angle between the normal vectors is used as the second-dimensional feature to identify the directional changes of linear structures on the rock wall surface. Regions where the direction of the angle between the point cloud normal vectors changes significantly are detected as significant markers of linear structures, and initial seed points are obtained through screening. P ps .

[0062] In the normal vector angle analysis, the neighborhood parameter is set to an effective scale that covers the local geometric features (preferably 8–15, set to 10 in this example). A point cloud nearest neighbor search method is used to process the point cloud of the concave region, searching for the nearest neighbors of the feature points in the concave region, and selecting the normal vector angle between all nearest neighbors. i All are greater than the included angle threshold i The feature points of the concave region of 1 are used as the initial seed points. P ps Based on the present invention and preliminary analysis and testing, 30º≤ i 1 ≤ 50º. Within this interval, i 1. The value can be adjusted based on the specific structure of the linear structure on the target rock wall surface.

[0063] In this example, based on the surface texture characteristics of the target rock wall and the results of previous tests, the following settings were established. i 1 = 42º. Under this high threshold constraint, the algorithm selects only about 6.7% (468 points) of the feature points in the approximately 6,900 concave regions as initial seed points. P ps Upon examination, these points are mainly located in abrupt transition zones or deep inflection points at the edges of linear structures, exhibiting strong geometric directional indicativeness and high confidence. They can effectively guide subsequent region growing algorithms to extend into connected regions without causing erroneous growth in non-structured regions.

[0064] 3.3 Constructing the initial linear construction point set F 0 Construct an initial linear set of construction points F 0, set 468 initial seed points P ps deposit F 0.

[0065] 4. Extracting linear structural regions R linear Obtaining the initial seed point P ps Based on this, an improved region growing algorithm is used to construct the initial identification region from a linear structure. R initial Extracting linear structural regions from point clouds R linear .

[0066] 4.1 Creating a Seed Point Queue Q With access token array Create seed point queue Q The initial linear construction point set F 0 Midpoint Push In Q , Q It is a first-in, first-out dynamic queue.

[0067] To ensure the completeness, accuracy, and efficiency of linear structure detection through a stable breadth-first search strategy in subsequent iterative growth operations, this example uses a dynamic first-in-first-out (FIFO) seed point queue. Q Manage seed points. First, set... F Initial seed point in 0 P ps Push the point into the seed point queue Q During the growth operation, new seed points that pass the verification are continuously added to the queue. Q The system calculates the "growth potential" of all seed points by sequentially reading them from the head of the queue. The system also maintains a seed point queue. Q Access marker array, recording Q Check whether each point has been processed to avoid the same point being repeatedly pushed into the queue due to being added to the calculation of different adjacent points, which could lead to duplicate calculations or infinite loops.

[0068] 4.2 Determine the seed point P s Local linear structure growth direction

[0069] From seed point queue Q Read sequentially P s The model is determined using the system of equations 1. P s Local linear structure growth direction In Equation 1, a global reference vector is introduced. (A pre-defined vector that is not normal to the seed point) Parallel fixed 3D vectors are used, and the geometric orthogonality of the vector cross product is utilized to force the growth direction to always be perpendicular to the normal vector, thus strictly constraining the growth search path within the local tangent plane at the seed point; simultaneously, by leveraging... The global direction is constant, eliminating the rotational uncertainty in the selection of the in-plane tangent vector, thereby jointly ensuring the consistency between the growth direction and the local surface geometry.

[0070] 4.3 Constructing the Candidate Point Directional Search Area R search Using seed points P s Determined growth direction of local linear structures The initial understanding of the linear region is constructed based on the system of equations 2. R initial Candidate point directional search area in point cloud R search . Rsearch Therefore P s Centered on, along the direction An extended linear space.

[0071] w It is a parameter related to the density characteristics of the original 3D point cloud, which controls the lateral range of linear structure growth with a fixed value throughout the point cloud growth process after system initialization. w and The jointly constrained directional search strategy not only effectively limits the growth range and avoids erroneous expansion in irrelevant areas, but also maintains the natural trend of linear structure extension, making it particularly suitable for the detection of curved linear structures. Based on the preliminary analysis and testing of this invention, w Can be used w 0 Confirmed.

[0072] Candidate point targeted search area R search Calculation examples are shown in Table 3 below.

[0073] Table 3 R search Relevant calculation data (partial)

[0074] 4.4 Constructing a linear set of construction points using a region growing algorithm F Create an empty linear construction point set. F At the seed point P s Candidate point directional search area R search Within, search for its nearest neighbors. P j Combine the access marker array to perform a status query, if P j Meeting the growth conditions (expressed according to Formula 3) and not having been pushed into the seed point queue Q (This is manifested as not being marked as visited in the access marker array), and is identified as a seed point. P s Add linear construction point set F And push it into the seed point queue. Q At the tail, its status in the access marker array is updated to "visited"; otherwise, the nearest neighbor is ignored. P j Continue searching to identify the next nearest neighbor. P j .

[0075] By utilizing a dual verification mechanism comprised of spatial conditions (belonging to nearest neighbors) and geometric conditions (curvature condition and normal vector angle condition), the extracted linear construction point set is guaranteed to be valid. F The accuracy.

[0076] 4.5 Iterative Calculation and Growth Termination Repeat "4.2 Determine the seed point" P s Local linear structure growth direction "~"4.4 Constructing a linear set of construction points using a region growing algorithm F "Perform iterative calculations until the growth termination condition is met. The growth termination condition is that the algorithm exhausts all possibilities starting from the seed point." P s Starting from a candidate region that is spatially connected and satisfies geometric constraints, the growth front is completely closed. After the growth operation terminates, the extracted linear construction point set is obtained. F The maximum connected component under the current constraints has been reached, completing the extraction of this independent linear structure. F This refers to the extracted complete linear structural zone on the surface of the target rock wall. R linear .

[0077] Because a dynamic queue is introduced to manage the growth front and an access marker array is maintained to avoid repeated accesses, the growth termination condition is represented by a seed point queue. Q Empty.

[0078] In this example, to meet the requirements of the breadth-first search strategy and ensure the linear structural region... R linear To achieve uniform expansion and support the detection of complex branching linear structures, the system employs spatial indexing technology to accelerate the search process for neighboring points.

[0079] Figure 4 This is a linear structural zone map. Red dots represent extracted points, and green dots represent other points within the target rock wall area.

[0080] The improved region growing algorithm connects discrete candidate points to form a complete linear structure region by simulating the natural extension characteristics of linear structures in three-dimensional space. R linear The improved region growing algorithm offers superior performance in maintaining the continuity of linear structures compared to existing technologies (refer to existing technology document CN119539487A). Under the dual constraints of spatial and geometric conditions, iterative growth achieves accurate expansion, maintaining 90.7% continuity of linear structures while keeping the error expansion rate below 8.3%.

[0081] Example 2 Measurement of linear structural zones on rock walls.

[0082] Utilizing linear structural regions R linear The following tools were used for quantitative analysis of the rock wall surface engineering: Dip Direction and Dip Angle Measurements: Principal Component Analysis (PCA) or Random Sample Consensus (RANSAC) plane fitting algorithms are used to measure the extracted linear structural regions. R linear The point cloud is fitted with a microplane to determine the normal vector of the best-fit plane. Then, using analytical geometric formulas, the three-dimensional components of the normal vector are directly converted into geological dip and slope.

[0083] Projected area measurement: Using coordinate transformation and computational geometry tools, the linear structural area is first... R linear The point cloud is projected along its normal vector onto the best-fit plane, reducing its dimensionality from three to two. Then, the two-dimensional outer boundary of the projected point cloud is extracted using the convex hull or Alpha Shape algorithm, and the projected area is calculated using the polygon area formula.

[0084] Average roughness measurement: Calculates linear structural regions using surface statistical parameters or fractal geometry analysis tools. R linear The root mean square (RMS) or standard deviation of the distance is calculated from the perpendicular distance (i.e., the residual) from each point to the best-fit plane. Alternatively, a two-dimensional profile of the linear structure along a certain direction is extracted, the undulation or fractal dimension of its profile is calculated, and then it is mapped to a roughness coefficient using the classic Barton empirical formula.

[0085] The quantitative analysis process can be automated by computer programs (such as scripts written on Python or MATLAB platforms) or assisted by point cloud processing tools such as CloudCompare. Measurements are used to obtain linear structural regions. R linear The information is shown in Table 4 below.

[0086] Table 4 R linear Measurement data (partial)

[0087] Example 3

[0088] The linear structural region obtained in Example 1 R linear Based on the data, we continue to extract linear construction paths.

[0089] like Figure 5 As shown, a complete linear structural zone extends from the rock face surface.R linear Extracting linear paths from point cloud data P linear .

[0090] In obtaining the complete linear structural region R linear After the data is collected, the discrete linear structural points that make up the point cloud (i.e., the set of linear structural points) need to be processed. F (Points within) are transformed into smooth, continuous linear paths with a well-defined topological structure. P linear Only by transforming messy point cloud data into standardized paths can the measurement of linear structures be completed.

[0091] 1. Extraction of linear structural clusters C i The linear construction point set obtained based on Example 1 F The DBSCAN clustering algorithm was used to construct the linear region. R linear Euclidean distances between points within a linear structural region R linear The point cloud is clustered and grouped into several independent linear structural clusters. C i Each C i This represents an independent linear structure on the surface of the rock wall, enabling [the observation of]... R linear Effective differentiation of the various linear structures within.

[0092] 2. Constructing linear structural clusters C i graphical model G i ( V i , E i ) For any linear structure cluster C i Convert it into a weighted graph model G i ( V i , E i ).Depend on C i Vertices are directly generated from points within the vertices. V i Then, a Delaunay triangulation is used to establish... C i Topological connections between interior points are used to define edge sets.E i This forms the initial network structure. Based on this, the Euclidean distance between points is used as the weight of the corresponding edges in the graph. w ij In Equation 8, P c,i and P c,j They are two adjacent linear structural points P c .

[0093] Formula 8 Thus, the weighted graph model of each linear structure instance is obtained. G i The weights in the graph w ij The definition method ensures that the subsequent path extraction process can prioritize the selection of point pairs that are closer together, which is more in line with the natural continuity of the path structure.

[0094] After DBSCAN clustering, spatially adjacent and density-reachable linear structural points are labeled as the same cluster. The table below shows the linear structural clusters. C 05 Spatial coordinate data of key nodes in the middle section. This cluster is spatially represented as a set of discrete points extending along the Y-axis, and is related to other surrounding clusters (such as...). C 04 , C 06 There are obvious breaks in Euclidean space. Examples of calculations for linear structural clusters are shown in Table 5 below.

[0095] Table 5 C 05 Relevant data (nodes)

[0096] Based on the vertices in the table above, a topological connection edge set is established using Delaunay triangulation. E 05 And calculate the weights of each side. w ij The final weighted graph model generated G 05 The partial adjacency matrix data is shown in the table below. An example of weighted graph model calculation is shown in Table 6 below.

[0097] Table 6 G 05 Relevant data (adjacency matrix)

[0098] 3. Extracting the initial path of the linear structure Pinitial For any weighted graph model G i Extract the minimum spanning tree from it, and extract the longest path from it as the initial path for linear construction. P initial ,right P initial Smoothing is performed to obtain the extracted linear structural paths. P linear .

[0099] The minimum spanning tree algorithm is used to extract the main path of a linear structure. Specifically, it involves solving... G i Minimum Spanning Tree T i This spanning tree contains the set of minimum-weighted edges connecting all vertices. From T i The longest path is further extracted as the initial path for linear structure identification. P initial The initial path best represents the extension direction and overall shape of the linear structure, forming its main framework. The system records all possible path branches, maintaining the complete topological structure of the linear network. Solve... G i of T i The Prim algorithm or the Kruskal algorithm can be used.

[0100] 4. Construct linear construction paths P initial Initial understanding of linear structures P initial Smoothing is performed. Based on previous experimental analysis, a cubic spline interpolation algorithm is used for smoothing. This achieves the dual purpose of effectively eliminating the jagged path phenomenon caused by the discreteness of the point cloud and preserving the natural details of the linear structure's geometric features, resulting in a naturally smooth linear structure path. P linear .

[0101] Given a series of ordered points on a path { Q 0, Q 1, ..., Q m Construct parameterized cubic spline curves S ( t This ensures that the curve passes through all path points and remains second-order continuous at the connections. For each interval [ t i , t i+1The spline curve segment is represented by Equation 9, where, t These are normalized parameter variables, coefficients. a i , b i , c i , d i The solution is obtained by applying boundary conditions and continuity requirements.

[0102] Formula 9

[0103] The final linear construction path obtained P linear It possesses a natural and smooth geometric shape and contains complete topological information and attribute data, enabling it to support engineering quantitative analysis and measurement.

[0104] Figure 5 This is a diagram of linear structural paths. The green lines in the diagram represent linear structural regions. R linear Linear construction paths extracted from point cloud data P linear The remaining areas remain consistent with the target rock wall area.

[0105] Ultimately from Figure 2 37 spatially continuous linear structural paths with clear topological relationships were successfully identified in the scene's colored point cloud.

[0106] Example 4 Measurement of linear structural paths in rock walls.

[0107] Linear construction path extracted using Example 3 P linear The tools used for quantitative engineering analysis of linear structural features are: The actual length of the trace is measured: Based on the cubic spline smooth curve S(t) constructed in the previous step, the actual extension length of the path is accurately calculated using the curve integral algorithm (or by discretizing the smooth path into a dense set of points and summing the Euclidean distances between adjacent points).

[0108] Endpoint straight-line distance measurement: The three-dimensional coordinates of the first and last nodes of the linear construction path are directly extracted and calculated using the Euclidean distance formula between two points in space.

[0109] Tortuosity coefficient measurement: The tortuosity coefficient is the ratio of the true length of the trace to the straight-line distance of the endpoint.

[0110] Overall trajectory measurement: Extract the direction vector formed by the beginning and end points of the path, project it onto the horizontal plane (usually the XY plane), and use the arctangent function to analytically calculate the angle between the projected vector and the due north direction to obtain the azimuth of the trajectory.

[0111] The quantization calculation process can be automated using the spatial data analysis modules of scientific computing platforms such as Python (e.g., with the help of SciPy and NumPy libraries) or MATLAB. This allows for the acquisition of the measured linear structural paths. P linear The information is shown in Table 7 below.

[0112] Table 7 P linear Measurement data (partial)

[0113] Comparative Example

[0114] like Figure 6~Figure 8 As shown, the linear structure of the target rock wall surface in Example 1 is extracted using the existing two-dimensional image recognition method, and the key advantages of the present invention compared with the existing technical solutions are compared.

[0115] The two-dimensional image recognition method for linear structures on the target rock wall surface is based on existing technologies (Yi Xiaoyu, Chengdu University of Technology, 2022 thesis, Digital Recognition of Rock Mass Structure and Characterization of Variability of Mechanical Parameters and its Engineering Application).

[0116] 1. Comparison of original data dimensions and information completeness The technical solution of this invention is entirely based on three-dimensional point cloud detection technology, which supports arbitrary angle rotation observation and multi-view analysis of the original data of the target rock wall surface, and can completely preserve the three-dimensional spatial information of the rock surface. The collected original three-dimensional point cloud data can directly provide spatial coordinates with millimeter-level accuracy as the data basis for linear structural engineering measurement.

[0117] Existing technical solutions are based on two-dimensional image detection technology, which only supports a single shooting perspective for the raw data of the target rock wall. This places high demands on the raw data acquisition equipment, operation, and environment, and is prone to scale distortion and perspective deformation. Furthermore, it is susceptible to severe information loss due to factors such as the device's viewing angle, distance, and environment. The acquired raw two-dimensional images can only provide image coordinates based on image accuracy restoration analysis.

[0118] Figure 6 This is a comparison chart of the quality of the original data. Figure 6 (a) is an example of the original data diagram of the technical solution of the present invention. The original data diagram can be collected from multiple angles. Figure 6(b) is an example of the original data image of existing two-dimensional image recognition technology. The original data must be captured by orthographic projection as much as possible.

[0119] 2. Comparison of geometric accuracy and measurement accuracy The technical solution of this invention provides a direct analysis of millimeter-precision real spatial coordinate data from point cloud measurements for the engineering quantitative analysis of linear structures. The quantitative results do not depend on the selection of the viewing angle of the target rock wall, resulting in high geometric accuracy and high measurement accuracy.

[0120] Existing technologies for identifying linear structures are significantly affected by perspective distortion and lens distortion. The quantitative nature of the identified linear structures, namely the measurement of image coordinates, will yield significantly different measurement results for the same structure under different shooting angles. Therefore, the accuracy and precision are limited by the degree of reproduction of the shooting angle.

[0121] Figure 7 It is a line-shaped structure map extracted by a two-dimensional image recognition method. In two-dimensional image recognition technology, it is often necessary to first pre-label the line-shaped regions (usually represented by using a frame diagram to outline the recognition area), and then use deep learning to perform recognition. Figure 7 In the middle (a), there is a pre-annotated image in the existing two-dimensional image recognition process. Figure 7 (b) is a result of existing two-dimensional image recognition technology.

[0122] 3. Comparison of environmental adaptability and robustness The geometric features of 3D point clouds possess natural illumination invariance, maintaining stable detection performance even under complex lighting conditions such as shadows and reflections. 3D detection, through spatial geometric analysis, can effectively distinguish surface textures from actual structures. In contrast, 2D image detection is extremely sensitive to lighting conditions, and structural features inconsistent with the training data are often difficult to extract accurately. 2D detection is also easily affected by surface features such as rock color variations and moss coverage. Compared to these, the solution presented in this invention exhibits strong environmental adaptability, making it particularly suitable for engineering surveying tasks in complex outdoor environments, such as rock face terrain.

[0123] Figure 8 This is a comparison chart showing how lighting conditions affect the recognition results. Figure 8 (a) is a result diagram of the technical solution of the present invention, in which the linear structure of the reflective area (with a yellow frame) is accurately extracted; Figure 8 In Figure (b), the image is a pre-annotated image in the existing two-dimensional image recognition process. The reflective area (highlighted in yellow) was not successfully detected in the pre-annotated linear area step, and the linear structure could not be identified in subsequent steps.

[0124] 4. Capability to handle complex scenarios For complex linear structures with occlusion and overlap, the three-dimensional detection of the present invention can accurately separate each structural element through spatial topology analysis, and has a strong ability to process complex geological scenes; two-dimensional images lack corresponding strategies for such situations and are prone to misjudgment.

Claims

1. A method for extracting linear structural information of rock walls from point cloud data, including: Step S100: Use a drone to acquire the original three-dimensional point cloud of the rock wall, and obtain the preserved three-dimensional point cloud after preprocessing. The preprocessing includes voxel downsampling. Its features are: Step S200, based on the reserved three-dimensional point cloud, clustering analysis is carried out in the RGB color space by using the DBSCAN algorithm, and outliers in the clustering process are extracted to obtain a preliminary line structure area R initial ; Step S300, extracting the wall geometry from the lineament preliminary area R initial identifying the initial seed point P ps obtaining the initial lineament point set F 0, including, Step S310, screening out points with curvature less than a curvature threshold value as concave region feature points through local curvature feature analysis from the line structure preliminary recognition area R initial κ 1 extracting a concave region.​ Curvature threshold κ The determination method is as follows: calculate the initial identification region of the linear structure. R initial The curvature of all points within the area is calculated and a frequency histogram is plotted. Local minima are then used to determine the local minimum values. κ 1; Step S320: Based on the point cloud data of the concave region, search for the nearest neighbors of any point and filter out the angle between the normal vectors of any point and all its neighbors. θ All are greater than the included angle threshold θ Points with a value of 1 are identified as initial seed points. P ps 30º≤ θ 1≤50º; Step S330: Construct an initial linear construction point set. F 0, set the initial seed point P ps deposit F 0; Step S400, Initial identification of the linear structure region R initial Extracting linear structural regions R linear ,include, Step S410, Create a seed point queue Q ,Will F 0 Midpoint Push In Q , Q It is a first-in, first-out dynamic queue; Step S420: Read the seed point queue Q Points in P s Determined by the system of equations 1 P s Local linear structure growth direction , Equation 1-1 Formula 1-2 In the formula, Seed point P s The normal vector, - Global reference vector; Step S430: Construct seed points according to the system of equations 2. P s Introduction to Linear Structures R initial Candidate point directional search area R search , Equation 2-1 Equation 2-2 Equation 2-3 In the formula, R search - Candidate point targeted search area, P - Introduction to Linear Structures R initial any point in, w - R search Width parameter, q - Coefficient, with a value of 1.2 ≤ q ≤1.5, w 0 – Expected width of the linear structure. k - Coverage factor, 2≤ k ≤5, d voxel - Voxel downsampling size during the preprocessing stage of raw 3D point cloud data; Step S440: Create an empty linear construction point set. F ; In the candidate point directional search area R search Inside, search for seed points P s nearest neighbor P j ,like P j Meeting the growth conditions and not having been pushed into the seed point queue Q Identified as a seed point P s ,join in F And push in Q Otherwise ignore this P j Continue searching and identifying the next one. P j The growth conditions are expressed according to Formula 3. Formula 3 In the formula, κ j - Neighboring points P j curvature, κ grow - Curvature growth threshold, κ 1< κ grow <0.06, θ sj Seed point P s and neighboring points P j The difference in the angle of the normal vector θ grow - Normal vector angle difference growth threshold, 30º < θ grow < θ 1; Step S450: Repeat steps S420 to S440 and perform iterative calculations until the growth termination condition is met. The growth termination condition is that the algorithm exhausts all possibilities starting from the seed point. P s Starting from a candidate region that is spatially connected and satisfies geometric constraints, the growth front achieves complete closure; After the growth operation terminates, the linear structure point set F This refers to the extracted linear structural region. R linear .

2. The method for extracting linear structural information of rock walls from point cloud data according to claim 1, characterized in that: Step S200 includes, Step S210: Normalize the RGB value of each point to the [0,1] interval to obtain a three-dimensional color feature vector; subsequent clustering is based only on the color space, and the distance between color components is measured by three-dimensional Euclidean distance. Step S220: Randomly select M points from the preserved 3D point cloud to form a sample set, calculate the distance between the color components of all point pairs in the sample set, and obtain the distance set; plot the frequency histogram of the distance set with the group interval Δd, locate the main peak of the histogram, and search for the first local minimum point to the right of the main peak. d val ,Pick r = d val Used as the global neighborhood radius; sets the minimum number of neighborhood points, MinPts. Step S230, using parameters r Using MinPts, DBSCAN clustering is performed on all points in the color space to extract outliers as linear constructs of initial recognition regions. R initial .

3. The method for extracting linear structural information of rock walls from point cloud data according to claim 1, characterized in that: In step S310, the curvature threshold κ 1. Determined by the system of equations 4. Equation 4-1 Equation 4-2 In the formula, H ( κ - Introduction to Linear Structures R initial The frequency histogram function of point cloud curvature distribution. κ speak - The curvature value corresponding to the maximum frequency in the frequency histogram. κ i - Discrete curvature values ​​of the frequency histogram interval.

4. The method for extracting linear structural information of rock walls from point cloud data according to claim 1, characterized in that: In step S440 κ grow , θ grow Satisfying equations 5 and 6 respectively, Formula 5 Formula 6.

5. The method for extracting linear structural information of rock walls from point cloud data according to claim 1, characterized in that: In step S450, the growth termination condition is the seed point queue. Q Empty.

6. The method for extracting linear structural information of rock walls from point cloud data according to any one of claims 1 to 5, characterized in that: It also includes step S500, from the linear structural region R linear Extracting linear construction paths P linear ,include, Step S510: Based on the linear construction point set obtained in step S450 F The DBSCAN clustering algorithm was used to construct the linear region. R linear The Euclidean distance between each point in the interior will R linear Point cloud clustering and grouping yields segmented, independent linear structural clusters. C i ; Step S520, for any linear structural cluster C i Convert it into a weighted graph model G i ( V i , E i ),Depend on C i Vertices are directly generated from points within the vertices. V i Established through Delaunay triangulation C i Topological connections between interior points define edge sets. E i Then, the Euclidean distance between points is used as the weight of the corresponding edges in the graph. w ij ; Step S530, from the weighted graph model G i Extract the minimum spanning tree and the longest path as the initial path for linear construction. P initial ; Step S540: Initial identification of the linear structure path P initial Smoothing process is performed to obtain linear structural paths. R linear .

7. The method for extracting linear structural information of rock walls from point cloud data according to claim 6, characterized in that: Step S540 also includes a curvature constraint optimization mechanism, which includes: detecting abnormal curvature abrupt change points in the path, moving their coordinates closer to the middle position of their adjacent points, straightening out minor jitters, and avoiding unnatural excessive distortion of the path.

8. The method for extracting linear structural information of rock walls from point cloud data according to claim 7, characterized in that: Step S540 further includes a redundancy detection and optimization mechanism, which includes: detecting redundant points in the path that do not actually contribute to describing the shape of the path, and deleting the redundant points while ensuring that the shape of the path is not deformed.

9. A method for measuring linear structures on rock walls, characterized in that: This is achieved using the method for extracting linear structural information of rock walls from point cloud data as described in claim 8, including: First, determine the rock wall surface to be measured; Secondly, the method for extracting linear structural information of rock walls from the point cloud data is implemented, and the linear structural area is output at the end of step S400. R linear Data and / or output linear construction path at the end of step S500 P linear data; Finally, regarding the linear structural region R linear and / or linear structural paths P linear Conduct engineering quantitative analysis to obtain corresponding measurement data.

10. A rock wall linear structure measurement system, comprising a memory, a processor, and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement the steps of the linear rock wall structure measurement method of claim 9.

Images