Dangerous rock crack feature extraction method and system based on live-action three-dimensional model

Abstract

The invention is suitable for the technical field of geological monitoring, and provides a dangerous rock crack feature extraction method and system based on a live-action three-dimensional model, and the method comprises the following steps: carrying out the collection based on an oblique photography camera to obtain multi-view image data, and carrying out the collection based on a laser radar to obtain point cloud data; determining an oblique photography model based on the multi-view image data, determining a laser radar point cloud model based on the point cloud data, and obtaining a three-dimensional model of the dangerous rock area; performing crack identification on the three-dimensional model through an edge detection algorithm, and extracting crack types and crack geometric features through a deep learning algorithm; and geological background data are called, and a risk assessment index system is established according to the geological background data, the crack type and the crack geometric characteristics. According to the method, the high-precision live-action three-dimensional model is constructed, the limitation of a single data source on precision, details and coverage range is overcome through multi-source data fusion, and the integrity and accuracy of the three-dimensional model are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of geological monitoring, and in particular, to a method and system for extracting characteristics of dangerous rock cracks based on a real-scene three-dimensional model. Background Technique

[0002] At present, the detection of dangerous rock cracks mainly relies on manual inspections or traditional two-dimensional remote sensing image analysis. Manual inspections are time-consuming and laborious, difficult to cover large areas, and the detection results are greatly affected by subjective factors, making it difficult to ensure consistency; two-dimensional images cannot fully reflect the spatial distribution and morphological characteristics of cracks, and due to the lack of three-dimensional information, it is difficult to accurately reflect the spatial distribution and depth information of cracks.

[0003] In the prior art, a LiDAR crack detection method is also used. High-density point cloud data of the ground surface is obtained by LiDAR, and the spatial distribution and morphological characteristics of cracks are analyzed. The position, length, width and other characteristics of cracks are extracted from the point cloud data. A three-dimensional model of the cracks is generated by combining three-dimensional modeling technology. Although the detection accuracy is high and the spatial distribution of cracks can be reflected. However, the equipment cost is high, and professional data processing technology is required. In addition, an infrared sensor crack detection method can also be used, but the detection range is limited, it is difficult to cover large areas, and it is greatly affected by environmental conditions (such as temperature, light).

[0004] It can be seen that the prior art generally lacks efficiency, accuracy and applicability. Traditional methods rely on manual operations, with low efficiency and high costs; although modern technologies have improved the detection accuracy, the equipment costs are high and the applicability is limited. Therefore, a method and system for extracting characteristics of dangerous rock cracks based on a real-scene three-dimensional model are needed to solve the above problems. Summary of the Invention

[0005] Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide a method and system for extracting characteristics of dangerous rock cracks based on a real-scene three-dimensional model to solve the problems existing in the above background technique.

[0006] The present invention is implemented as follows. A method for extracting characteristics of dangerous rock cracks based on a real-scene three-dimensional model, the method comprising the following steps:

[0007] Based on multi-view image data collected by an oblique photography camera and point cloud data collected by LiDAR;

[0008] Based on the multi-view image data, an oblique photography model is determined, and based on the point cloud data, a LiDAR point cloud model is determined to obtain a three-dimensional model of the dangerous rock area;

[0009] Crack identification is performed on the three-dimensional model through an edge detection algorithm, and the crack type and crack geometric characteristics are extracted through a deep learning algorithm;

[0010] Retrieve geological background data and establish a risk assessment index system based on the geological background data, fracture types, and fracture geometric characteristics.

[0011] As a further solution of the present invention: The steps of collecting multi-view image data based on an oblique photography camera and collecting point cloud data based on lidar specifically include:

[0012] Use a drone equipped with an oblique photography camera and lidar to take multi-angle photos and collect point cloud data of the dangerous rock area to obtain multi-view image data and point cloud data;

[0013] Perform denoising, registration, and stitching processing on the multi-view image data to ensure the geometric consistency and texture integrity of the images;

[0014] Perform noise reduction, filtering, and smoothing processing on the point cloud data to remove abnormal points and noise in the point cloud data.

[0015] As a further solution of the present invention: The steps of determining an oblique photography model based on the multi-view image data, determining a lidar point cloud model based on the point cloud data, and obtaining a three-dimensional model of the dangerous rock area specifically include:

[0016] Process the multi-view image data through multi-view image matching and stereo reconstruction techniques to construct a three-dimensional geometric structure of the dangerous rock area and determine the oblique photography model;

[0017] Generate a digital elevation model using the point cloud data, and extract the terrain features of the dangerous rock area through point cloud classification and segmentation techniques to determine the lidar point cloud model;

[0018] Fuse the oblique photography model and the lidar point cloud model to generate a three-dimensional model of the dangerous rock area.

[0019] As a further solution of the present invention: The steps of identifying fractures in the three-dimensional model through an edge detection algorithm and extracting fracture types and fracture geometric characteristics through a deep learning algorithm specifically include:

[0020] Use the Canny edge detection algorithm to identify the location of the fractures and the edge areas of the fractures;

[0021] Identify fracture types through a convolutional neural network model and extract fracture geometric characteristics, where the fracture geometric characteristics include length L, width W, and orientation θ;

[0022] Perform statistical analysis on the extracted fracture geometric characteristics to calculate the number, density, and propagation rate SR of the fractures.

[0023] As a further solution of the present invention: the step of retrieving geological background data and establishing a risk assessment index system according to the geological background data, fracture types and fracture geometric characteristics specifically includes:

[0024] Determine the comprehensive risk index CRI, where the geological background data includes rock type R, rainfall P and seismic activity intensity E, CRI = w L ×L+w W ×W+w SR ×SR+w θ ×θ+w R ×R+w P ×P+w E ×E,w L 、w W 、w SR 、w θ 、w R 、w P and w E are the weight coefficients of each indicator;

[0025] Cracks are divided into three risk levels according to the comprehensive risk index, and risk visualization is performed.

[0026] As a further solution of the present invention: when determining the oblique photography model, a sparse point cloud is generated based on SIFT / ORB feature matching, a dense point cloud is generated through semi-global matching, a three-dimensional mesh model is generated by Poisson surface reconstruction, and a textured three-dimensional model is obtained by combining UV mapping projection texture; when determining the lidar point cloud model, the ICP algorithm is used to align multi-view point clouds, the terrain features are classified through random forest / PointNet++, and the digital elevation model is generated through RANSAC plane fitting.

[0027] Another object of the present invention is to provide a dangerous rock crack feature extraction system based on a real-scene three-dimensional model, the system comprising:

[0028] A data acquisition and processing module is used to acquire multi-view image data based on the oblique photography camera and to acquire point cloud data based on the laser radar;

[0029] The 3D model fusion module is used to determine the oblique photography model based on the multi-view image data and the laser radar point cloud model based on the point cloud data to obtain the 3D model of the dangerous rock area;

[0030] The crack feature determination module is used to identify cracks in the 3D model through edge detection algorithm and extract crack types and crack geometric features through deep learning algorithm;

[0031] The risk assessment system module is used to retrieve geological background data and establish a risk assessment indicator system based on geological background data, fracture types and fracture geometric characteristics.

[0032] As a further solution of the present invention: the three-dimensional model fusion module includes:

[0033] The oblique photography model unit is used to process the multi-view image data through multi-view image matching and stereo reconstruction technology, construct the three-dimensional geometric structure of the dangerous rock area, and determine the oblique photography model;

[0034] The laser point cloud model unit is used to generate a digital elevation model using point cloud data, extract the terrain features of the dangerous rock area through point cloud classification and segmentation technology, and determine the laser radar point cloud model;

[0035] The model fusion unit is used to fuse the oblique photography model with the lidar point cloud model to generate a three-dimensional model of the dangerous rock area.

[0036] As a further solution of the present invention: the crack feature determination module includes:

[0037] An edge detection and recognition unit, used to recognize the position of the crack and the edge area of ​​the crack using a Canny edge detection algorithm;

[0038] A crack feature extraction unit, used to identify the crack type through a convolutional neural network model and extract crack geometric features, wherein the crack geometric features include length L, width W and strike θ;

[0039] The characteristic statistical analysis unit is used to perform statistical analysis on the extracted fracture geometric characteristics and calculate the number, density and expansion rate SR of the fractures.

[0040] As a further solution of the present invention: the risk assessment system module includes:

[0041] The risk index calculation unit is used to determine the comprehensive risk index CRI, where the geological background data includes rock type R, rainfall P and seismic activity intensity E, CRI = w L ×L+w W ×W+w SR ×SR+w θ ×θ+w R ×R+w P ×P+w E ×E,w L 、w W 、w SR 、w θ 、w R 、w P and w E are the weight coefficients of each indicator;

[0042] A risk level classification unit is used to divide cracks into three risk levels according to the comprehensive risk index for risk visualization.

[0043] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0044] The present invention combines the real scene three-dimensional technology with the deep learning algorithm to realize the automatic identification and analysis of dangerous rock cracks. By combining the oblique photography data with the lidar point cloud data, a high-precision real scene three-dimensional model is constructed. Through multi-source data fusion, the limitations of a single data source in terms of accuracy, details, and coverage are overcome, and the integrity and accuracy of the three-dimensional model are significantly improved. A comprehensive risk index evaluation system based on crack characteristics and geological background data is established, which can quantitatively evaluate the geological disaster risk level of the dangerous rock area. Description of the Drawings

[0045] Figure 1 It is a flowchart of a method for extracting dangerous rock crack characteristics based on a real scene three-dimensional model.

[0046] Figure 2 It is a flowchart of data collection in a method for extracting dangerous rock crack characteristics based on a real scene three-dimensional model.

[0047] Figure 3 It is a flowchart of determining a three-dimensional model in a method for extracting dangerous rock crack characteristics based on a real scene three-dimensional model.

[0048] Figure 4 It is a flowchart of determining crack characteristics in a method for extracting dangerous rock crack characteristics based on a real scene three-dimensional model.

[0049] Figure 5 It is a flowchart of establishing a risk assessment index system in a method for extracting dangerous rock crack characteristics based on a real scene three-dimensional model.

[0050] Figure 6 It is a structural schematic diagram of a system for extracting dangerous rock crack characteristics based on a real scene three-dimensional model. Detailed Embodiments

[0051] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the following further details the present invention in combination with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0052] The following details the specific implementation of the present invention in combination with specific embodiments.

[0053] As Figure 1 shown, the embodiment of the present invention provides a method for extracting dangerous rock crack characteristics based on a real scene three-dimensional model, and the method includes the following steps:

[0054] S100, Collect multi-view image data based on an oblique photography camera and point cloud data based on lidar.

[0055] S200, Determine an oblique photography model based on the multi-view image data and a lidar point cloud model based on the point cloud data to obtain a three-dimensional model of the dangerous rock area.

[0056] S300, Identify cracks in the three-dimensional model through an edge detection algorithm, and extract crack types and crack geometric features through a deep learning algorithm.

[0057] S400, Retrieve geological background data and establish a risk assessment index system based on the geological background data, crack types, and crack geometric features.

[0058] It should be noted that UAV oblique photography is a technology for obtaining high-resolution image data through multi-angle shooting. Here, a multi-rotor UAV (such as DJI Mavic Pro or DJI Phantom 4) is used to carry an oblique photography camera (such as DJI Zenmuse XT series). Set the camera parameters: focal length, aperture, shutter speed, etc. to ensure image quality. Use flight control software (such as DJI GS Pro) to plan the flight route. Set the flight altitude (such as 100 meters), flight speed (such as 5 m / s), and overlap rate (such as 70%). Ensure that the entire dangerous rock area is covered and multi-angle images (such as in the front, back, left, and right directions) are taken. The UAV flies according to the planned route to capture multi-view image data. Each image file contains geographical location information (such as longitude, latitude, and altitude).

[0059] In the embodiments of the present invention, lidar measures the three-dimensional coordinates of the object surface by emitting laser pulses to generate high-precision point cloud data. The following are the specific implementation steps: 1. Lidar configuration: Use a ground lidar device (such as Leica BLK360) or an airborne lidar (such as Riegl VQ-1560i). Set the scanning parameters: scanning frequency, scanning angle, point cloud density, etc. 2. Data acquisition: Conduct a full-range scan of the dangerous rock area to obtain high-density point cloud data. The point cloud data contains the three-dimensional coordinates (X, Y, Z) and reflection intensity of each point. 3. Collection of geological background data: In addition to multi-source images and point cloud data, relevant geological background data also needs to be collected, such as: Rock mass type: Obtained through on-site investigation or geological maps; Rainfall: Obtained through weather stations or online meteorological data; Seismic activity: Obtained through seismic monitoring networks.

[0060] As Figure 2 shown, as a preferred embodiment of the present invention, the step of collecting multi-view image data based on an oblique photography camera and point cloud data based on lidar specifically includes:

[0061] S101, Use a drone equipped with an oblique photography camera and a lidar to take multi-angle photos and collect point cloud data of the dangerous rock area, obtaining multi-view image data and point cloud data;

[0062] S102, Denoise, register, and splice the multi-view image data to ensure the geometric consistency and texture integrity of the images;

[0063] S103, Denoise, filter, and smooth the point cloud data to remove outliers and noise in the point cloud data.

[0064] In the embodiment of the present invention, when processing images, OpenCV is used for image denoising (Fast Non-Local Means Denoising algorithm), registration (ORB feature matching), and splicing (multi-view image fusion). When processing point clouds, statistical filtering (SOR algorithm) is used to remove outliers, Gaussian filtering is used to smooth the noise, and DBSCAN clustering is used to segment the point cloud. Finally, the ICP algorithm is used to align the coordinate system, and the oblique photography model and LiDAR point cloud are fused to generate standardized preprocessed data.

[0065] Specifically, 1. Image denoising: Use an image processing algorithm to remove noise (such as salt-and-pepper noise, Gaussian noise) from the image, algorithm: Fast Non-Local Means Denoising. 2. Image registration: Align images from different perspectives to the same coordinate system through feature point matching, algorithm: ORB (Oriented FAST and Rotated BRIEF), where the key steps include: FAST key point detection: Fast corner detection based on pixel gray values; BRIEF descriptor: Generation of binary descriptors by comparing the gray value differences of pixel pairs; Hamming distance matching: Calculating the differences between descriptors. 3. Image splicing: Splice multi-view images into a continuous orthophoto to ensure geometric consistency and texture integrity, algorithm: Image transformation based on the homography matrix. 4. Denoising processing of point cloud data: Use statistical filtering or neighborhood-based methods to remove outliers and noise, algorithm: Statistical Outlier Removal. 5. Point cloud filtering: Smooth the point cloud data through high-pass filtering or low-pass filtering to remove high-frequency noise, algorithm: Gaussian filtering for smoothing. 6. Point cloud segmentation: Segment the point cloud into different regions (such as rock regions, vegetation regions, etc.) for subsequent processing, algorithm: DBSCAN (Density-Based Spatial Clustering). 6. Data fusion preparation: For subsequent multi-source data fusion, it is necessary to align the image data and point cloud data to the same coordinate system and perform necessary format conversions. Coordinate system alignment algorithm: ICP (Iterative Closest Point algorithm).

[0066] As Figure 3 shown, as a preferred embodiment of the present invention, the steps of determining an oblique photography model based on multi-view image data, determining a lidar point cloud model based on point cloud data, and obtaining a three-dimensional model of the dangerous rock area specifically include:

[0067] S201, processing the multi-view image data through multi-view image matching and stereo reconstruction techniques, constructing a three-dimensional geometric structure of the dangerous rock area, and determining an oblique photography model;

[0068] S202, generating a digital elevation model (DEM) using the point cloud data, extracting the terrain features of the dangerous rock area through point cloud classification and segmentation techniques, and determining a lidar point cloud model;

[0069] S203, fusing the oblique photography model and the lidar point cloud model to generate a three-dimensional model of the dangerous rock area.

[0070] In the embodiment of the present invention, when determining the oblique photography model, a sparse point cloud is generated based on SIFT / ORB feature matching, a dense point cloud is generated through semi-global matching, a three-dimensional mesh model is generated by Poisson surface reconstruction, and a textured three-dimensional model is obtained by combining UV mapping projection texture; when determining the lidar point cloud model, the ICP algorithm is used to register multi-view point clouds, the terrain features are classified through random forest / PointNet++, and a digital elevation model is generated by RANSAC plane fitting; finally, the texture model and the DEM are fused to generate a high-precision three-dimensional mine model.

[0071] Specifically, one, oblique photography modeling: Use professional three-dimensional modeling software (such as Agisoft Metashape or Pix4Dmapper) for data processing, mainly including the following steps: 1. Image matching: Generate a sparse point cloud through feature point matching, algorithms: SIFT (Scale-Invariant Feature Transform) or ORB (Oriented FAST and Rotated BRIEF), where the key steps are: SIFT: Detect key points through a Gaussian difference pyramid and generate a 128-dimensional feature descriptor; ORB: Based on FAST corner detection and BRIEF descriptor, introduce direction information (gray centroid method). Matching error formula: Matching error = ∑||H·p i - p′ i || 2, where H is the homography matrix, and pi and pi′ are matching point pairs. 2. Dense matching: Generate a dense point cloud through dense matching; algorithm: Semi-Global Matching (SGM). 3. Mesh generation: Convert the point cloud into a three-dimensional mesh model; algorithm: Poisson Surface Reconstruction, where the core formula is: φ: An implicit function representing the three-dimensional surface; V: The normal vector field of the point cloud, generating a continuous surface through divergence calculation. 4. Texture mapping: Project the original image onto the mesh model to generate a three-dimensional model with rich textures; algorithm: UV mapping and image projection.

[0072] II. LiDAR point cloud modeling: 1. Data preprocessing: Preprocess the point cloud data, mainly including noise reduction, filtering, and registration. 2. Point cloud classification and segmentation: Use machine learning algorithms (such as random forest or deep learning) to classify and segment the point cloud, and extract the terrain features of the dangerous rock area; algorithm: random forest / PointNet++. Perform terrain feature extraction; algorithm: RANSAC (Random Sample Consensus) plane fitting.

[0073] As Figure 4 shown, as a preferred embodiment of the present invention, the steps of identifying cracks in the three-dimensional model through the edge detection algorithm and extracting crack types and crack geometric features through the deep learning algorithm specifically include:

[0074] S301, Use the Canny edge detection algorithm to identify the location of the cracks and the edge regions of the cracks;

[0075] S302, Identify the crack types through a convolutional neural network model and extract the crack geometric features, where the crack geometric features include length L, width W, and orientation θ;

[0076] S303, Perform statistical analysis on the extracted crack geometric features, and calculate the number, density, and propagation rate SR of the cracks.

[0077] In the embodiment of the present invention, edge detection is the first step in crack feature extraction, used to identify the location and orientation of the cracks. We adopt the classic Canny edge detection algorithm. It mainly includes the following steps: 1. Gaussian filtering: Smooth the image to reduce noise. 2. Calculate the gradient: Use the Sobel operator to calculate the gradient magnitude and direction of the image. 3. Non-maximum suppression: Retain the pixels with the maximum gradient magnitude and suppress other edge responses. 4. Double-threshold detection: Use two thresholds, high and low, to distinguish strong edges and weak edges, and connect the weak edges to form complete edges.

[0078] Deep learning algorithms are used to further extract fine features of fractures, such as fracture types (tensile fractures, shear fractures, etc.) and geometric parameters of fractures (such as length, width, etc.). Here, a convolutional neural network (CNN) is used for classification and feature extraction, and the CNN model is pre-trained. The CNN model structure includes the following layers: 1. Convolutional layer: Extract local features of the image. 2. Pooling layer: Reduce computational complexity and improve the robustness of the model. 3. Fully connected layer: Map features to the category space. 4. Output layer: Use the Softmax function for classification.

[0079] After the edges of the fractures are extracted, it is necessary to further analyze the geometric features of the fractures, such as length, width, orientation, etc. 1. Fracture length calculation: Use the cumulative Euclidean distance formula to find the coordinates of all edge points and calculate the longest distance as the fracture length; 2. Fracture width calculation: Use the average spacing formula to take multiple cross-sections along the fracture orientation and calculate the average distance between the edge points on both sides. 3. Fracture orientation angle calculation: Use the direction angle formula to reflect the overall extension direction of the fracture.

[0080] As Figure 5 shown, as a preferred embodiment of the present invention, the steps of retrieving geological background data and establishing a risk assessment index system according to the geological background data, fracture types, and fracture geometric features specifically include:

[0081] S401, Determine the comprehensive risk index CRI. Among them, the geological background data includes rock mass type R, rainfall P, and seismic activity intensity E, and CRI = w L ×L + w W ×W + w SR ×SR + w θ ×θ + w R ×R + w P ×P + w E ×E, where w L , w W , w SR , w θ , w R , w P and w E are all weight coefficients of each index;

[0082] S402, Divide the fractures into three risk levels according to the comprehensive risk index and perform risk visualization.

[0083] In the embodiments of the present invention, the following factors will be considered: crack length, i.e., the extended length of the crack; crack width, i.e., the maximum width of the crack; crack propagation rate, i.e., the propagation speed of the crack within a certain period of time; crack orientation, i.e., the angle between the crack and the horizontal plane. Rock mass type: Different rock mass types have different effects on crack propagation; rainfall: An increase in rainfall may lead to an accelerated crack propagation; seismic activity: Seismic activity may induce or exacerbate crack propagation. According to the comprehensive risk index (CRI), cracks are divided into three risk levels: low risk (CRI < 30): The crack propagates slowly and will not cause significant harm in the short term. Medium risk (30 ≤ CRI < 70): The crack has a certain propagation trend and requires enhanced monitoring. High risk (CRI ≥ 70): The crack propagates rapidly, posing a relatively large safety hazard and immediate prevention and control measures are required. In the 3D model, the risk levels are marked with colors (red: high risk; orange: medium risk; green: low risk) to generate a risk heat map.

[0084] As Figure 6 shown, the embodiments of the present invention further provide a system for extracting dangerous rock crack characteristics based on a real-scene 3D model, and the system includes:

[0085] A data acquisition and processing module 100, configured to collect multi-view image data based on an oblique photography camera and collect point cloud data based on a lidar;

[0086] A 3D model fusion module 200, configured to determine an oblique photography model based on the multi-view image data, determine a lidar point cloud model based on the point cloud data, and obtain a 3D model of the dangerous rock area;

[0087] A crack feature determination module 300, configured to identify cracks in the 3D model through an edge detection algorithm and extract crack types and crack geometric features through a deep learning algorithm;

[0088] A risk assessment system module 400, configured to retrieve geological background data and establish a risk assessment index system according to the geological background data, crack types, and crack geometric features.

[0089] As a preferred embodiment of the present invention, the 3D model fusion module 200 includes:

[0090] An oblique photography model unit, configured to process the multi-view image data through multi-view image matching and stereoscopic reconstruction techniques, construct a 3D geometric structure of the dangerous rock area, and determine the oblique photography model;

[0091] A laser point cloud model unit, configured to generate a digital elevation model using the point cloud data, extract the terrain features of the dangerous rock area through point cloud classification and segmentation techniques, and determine the lidar point cloud model;

[0092] The model fusion unit is used to fuse the oblique photography model with the lidar point cloud model to generate a three-dimensional model of the dangerous rock area.

[0093] As a preferred embodiment of the present invention, the crack feature determination module 300 includes:

[0094] An edge detection and recognition unit, used to recognize the position of the crack and the edge area of ​​the crack using a Canny edge detection algorithm;

[0095] A crack feature extraction unit, used to identify the crack type through a convolutional neural network model and extract crack geometric features, wherein the crack geometric features include length L, width W and strike θ;

[0096] The characteristic statistical analysis unit is used to perform statistical analysis on the extracted fracture geometric characteristics and calculate the number, density and expansion rate SR of the fractures.

[0097] As a preferred embodiment of the present invention, the risk assessment system module 400 includes:

[0098] The risk index calculation unit is used to determine the comprehensive risk index CRI, where the geological background data includes rock type R, rainfall P and seismic activity intensity E, CRI = w L ×L+w W ×W+w SR ×SR+w θ ×θ+w R ×R+w P ×P+w E ×E,w L 、w W 、w SR 、w θ 、w R 、w P and w E are the weight coefficients of each indicator;

[0099] The risk level classification unit is used to classify cracks into three risk levels according to the comprehensive risk index and to visualize the risks.

[0100] In the embodiments of the present invention, through the use of drones and multi-source data acquisition technologies, the efficiency of crack detection is significantly improved. Based on high-precision three-dimensional models and deep learning algorithms, the accuracy of crack feature extraction is remarkably enhanced. An intuitive three-dimensional visualization interface is provided to facilitate decision-makers' understanding of the spatial distribution and morphological characteristics of cracks. By combining real-scene three-dimensional technology with deep learning algorithms, automatic identification and analysis of dangerous rock cracks are achieved. A method for crack feature extraction based on multi-source data fusion is proposed, improving the accuracy and reliability of detection. By combining oblique photography data with lidar point cloud data, a high-precision real-scene three-dimensional model is constructed. Through multi-source data fusion, the limitations of a single data source in terms of accuracy, details, and coverage are overcome, significantly enhancing the integrity and accuracy of the three-dimensional model. A method for crack feature extraction based on edge detection and machine learning is proposed, enabling the automatic identification and extraction of geometric features such as the location, length, and width of cracks from the three-dimensional model. Compared with traditional manual visual interpretation methods, the work efficiency and accuracy are greatly improved. A comprehensive risk index (CRI) assessment system based on crack features and geological background data is established, which can quantitatively evaluate the geological disaster risk level of dangerous rock areas. Through scientific risk assessment methods, decision-making support for geological disaster prevention and control is provided.

[0101] The above only describes the preferred embodiments of the present invention in detail and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

[0102] It should be understood that although the steps in the flowcharts of the embodiments of the present invention are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limit, and these steps can be executed in other orders. Moreover, at least some of the steps in each embodiment may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential either, but can be executed alternately or in rotation with at least a part of other steps or sub-steps or stages of other steps.

[0103] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0104] After considering the specification and the disclosure of the embodiments, those skilled in the art will readily conceive of other embodiments of the present disclosure. This application is intended to cover any variations, uses, or adaptations of the present disclosure, which follow the general principles of the present disclosure and include known common knowledge or conventional technical means in the technical field not disclosed in the present disclosure. The specification and the embodiments are only regarded as exemplary, and the true scope and spirit of the present disclosure are pointed out by the claims.

Claims

1. A method for extracting dangerous rock crack features based on a real-life three-dimensional model, characterized in that: The method comprises the following steps: Multi-view image data is acquired based on the oblique photography camera, and point cloud data is acquired based on the lidar. Determine the oblique photography model based on the multi-view image data, determine the lidar point cloud model based on the point cloud data, and obtain a three-dimensional model of the dangerous rock area; Cracks are identified in the 3D model using edge detection algorithms, and crack types and geometric features are extracted using deep learning algorithms. The geological background data is retrieved, and a risk assessment index system is established based on the geological background data, fracture types and fracture geometric characteristics.

2. The method for extracting dangerous rock crack features based on a real-scene three-dimensional model according to claim 1 is characterized in that: The step of acquiring multi-view image data based on the oblique photography camera and acquiring point cloud data based on the laser radar specifically includes: Use a drone equipped with an oblique camera and a laser radar to take multi-angle photos of dangerous rock areas and collect point cloud data to obtain multi-view image data and point cloud data; De-noising, registration and stitching of multi-view image data to ensure geometric consistency and texture integrity of the images; The point cloud data is subjected to denoising, filtering and smoothing to remove abnormal points and noise in the point cloud data.

3. The method for extracting dangerous rock crack features based on a real-scene three-dimensional model according to claim 1 is characterized in that: The step of determining the oblique photography model based on the multi-view image data, determining the laser radar point cloud model based on the point cloud data, and obtaining the three-dimensional model of the dangerous rock area specifically includes: The multi-view image data is processed through multi-view image matching and stereo reconstruction technology to construct the three-dimensional geometric structure of the dangerous rock area and determine the oblique photography model; Generate a digital elevation model using point cloud data, extract the topographic features of the dangerous rock area through point cloud classification and segmentation technology, and determine the lidar point cloud model; The oblique photography model is integrated with the LiDAR point cloud model to generate a three-dimensional model of the dangerous rock area.

4. The method for extracting dangerous rock crack features based on a real-scene three-dimensional model according to claim 1, characterized in that: The steps of identifying cracks in the three-dimensional model by using an edge detection algorithm and extracting crack types and crack geometric features by using a deep learning algorithm specifically include: Use the Canny edge detection algorithm to identify the location of the crack and the edge area of ​​the crack; Identify the fracture type through a convolutional neural network model and extract fracture geometric features, wherein the fracture geometric features include length L, width W and strike θ; The extracted fracture geometric features are statistically analyzed to calculate the number, density and propagation rate SR of the fractures.

5. The method for extracting dangerous rock crack features based on a real-scene three-dimensional model according to claim 4 is characterized in that: The step of retrieving geological background data and establishing a risk assessment index system according to the geological background data, fracture types and fracture geometric characteristics specifically includes: Determine the comprehensive risk index CRI, where the geological background data includes rock type R, rainfall P and seismic activity intensity E, CRI = w L ×L+w W ×W+w SR ×SR+w θ ×θ+w R ×R+w P ×P+w E ×E,w L 、w W 、w SR 、w θ 、w R 、w P and w E are the weight coefficients of each indicator; Cracks are divided into three risk levels according to the comprehensive risk index, and risk visualization is performed.

6. The method for extracting dangerous rock crack features based on a real-scene three-dimensional model according to claim 3 is characterized in that: When determining the oblique photography model, sparse point clouds are generated based on SIFT / ORB feature matching, dense point clouds are generated through semi-global matching, three-dimensional mesh models are generated through Poisson surface reconstruction, and textured three-dimensional models are obtained by combining UV mapping projection textures. When determining the lidar point cloud model, the ICP algorithm is used to align multi-view point clouds, terrain features are classified through random forest / PointNet++, and digital elevation models are generated through RANSAC plane fitting.

7. A dangerous rock crack feature extraction system based on real-scene three-dimensional model, characterized in that: The system comprises: A data acquisition and processing module is used to acquire multi-view image data based on the oblique photography camera and to acquire point cloud data based on the laser radar; The 3D model fusion module is used to determine the oblique photography model based on the multi-view image data and the laser radar point cloud model based on the point cloud data to obtain the 3D model of the dangerous rock area; The crack feature determination module is used to identify cracks in the 3D model through edge detection algorithm and extract crack types and crack geometric features through deep learning algorithm; The risk assessment system module is used to retrieve geological background data and establish a risk assessment indicator system based on geological background data, fracture types and fracture geometric characteristics.

8. The dangerous rock crack feature extraction system based on real-scene three-dimensional model according to claim 7 is characterized in that: The three-dimensional model fusion module includes: The oblique photography model unit is used to process the multi-view image data through multi-view image matching and stereo reconstruction technology, construct the three-dimensional geometric structure of the dangerous rock area, and determine the oblique photography model; The laser point cloud model unit is used to generate a digital elevation model using point cloud data, extract the terrain features of the dangerous rock area through point cloud classification and segmentation technology, and determine the laser radar point cloud model; The model fusion unit is used to fuse the oblique photography model with the lidar point cloud model to generate a three-dimensional model of the dangerous rock area.

9. The dangerous rock crack feature extraction system based on real-scene three-dimensional model according to claim 7, characterized in that: The crack feature determination module comprises: An edge detection and recognition unit, used to recognize the position of the crack and the edge area of ​​the crack using a Canny edge detection algorithm; A crack feature extraction unit, used to identify the crack type through a convolutional neural network model and extract crack geometric features, wherein the crack geometric features include length L, width W and strike θ; The characteristic statistical analysis unit is used to perform statistical analysis on the extracted fracture geometric characteristics and calculate the number, density and expansion rate SR of the fractures.

10. The dangerous rock crack feature extraction system based on real-scene three-dimensional model according to claim 9, characterized in that: The risk assessment system modules include: The risk index calculation unit is used to determine the comprehensive risk index CRI, where the geological background data includes rock type R, rainfall P and seismic activity intensity E, CRI = w L ×L+w W ×W+w SR ×SR+w θ ×θ+w R ×R+w P ×P+w E ×E,w L 、w W 、w SR 、w θ 、w R 、w P and w E are the weight coefficients of each indicator; The risk level classification unit is used to classify cracks into three risk levels according to the comprehensive risk index and to visualize the risks.

Images