An open-pit mine landslide body emergency monitoring system and method based on smartphone photogrammetry

Abstract

The present application belongs to the technical field of open-pit mine safety monitoring and photogrammetry, and discloses an open-pit mine landslide emergency monitoring system and method based on smartphone photogrammetry, which comprises the following steps: firstly, calibrate the smartphone camera using a calibration board in advance to obtain the intrinsic matrix and distortion coefficient of the camera; secondly, take pictures automatically and synchronously at multiple stations, and return the images; thirdly, extract and match the features of the returned images to construct a joint sparse homonymic point set; fourthly, estimate the essential matrix and recover the relative attitude based on the joint sparse homonymic point set; fifthly, calculate the spatial coordinates of the sparse three-dimensional point cloud and the center of the marker circle by the line-of-sight forward intersection method based on the essential matrix and the relative attitude; sixthly, restore the real physical scale by using the baseline length; and finally, calculate the cross-time three-dimensional displacement relative to the reference based on the real physical scale.

Description

Technical Field

[0001] This invention belongs to the field of open-pit mine safety monitoring and photogrammetry technology, specifically relating to an emergency monitoring system and method for open-pit mine landslides based on smartphone photogrammetry. Background Technology

[0002] In open-pit mining, deep subsidence mining often leads to slope instability and the formation of large landslides. Once formed, these landslides tend to be in a state of continuous creep, making them extremely prone to sudden collapses and posing a very high safety risk. Current technologies for emergency monitoring in such scenarios have the following limitations: 1. High risks associated with contact surveying: Total station prism monitoring or GPS monitoring both require surveyors to carry equipment onto the surface of the sliding body for deployment. Since the sliding body may collapse at any time, personnel cannot enter, rendering these conventional methods ineffective.

[0003] 2. Expensive equipment is difficult to deploy: Although slope radar is a non-contact monitoring device, the equipment is extremely expensive and bulky, and it cannot track the displacement of material points, making it difficult to quickly and effectively monitor landslides in emergency situations.

[0004] 3. Traditional photogrammetry has many limitations: Existing UAV or ground-based photogrammetric monitoring technologies typically rely on large commercial software (such as ContextCapture, Agisoft Metashape, etc.), which involves excessive computation and cannot meet the requirements of "real-time monitoring." UAV aerial photography requires regular flights, involves a large amount of fieldwork, is easily affected by weather, and suffers from data processing delays, making it difficult to achieve "real-time" monitoring. More critically, traditional methods require the establishment of ground image control points (GCPs) within the survey area to recover absolute coordinates through aerial triangulation. However, it is impossible to establish GCPs in sliding body areas, resulting in the inability to obtain accurate physical displacement data. Summary of the Invention

[0005] To address the challenges of inaccessible personnel, expensive equipment, and difficulty in real-time calculation in existing emergency monitoring of landslides in open-pit mines, this invention provides an emergency monitoring system and method for landslides in open-pit mines based on smartphone photogrammetry. This solution is a contactless, image control-free, high-precision, and low-cost real-time landslide monitoring technology. By integrating natural feature points and artificial markers, it achieves rapid and robust pose tracking of sparse point clouds on ordinary equipment.

[0006] To achieve the above objectives, the present invention provides the following solution: An emergency monitoring system for landslides in open-pit mines based on smartphone photogrammetry is described. The system construction process includes: setting up multiple monitoring stations on the safety steps around the landslide, with each monitoring station consisting of a fixed baseline stereo camera composed of two smartphones and a rigid support, combined with a flat dot array marker dropped by a drone onto the surface of the landslide, to form a non-contact displacement monitoring system that does not require personnel to enter the landslide to set up image control points, i.e., an emergency monitoring system for landslides in open-pit mines.

[0007] This invention also provides an emergency monitoring method for landslides in open-pit mines based on smartphone photogrammetry. The method is implemented using the aforementioned system and includes: The smartphone camera was calibrated in advance indoors using a calibration board to obtain the camera's intrinsic parameter matrix and distortion coefficients; Multiple mobile phones can automatically take photos at set times and send the images back; Feature extraction and matching are performed on the returned images to construct a joint sparse set of corresponding points; Based on the joint sparse set of corresponding points, perform essential matrix estimation and relative pose recovery; Based on the essential matrix and relative attitude, the spatial coordinates of the sparse 3D point cloud and the center of the marker are calculated by the line-of-sight intersection method. Based on the spatial coordinates of the center of the sparse 3D point cloud and the marker, the true physical scale is recovered using the baseline length. Based on real physical scales, the time-series three-dimensional displacement relative to the initial reference is calculated in a fixed camera coordinate system.

[0008] Preferably, the method for calibrating a smartphone camera and obtaining the camera's intrinsic parameter matrix and distortion coefficients includes: Global coordinates of the object point Provided by a flat checkerboard pattern, the flat checkerboard pattern is displayed on the monitor; Images were taken from different angles, and the image coordinates of the checkerboard corner points were automatically identified using OpenCV open-source functions, along with the object coordinates of the corner points. Input the central perspective projection equation together In the formula , Pixel focal length, , For the coordinates of the optical center, , , , Given the rotation and translation transformation matrices, i,j=1,2,3, we obtain the intrinsic parameter matrix. K and distortion coefficient .

[0009] Preferably, the method for extracting and matching features from the returned image to construct a joint sparse set of corresponding points includes: use Operators extract textured blobs from the rock surface; Using OpenCV library functions and algorithms, the center of a dot array on a flat marker was identified. By merging the texture blobs and the center of the dot array, and matching the left and right images, we can obtain pairs of images from the same source.

[0010] Preferred, use Methods for extracting textured blobs from rock surfaces using operators include: For rock mass image functions f ( x , y ), Introduction LoG Convolution kernel: ; ; In the formula, For planar image coordinates, Gaussian function variance Gaussian function Convolution of an image; Perform convolution operation on the image: ; in, For gradient; use Sensitivity to spotted blobs, through 3D spatial detection response function Local extrema are used to locate the characteristic points of a blob; The SIFT algorithm's descriptive vectors are introduced for feature point matching to find pairs of originating points.

[0011] Methods for relative pose recovery based on a joint sparse set of corresponding points include: Essential Matrix Tie Points of the same origin ( Connect them: ; ; in , Using at least five pairs of points of common origin, assemble them into a system of linear algebraic equations: ; ; Using SVD matrix factorization technique Solve the equation. U =[ u 1 ,u 2 ,…,u m ] is an m-dimensional orthogonal matrix composed of left singular vectors. V =[ v 1 ,v 2 ,…,v n ] is an n-dimensional orthogonal matrix composed of right singular vectors. It is an m*n Victoire singular matrix, and the solution to E is the last singular vector of V; the essential matrix is ​​obtained through... Encapsulated coordinate system rotation matrix R With translation vector t ; Perform SVD decomposition on the essential matrix: ; Introduction ; This yields four relative orientation modes: ; ; In the formula, , Rotation matrix of coordinate system R Alternative solutions, yes The third column; based on the cherality condition, select the solution where the intersection of the lines of sight is in front of the two cameras ( ).

[0012] Methods for calculating the spatial coordinates of sparse 3D point clouds and the centers of landmarks using the line-of-sight intersection method include: Based on the camera's interior and exterior orientation projection equations The full projection equation is obtained as follows: ; In the formula, This is the scaling factor; Point of object The homogeneous coordinates in the global coordinate system are: ; object point Xi In the camera j The pixel coordinates are Substitute into the projection equation and the cameraj projection matrix Write it in line format: ,get: ; Two linear equations were obtained: ; A system of linear algebraic equations expressed in matrix form: ; In the formula, It is a coefficient matrix; If every point is seen in at least two photographs, then a unified system of equations can be assembled: ; Using the DLT method Solve for the least squares solution to obtain the global coordinates of the object point. .

[0013] Methods for recovering the true physical scale using baseline length include: Based on the essential matrix estimated by the corresponding points and the relative attitude, the distance between cameras ||t|| at the current scale is calculated using the dimensionless translation vector t. Then, the ratio of the measured baseline length B to ||t|| is used as the scale factor to uniformly scale the coordinates of all three-dimensional points to obtain the true physical scale.

[0014] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Existing smartphone-based systems primarily target indoor slope models or small, safely accessible slopes, typically requiring the application of coded markers or artificial targets to the surface of the object being monitored. This invention addresses the emergency monitoring scenario of large landslides in open-pit mines, where the landslide is continuously moving and at risk of collapse at any time, making it impossible for personnel to enter the landslide to affix markers or deploy control points. Therefore, this invention proposes using drones to airdrop flat-panel markers (with a planar array of dots) onto the landslide surface, allowing for the deployment of stable markers without requiring personnel to ascend the slope.

[0015] 2. Existing multi-smartphone systems generally require ground control points or external benchmarks to recover the true scale of the 3D model. This invention uses a rigid support set up on a stable step around the sliding body to form a pair of stereo cameras with a fixed baseline. By accurately measuring the baseline length between the optical centers of the two cameras, and solving for the relative attitude, the true scale of the point cloud is recovered using the baseline length. Thus, the true physical displacement of the sliding body can be obtained without setting up traditional image control points.

[0016] 3. Existing systems mostly use general-purpose commercial SfM software to process large amounts of imagery, resulting in high computational costs and complex processes. This invention, targeting emergency monitoring scenarios, designs a lightweight photogrammetry algorithm based on Python and OpenCV. It utilizes SIFT features and a marker dot array to construct sparse corresponding points for stereo image pairs. First, it solves for the essential matrix and relative pose, then directly generates sparse point clouds of the sliding body and markers through two-view forward intersection. This avoids complex multi-view joint adjustment and dense reconstruction, significantly reducing computational load and achieving near real-time displacement monitoring.

[0017] 4. This invention deploys several multi-smartphone monitoring stations on multiple safety steps around the landslide and enables remote, automatic, timed photo taking and data transmission via WIFI or 4G / 5G wireless networks, forming a rapidly deployable and scalable landslide emergency monitoring network. It is particularly suitable for monitoring the needs of multi-step, large-scale landslides in open-pit mines.

[0018] In summary, this invention has significant differences and improvements over the closest existing technology in terms of the method of landslide marker placement, scale restoration mechanism, algorithm process and system deployment, and can solve the core problems existing in the emergency monitoring of landslides in open-pit mines. Attached Figure Description

[0019] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is an overall layout diagram of the mobile phone photography measurement system according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the dual-smartphone monitoring station structure according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the operation and data processing of the monitoring system according to an embodiment of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0023] Example 1 like Figure 1 , Figure 2 As shown, this invention provides an emergency monitoring system for landslides in open-pit mines based on smartphone photogrammetry. In particular, it relates to a system that utilizes stereo images from smartphones combined with markers dropped by drones to perform non-contact real-time displacement monitoring of dangerous landslides without ground control points (GCPs). The system construction process includes: setting up multiple monitoring stations on safety steps around the landslide, each monitoring station consisting of a fixed baseline stereo camera composed of two smartphones and a rigid support, combined with a flat dot array marker dropped by a drone onto the surface of the landslide, to form a non-contact displacement monitoring system that does not require personnel to enter the landslide to set up control points, i.e., an emergency monitoring system for landslides in open-pit mines.

[0024] Specifically, several monitoring stations were set up on the safety steps surrounding the landslide. Each monitoring station contained at least two smartphones, fixed on a bracket, taking pictures of the adjacent landslide to form a "stereoscopic image pair." The two phones transmitted the images to the monitoring center computer via Wi-Fi. In addition, drones were used to airdrop flat markers with a grid of circles printed on their surface onto the landslide surface as the targets to be monitored.

[0025] Example 2 like Figure 3 As shown, the present invention also provides an emergency monitoring method for landslides in open-pit mines based on smartphone photogrammetry. This method is implemented using the system described in Example 1, and includes: The smartphone camera was calibrated in advance indoors using a calibration board to obtain the camera's intrinsic parameter matrix and distortion coefficients; Multiple mobile phones can automatically take photos at set times and send the images back; Feature extraction and matching are performed on the returned images to construct a joint sparse set of corresponding points; Based on the joint sparse set of corresponding points, perform essential matrix estimation and relative pose recovery; Based on the essential matrix and relative attitude, the spatial coordinates of the sparse 3D point cloud and the center of the marker are calculated by the line-of-sight intersection method. Based on the spatial coordinates of the center of the sparse 3D point cloud and the marker, the true physical scale is recovered using the baseline length. Based on real physical scales, calculate the time-series three-dimensional displacement relative to a reference.

[0026] The specific implementation process is as follows: ① Camera calibration: The smartphone camera is precisely calibrated indoors using a calibration board to obtain the camera's intrinsic parameter matrix. K ) and distortion coefficient ( ), to eliminate the effects of lens distortion.

[0027] In the formula , Pixel focal length, , For the coordinates of the optical center, , For image point coordinates, , For coordinate error, Let the distance be from the image point to the optical center. Scaling factor , , , For rotation and translation transformation matrices, i , j =1,2,3, global coordinates of the object point Provided by a planar checkerboard pattern, the planar checkerboard pattern is displayed on a large screen. Photos are then taken from different angles, and OpenCV open-source functions are used to automatically identify the image coordinates of the checkerboard corner points, along with their object coordinates. Enter the central perspective projection equation together: Solving the equation yields the intrinsic parameter matrix. K and distortion coefficient .

[0028] ② Hybrid Feature Extraction: For the returned stereo image pairs, a new approach is proposed to achieve high-frequency real-time monitoring by fully utilizing the natural texture characteristics of the rock surface and the features of this photogrammetric system, and employing... The Laplacian of Gaussian operator directly extracts texture spots from the rock surface on a single-scale image. Since the distances of the two phones to the target object are approximately the same, operation can be performed at a single scale, saving significant computation time and enabling real-time computation. This is applicable to two-dimensional grayscale image functions of rock masses. f ( x , y Introduce the following LoG convolution kernel: in, For planar image coordinates, Gaussian function variance Gaussian function Convolution of an image. Perform the following convolution operation on the image: in, For gradient.

[0029] use Sensitivity to spotted blobs, through 3D spatial detection response function Local extrema can be used to locate the feature points of a blob. Then, the feature description vectors of the SIFT algorithm are directly introduced for feature point matching. By calculating the spatial Euclidean distance, the similarity between the two description vectors is compared to find pairs of originating points.

[0030] On the other hand, OpenCV's open-source library functions are used to detect the center coordinates of the dot array on the flat marker. Since the source point of a dot in another photo must lie on its epipolar line, dot matching on the left and right images can be performed through this epipolar geometric constraint. Texture spots and dots together form image pair tiepoints.

[0031] ③ Pose calculation and point cloud generation: Solving the essential matrix using hybrid Tie Points. (EssentialMatrix) is used to recover the relative poses of the left and right cameras (rotation matrix R and translation vector t). Then, the spatial coordinates of the sparse 3D point cloud and the center of the marker circle are calculated using the line-of-sight intersection method (Triangulation). Essential Matrix Tie Points can be from the same source ( Connect them: in , Using at least five pairs of points of common origin, a system of linear algebraic equations can be assembled: Singular Value Decomposition (SVD) technique of matrices is employed. Solving the above homogeneous linear algebra equations, for any m*n dimensional real matrix, can be done by the following decomposition: in U=[ u 1 ,u 2 ,…,u m ] is an m-dimensional orthogonal matrix composed of left singular vectors. V =[ v 1 ,v 2 ,…,v n ] is an n-dimensional orthogonal matrix composed of right singular vectors. It is an m*n Vickers unique matrix. The solution to E is... V The last singular vector v n The essential matrix encapsulates the coordinate system rotation matrix using the following formula. R With translation vector t : Perform Singular Value Decomposition (SVD) on the essential matrix: Introduction This yields four relative orientation modes: In the above formula, , Rotation matrix of coordinate system R Alternative solutions, yes The third column. Then, based on the cherality condition, select the solution where the intersection of the lines of sight is in front of both camera lenses ( This completes the relative orientation of the two cameras. To save computation and achieve real-time monitoring, relative orientation can be performed only once a day, after which the coordinates of the target points in the dot array can be calculated at high frequency.

[0032] Based on the camera's inward and outward orientation, the projection equation can be obtained. Now we can write the equations for the full projection: In the formula This is the scaling factor. (Object point) The homogeneous coordinates in the global coordinate system are: Point of object Xi In the camera j The pixel coordinates are Substitute into the projection equation and the camera j projection matrix Write it in line format: Therefore, we can conclude that: Simplifying, we obtain two linear equations: A system of linear algebraic equations expressed in matrix form: in, Let be the coefficient matrix. If each object point is seen in at least two photographs, then a unified system of equations can be assembled: Considering measurement errors, these equations are incompatible. Therefore, the mathematical technique of Singular Value Decomposition (SVD) is employed. Find the least squares solution, where U =[ u 1 ,u 2 ,…,u m ] is an m-dimensional orthogonal matrix composed of left singular vectors. V =[ v 1 ,v 2 ,…,v n ] is an n-dimensional orthogonal matrix composed of right singular vectors. It is an m*n Victoire unique matrix. The solution is V The last singular vector v n At this point, the global coordinates of the object point have been determined. .

[0033] ④ Scale restoration (control point-free technique): Measure the physical baseline distance between the two mobile phones in the stereo image pair, and use this distance to absolutize the relative scale of the reconstructed model, thereby obtaining the true physical coordinates.

[0034] ⑤ Displacement Calculation: Within the framework of close-range photogrammetry in this case, since the camera position is fixed, the camera coordinate system position is also fixed. By using one of the camera coordinate systems as the global coordinate system, the change in the coordinates of the center of the marker can be tracked in real time, enabling centimeter or even millimeter-level displacement monitoring of the mine rock mass.

[0035] In summary, the technical effects of this invention are as follows: 1. High safety: No contact is required throughout the process, and no personnel need to enter the sliding body.

[0036] 2. No image control points required: The scale is recovered by baseline distance, which breaks through the limitation of traditional photogrammetry that requires the establishment of GCPs.

[0037] 3. High accuracy and stability: By combining natural feature points blobs of the rock mass with an array of artificial marker circles, the robustness of the pose solution (due to the large number of points) is guaranteed, while also enabling precise tracking of specific points.

[0038] 4. Real-time and low cost: The algorithm only processes sparse point clouds, and the amount of computation is much less than that of commercial software. It can achieve minute-level real-time monitoring on ordinary computers; the cost of mobile phone hardware is extremely low, making it suitable for large-scale, multi-level deployment.

[0039] Example 3 The specific steps for implementing this invention are as follows: Step 1: Indoor camera calibration Before using mobile phones for monitoring, each smartphone camera was calibrated in a laboratory environment using Zhang Zhengyou's calibration method (photographing a checkerboard pattern). The intrinsic parameter matrix K (including focal lengths fx and fy, and optical centers cx and cy) and radial / tangential distortion coefficients were calculated. This step ensured the geometric accuracy of subsequent measurements.

[0040] Step Two: On-site Emergency Deployment 1. Site selection: The monitoring station shall be set up on a stable and safe step opposite the sliding body.

[0041] 2. Network setup: Two mobile phones (left and right cameras) are installed at each station and connected to the on-site WIFI.

[0042] 3. Baseline Measurement: Using a high-precision steel ruler or a handheld laser rangefinder, accurately measure the horizontal distance B_real (e.g., 1.500 meters) between the optical centers of the left and right cameras as the system's scale reference.

[0043] 4. Dropping markers: Operate a drone carrying specially designed flat markers (with an array of dots at known intervals) and airdrop them onto key monitoring areas on the surface of the landslide, ensuring that the mobile phone's field of view can simultaneously cover the landslide rock background and the markers.

[0044] Step 3: Data Acquisition and Processing (Python Program) 1. Image Acquisition: The phone takes photos and uploads them synchronously at set intervals (e.g., every 15 or 30 minutes).

[0045] 2. Distortion correction: Call the cv2.undistort function to correct the image using the intrinsic parameters from step one.

[0046] 3. Feature extraction and matching (hybrid mode): For rock mass areas: use The operator extracts a large number of natural feature points blobs from the rock mass. These points are invariant to illumination, rotation, scale, and perspective distortion, and can provide robust geometric constraints.

[0047] For the marker area: use cv2.findCirclesGrid to detect the center of the circle array. The center coordinate extraction accuracy can reach the sub-pixel level.

[0048] Matching: Merge the two types of points mentioned above, perform left and right image matching, and obtain the same image point pairs (Tie Points).

[0049] 4. Relative Orientation: Based on the matching point pairs, the essential matrix E is calculated using epipolar geometry constraints, and mismatches are eliminated using the RANSAC algorithm to obtain pure homologous image point pairs (Tie Points). Then, the rotation matrix R and the normalized translation vector t_norm are obtained through decomposition.

[0050] 5. Absolute coordinate restoration: Calculate the scaling factor: Scale = B_real / ||t_norm||.

[0051] Obtain the true translation vector t_real = t_norm * Scale.

[0052] Using the principle of line-of-sight intersection, the three-dimensional spatial coordinates (X, Y, Z) of the center of the marker are calculated.

[0053] Step 4: Displacement Analysis and Early Warning Using the coordinates of the marker center obtained in the initial calculation as a reference, the Euclidean distance between the current coordinates and the reference coordinates is calculated. When the cumulative displacement or displacement rate exceeds the set safety threshold (e.g., 10 mm / h), the system automatically triggers an alarm, providing data support for personnel and equipment evacuation and landslide control.

[0054] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. An emergency monitoring system for landslides in open-pit mines based on smartphone photogrammetry, characterized in that, The construction process of the system includes: setting up multiple monitoring stations on the safety steps around the landslide, each monitoring station consisting of two smartphones and a rigid bracket forming a fixed baseline stereo camera, combined with flat dot array markers airdropped by drones onto the surface of the landslide, together forming a non-contact displacement monitoring system that does not require personnel to enter the landslide to set up image control points, namely the open-pit mine landslide emergency monitoring system.

2. An emergency monitoring method for landslides in open-pit mines based on smartphone photogrammetry, wherein the method is implemented using the system described in claim 1, characterized in that, The method includes: The smartphone camera was calibrated in advance indoors using a calibration board to obtain the camera's intrinsic parameter matrix and distortion coefficients; Multiple mobile phones can automatically take photos at set times and send the images back; Feature extraction and matching are performed on the returned images to construct a joint sparse set of corresponding points; Based on the joint sparse set of corresponding points, perform essential matrix estimation and relative pose recovery; Based on the essential matrix and relative attitude, the spatial coordinates of the sparse 3D point cloud and the center of the marker are calculated by the line-of-sight intersection method. Based on the spatial coordinates of the center of the sparse 3D point cloud and the marker, the true physical scale is recovered using the baseline length. Based on real physical scales, calculate the time-series three-dimensional displacement relative to a reference.

3. The method according to claim 2, characterized in that, Methods for calibrating smartphone cameras and obtaining the camera's intrinsic parameter matrix and distortion coefficients include: Global coordinates of the object point Provided by a flat checkerboard pattern, the flat checkerboard pattern is displayed on the monitor; Images were taken from different angles, and the image coordinates of the checkerboard corner points were automatically identified using OpenCV open-source functions, along with the object coordinates of the corner points. Input the central perspective projection equation together In the formula , Pixel focal length, , For the coordinates of the optical center, , , , Given the rotation and translation transformation matrices, i,j=1,2,3, we obtain the intrinsic parameter matrix. K and distortion coefficient .

4. The method according to claim 3, characterized in that, Methods for extracting and matching features from returned images to construct a joint sparse set of corresponding points include: use Operators extract textured blobs from the rock surface; Using OpenCV library functions and algorithms, the center of a dot array on a flat marker was identified. By merging the texture blobs and the center of the dot array, and matching the left and right images, we can obtain pairs of images from the same source.

5. The method according to claim 4, characterized in that, use Methods for extracting textured blobs from rock surfaces using operators include: For rock mass image functions f ( x , y Introducing LoG convolution kernels: ； ； In the formula, For planar image coordinates, Gaussian function variance Gaussian function Convolution of an image; Perform convolution operation on the image: ； in, For gradient; use Sensitivity to spotted blobs, through 3D spatial detection response function Local extrema are used to locate the characteristic points of a blob; The SIFT algorithm's descriptive vectors are introduced for feature point matching to find pairs of originating points.

6. The method according to claim 5, characterized in that, Methods for relative pose recovery based on a joint sparse set of corresponding points include: Essential Matrix Tie Points of the same origin ( Connect them: ； ； in , Using at least five pairs of points of common origin, assemble them into a system of linear algebraic equations: ； ； Using SVD matrix factorization technique Solve the equation. U =[ u 1 ,u 2 ,…,u m ] is an m-dimensional orthogonal matrix composed of left singular vectors. V =[ v 1 ,v 2 ,…,v n ] is an n-dimensional orthogonal matrix composed of right singular vectors. It is an m*n Victoire singular matrix, and the solution to E is the last singular vector of V; the essential matrix is ​​obtained through... Encapsulated coordinate system rotation matrix R With translation vector t ; Perform SVD decomposition on the essential matrix: ； Introduction ; This yields four relative orientation modes: ； ； In the formula, , Rotation matrix of coordinate system R Alternative solutions, yes The third column; based on the cherality condition, select the solution where the intersection of the lines of sight is in front of the two cameras ( ).

7. The method according to claim 6, characterized in that, Methods for calculating the spatial coordinates of sparse 3D point clouds and the centers of landmarks using the line-of-sight intersection method in a fixed camera coordinate system include: Based on the camera's interior and exterior orientation projection equations The full projection equation is obtained as follows: ； In the formula, This is the scaling factor; object point The homogeneous coordinates in the global coordinate system are: ; object point Xi In the camera j The pixel coordinates are Substitute into the projection equation and the camera j projection matrix Write it in line format: ,get: ； Two linear equations were obtained: ； A system of linear algebraic equations expressed in matrix form: ； In the formula, It is a coefficient matrix; If every point is seen in at least two photographs, then a unified system of equations can be assembled: ； Using the DLT method Solve for the least squares solution to obtain the global coordinates of the object point. .

8. The method according to claim 7, characterized in that, Methods for recovering the true physical scale using baseline length in a fixed camera coordinate system include: Based on the essential matrix estimated by the corresponding points and the relative attitude, the distance between cameras ||t|| at the current scale is calculated using the dimensionless translation vector t. Then, the ratio of the measured baseline length B to ||t|| is used as the scale factor to uniformly scale the coordinates of all three-dimensional points to obtain the true physical scale.

Images