A high and steep slope dangerous rock monitoring method based on point cloud-tilt photography technology

Abstract

This invention discloses a method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology, comprising: firstly, using UAV oblique photography technology to collect oblique photography data of dangerous rocks in a steep slope area, and performing aerial triangulation on the oblique photography data; then, using a UAV equipped with a laser point cloud acquisition device to collect laser point cloud data of the steep slope area, and simultaneously preprocessing the laser point cloud data to obtain noise-reduced and redundancy-removed laser point cloud data; then, using a point cloud registration algorithm to register the aerial triangulation dense point cloud model and the laser point cloud data, thereby generating a high-precision three-dimensional mesh model and point cloud model; finally, using a mesh model measurement tool and a point cloud comparison algorithm to compare the measurement models of dangerous rocks on steep slopes at different times in multiple periods, thereby achieving effective monitoring of dangerous rocks on steep slopes.

Description

Technical Field

[0001] This invention relates to the field of disaster prevention and mitigation technology, and in particular to a method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology. Background Technology

[0002] Dangerous rocks refer to rock masses whose stability has been reduced due to natural or human factors, posing a risk of collapse, rolling, or other geological hazards. These geological disasters pose a serious threat to people's lives and property, with dangerous rockfalls on steep slopes being particularly severe. Therefore, monitoring dangerous rocks on steep slopes is crucial. Currently, dangerous rock monitoring technologies mainly include manual patrols, establishing field monitoring points, and installing monitoring equipment. However, due to the common problem of inaccessible areas and difficulties in establishing monitoring points, current real-time monitoring methods primarily utilize GPS or various sensor devices.

[0003] The GPS monitoring method utilizes global satellite positioning technology to determine the precise location changes of dangerous rock slopes in real time through GPS transceivers installed in the monitoring area. This method is fast, highly accurate, and can achieve all-weather automated real-time monitoring. It is particularly suitable for slope environments with large terrain undulations and poor visibility. Currently, many scholars are also working on improvements to the GPS monitoring method to further enhance monitoring accuracy. For example, when Li Long et al. constructed a high-precision positioning mathematical model integrating the BeiDou Navigation Satellite System (BDS) and the Global Positioning System (GPS), they designed a robust GNSS observation data preprocessing strategy based on the principle of consistency between pre- and post-episode data. This strategy aims to simultaneously ensure the high precision requirements and rapid response characteristics of the monitoring algorithm. By optimizing the data processing flow, redundancy and repetitive operations are reduced, thereby improving the overall monitoring efficiency and accuracy. Chen Yuejuan et al., addressing the limitations of GPS technology when applied alone to surface deformation monitoring, proposed an innovative fusion strategy. This strategy combines precise GPS location data with InSAR wide-area deformation information and employs a Kriging interpolation method optimized by a genetic algorithm. Specifically, it explores and selects the optimal variogram model to estimate the annual average deformation rate of the study area. Furthermore, a genetic algorithm is introduced to fine-tune the key parameters of the variogram, improving the prediction accuracy and adaptability of the interpolation model. Finally, using measured data from GPS stations as a benchmark, the performance of the fusion model is comprehensively verified and evaluated to ensure the reliability and accuracy of the monitoring results. GPS monitoring offers advantages such as high precision, high efficiency, and global coverage in monitoring dangerous rocks. However, it also has some drawbacks. For example, when a GPS receiver is located in a mountainous area, densely wooded area, or valley, it may receive reflected signals from multiple directions, a phenomenon known as multipath effect, leading to unstable or even failed positioning results. GPS is highly dependent on the number of satellites; insufficient or unevenly distributed satellites will affect the accuracy and reliability of monitoring. GPS devices consume a significant amount of power during continuous operation, especially when battery-powered, requiring frequent battery replacements or alternative power supply measures to ensure continuous monitoring. Sensor-based methods, on the other hand, involve deploying various sensors such as pressure gauges, crack measuring devices, and rain gauges in the monitoring area to acquire information on pressure, angle, displacement, and rainfall. This data is received and processed in a backend system to enable remote monitoring and early warning. Patent CN116448883B proposes a method for monitoring and early warning of dangerous rock masses on steep cliffs. This patent involves installing acoustic emission monitoring holes on the parent rock on both sides of the dangerous rock mass's rock bridge, with acoustic emission monitoring sensors inside the holes. This allows for dynamic monitoring of stress changes leading to fracture events during the propagation and penetration of cracks in the dangerous rock mass. This method not only improves the accuracy and safety of monitoring but also effectively avoids damage to the original stress state of the dangerous rock mass.Patent CN219392758U proposes a monitoring and early warning system for rock mass collapse. This patent utilizes a combination of pressure sensors, infrared obstacle avoidance sensors, and other equipment to achieve real-time monitoring and early warning of rock mass displacement, improving the timeliness and accuracy of early warnings. While sensor-based methods offer advantages such as high monitoring accuracy, low cost, and easy equipment portability, long-term operation in harsh environments may lead to aging of electronic components and poor stability. Furthermore, sensor installation may be difficult on complex terrains such as steep slopes. Additionally, monitoring can only target predicted danger points, making it less suitable for slopes with scattered, large areas of unstable rocks, or slopes prone to sudden rockfall disasters.

[0004] Laser point cloud and oblique photogrammetry are important technologies in modern surveying and geographic information systems, enabling effective monitoring of areas prone to rockfall. Chu Hongliang et al. addressed the problem that monitoring rock mass deformation on reservoir banks primarily relies on fixed-point observations, which struggles to detect overall rock mass changes. They utilized terrestrial three-dimensional laser scanning to acquire centimeter-precision point cloud data of the entire rock mass surface, effectively monitoring reservoir bank rock mass deformation. However, this method uses a terrestrial laser scanner, making it difficult to reach steep slopes for measurement, and terrestrial laser scanning technology also suffers from incomplete scanning and accuracy deviations. Kang Chenyun et al. used UAV oblique photogrammetry to identify, acquire geometric features, and analyze boundary conditions of dangerous rocks. They employed stereographic projection and rigid body limit equilibrium methods to analyze the stability of dangerous rock zones and masses, respectively, effectively monitoring high-level dangerous rocks. However, this method has lower monitoring accuracy compared to laser point cloud methods, exhibiting significant data deviations.

[0005] In general, existing technologies for monitoring dangerous rocks mainly rely on methods such as regular manual patrols, establishing fixed field monitoring points, and installing real-time monitoring devices. Among these, manual patrols generally suffer from problems such as untimely monitoring and blind spots; field monitoring points are generally difficult to place monitoring equipment on and are hard to reach; and installing monitoring devices has problems such as unstable monitoring, the need for frequent maintenance, or easy damage to the monitoring devices. Summary of the Invention

[0006] The purpose of this invention is to overcome the above-mentioned shortcomings and provide a method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology. This method is suitable for flexible monitoring of dangerous rocks in various complex steep slope landforms that are inaccessible to humans, thereby achieving non-contact, convenient, flexible, high-precision, and large-area monitoring of dangerous rocks on steep slopes.

[0007] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology, which includes the following steps:

[0008] Step 1: Use UAV oblique photography technology to collect oblique photography data of dangerous rocks in the steep slope area, and use ContextCapture to build an aerial triangulation dense point cloud model of the steep slope area.

[0009] Step 2: Use a drone equipped with a laser point cloud acquisition device to collect laser point cloud data in the steep slope area. At the same time, perform voxel filtering on the laser point cloud data to remove invalid points and reduce the amount of point cloud data.

[0010] Step 3: Use a point cloud registration algorithm to register the aerial triangulation dense point cloud model and the laser point cloud data to obtain a high-precision point cloud model of the steep slope area after the fusion of oblique photography and point cloud data;

[0011] Step 4: Based on the high-precision point cloud model of the steep slope area, use ContextCpature to reconstruct the three-dimensional mesh to obtain a high-precision three-dimensional mesh model of the steep slope area.

[0012] Step 5: Use grid model measurement tools or point cloud comparison algorithms to compare measurement models from different periods in the high and steep slope area, thereby achieving effective monitoring of dangerous rocks on high and steep slopes.

[0013] Furthermore, the specific process of step one is as follows:

[0014] UAV oblique photography technology was used to collect oblique photography data of dangerous rocks in the steep slope area. The collected oblique photography data was then imported into ContextCapture. Next, the integrity of the images was checked and substandard oblique photography images were removed. Finally, aerial triangulation was performed to obtain an aerial triangulation dense point cloud model of the steep slope area.

[0015] Furthermore, in step two, the specific steps for voxel filtering the laser point cloud data are as follows:

[0016] Step 2.1: Construct a spatial voxel mesh;

[0017] Step 2.2: Calculate the voxel index of the point cloud data;

[0018] Step 2.3: Calculate the centroid of each voxel mesh in the point cloud data;

[0019] Step 2.4: Replace the points contained in each voxel grid with the centroid of each voxel grid to complete the filtering process of the point cloud data.

[0020] Furthermore, the specific process of step 2.1 is as follows:

[0021] Assume the laser point cloud data is M = {m1, m2, ..., m} n}, where m1, m2, ..., m n For each individual point in the point cloud data, calculate the minimum cubic boundary of the point cloud data M using the following formula:

[0022]

[0023] In the formula: x max y max z max Let x, y, and z represent the maximum values ​​of the point cloud data M in the x, y, and z directions, respectively; x min y min z min Let M represent the minimum values ​​of point cloud data M in the x, y, and z directions, respectively; N x N y N z represents the number of grid cells in the x, y, and z directions, respectively; l represents the grid size.

[0024] Furthermore, the specific process of step 2.2 is as follows:

[0025] After constructing the spatial voxel mesh of the point cloud, each point m in the point cloud data M... i =(x i ,y i ,z i The formula for calculating the voxel index is as follows:

[0026]

[0027] In the formula: i is a one-dimensional code, through which the voxel grid containing each point in the point cloud data can be directly retrieved; i x i y i z These represent the voxel index values ​​in the x, y, and z directions, respectively; N x N y N z represents the number of grid cells in the x, y, and z directions, respectively; l represents the grid size.

[0028] Furthermore, the specific process of step 2.3 is as follows:

[0029] After determining the correspondence between each point in the point cloud data and the voxel grid using voxel index values, the centroid of each voxel grid is calculated. Then, the point in the voxel grid closest to the centroid is used as the centroid of the voxel grid. The formula for calculating the centroid of the voxel grid is as follows:

[0030]

[0031] In the formula: k is the number of points contained in the voxel grid; G i Denotes the centroid of the i-th voxel mesh; m iThis represents a point located in the i-th voxel grid.

[0032] Furthermore, the specific process of step three is as follows:

[0033] Step 3.1: Assume the aerial triangulation dense point cloud data is the source point cloud P, and the laser point cloud data is the target point cloud Q; firstly, randomly select 4 corresponding points from both the source point cloud P and the target point cloud Q to form four pairs of points. Then, calculate the rigid body transformation matrix H using the least squares method. Next, use the following formula to calculate the new point cloud set P′ of the source point cloud P under the rigid body transformation matrix H:

[0034] P′=R c P+t c

[0035] In the formula: R c t is the rotation matrix; c P' represents the translation amount; P' denotes the new point cloud set of the source point cloud P under the rigid body transformation matrix H.

[0036] Step 3.2: Calculate the distance threshold between corresponding points of P′ and the target point cloud Q. The calculation formula is as follows:

[0037] D = d 2 (p′ i ,Hp i i = 1, 2, 3, 4

[0038] In the formula: D is the distance threshold; d represents the distance between corresponding points of P′ and the target point cloud Q; H represents the transformation matrix of the corresponding points; p′ i This represents the new points in H corresponding to the four selected points in the source point cloud P; p i This represents four corresponding points randomly selected from the source point cloud;

[0039] After determining the distance threshold D between corresponding points of P′ and the target point cloud Q, it is then determined whether the distance between the four pairs of points between P′ and the target point cloud Q is greater than the distance threshold D. If it is greater, the pair of points is considered not to belong to the interior point set I; if it is less, the pair of points is considered to belong to the interior point set I. Then, the registration score S is calculated using the following formula:

[0040]

[0041] In the formula: S is the registration score between the source point cloud P and the target point cloud Q; card(I) represents the cardinality of the interior point set I; card(P) represents the cardinality of the source point cloud P; card(Q) represents the cardinality of the target point cloud Q;

[0042] Step 3.3: Set the registration score threshold S minRepeat steps 3.1 and 3.2 until the calculated registration score S reaches the set registration score threshold S. min The loop ends, and the source point cloud P after coarse registration is obtained. c ;

[0043] Step 3.4: After completing the coarse registration, the dense aerial triangulation point cloud data and the laser point cloud data have been initially aligned. Next, fine registration will be performed to further improve the registration accuracy of the two point cloud data. Fine registration will be performed using the iterative nearest point algorithm that takes into account the normal. The iterative nearest point algorithm uses the minimum distance between corresponding points as the objective function and takes into account the normal information on the basis of the iterative nearest point algorithm.

[0044] Furthermore, step 3.4 specifically includes the following:

[0045] (a) First, find the source point cloud P after coarse registration. c For each point in the target point cloud Q, the point that is closest to it is taken as its corresponding point;

[0046] (b) Calculate the source point cloud P c The optimal translation and rotation parameters for registration between corresponding points in the target point cloud Q are defined by the following objective optimization function:

[0047]

[0048] In the formula: q i Represents a point in the target point cloud Q; This represents the source point cloud P after coarse registration. c The point in; η i For q i arrive The projection onto the surface normal; t f R is the translation vector; f F(t) is the rotation matrix; f ,R f ) represents the optimal rigid body transformation between the corresponding point and the surface after the minimum transformation; F i (t f ,R f ) represents the rigid transformation in the i-th iteration;

[0049] Next, calculate the source point cloud P. c The centroid of the target point cloud Q is calculated using the following formula:

[0050]

[0051] In the formula: p cz Source point cloud P c The center of mass; q z Let n be the centroid of the target point cloud Q; and let n and m be the source point cloud P.c The number of points and the number of points in the target point cloud Q; Source point cloud P c The point in; q i For the points in the target point cloud Q;

[0052] Find P c After finding the center of mass of Q, then for P c Decentralizing Q is achieved using the following formula:

[0053]

[0054] In the formula: Represents the source point cloud P c Decentralized point cloud data; p represents the decentralized point cloud data of the target point cloud Q; cz Source point cloud P c The center of mass; q z The centroid of the target point cloud Q; q i Represents a point in the target point cloud Q; Represents the source point cloud P c The point in the middle;

[0055] Assuming the transformation matrix is ​​G, its calculation formula is as follows:

[0056]

[0057] In the formula: G is the transformation matrix; Represents the source point cloud P c Decentralized point cloud data; This represents the decentralized point cloud data of the target point cloud Q;

[0058] Finally, the translation vector t is calculated. f and rotation matrix R f The calculation formula is as follows:

[0059]

[0060] In the formula: R f Represents the rotation matrix; t f p represents the translation vector; cz Source point cloud P c The center of mass; q z Let Q be the centroid of the target point cloud; A and B represent the orthogonal matrices of the transformation matrix G after SVD decomposition;

[0061] (c) Set the maximum number of iterations N max And the error threshold E, then repeat steps (a) and (b) iteratively, and calculate the error between every two iterations, as shown in the following formula:

[0062] ε=F k (t f ,R f )-F k-1 (t f ,R f )

[0063] In the formula: ε represents the iteration error; F k (t f ,R f ), F k-1 (t f ,R f ) represent the optimal rigid body transformations between the corresponding points and surfaces after the k-th and (k-1)-th minimum transformations, respectively;

[0064] When the iteration error ε is less than or equal to E or the number of iterations reaches N max The iteration is completed, and a high-precision point cloud model of the steep slope area is obtained after fine registration of the aerial triangulation dense point cloud and the laser point cloud.

[0065] Furthermore, the specific process of step four is as follows:

[0066] Based on a high-precision point cloud model of a steep slope area, 3D mesh reconstruction is performed using ContextCpature. First, in the software's spatial frame options, the coordinate system of the dangerous rock monitoring area model is input, and the model's cutting mode is set to regular planar mesh. Using the clipping box function, the range of the output model is adjusted according to the dangerous rock monitoring area. Then, the output model type is selected as a 3D mesh model, the model output format is .OBJ format, texture compression is set to 100%, and adaptive tree is used for the level of detail. Finally, a high-precision 3D mesh model of the steep slope area is generated, resulting in a high-precision 3D mesh model of the steep slope area.

[0067] Furthermore, the specific process of step five is as follows:

[0068] Using the mesh model measurement tool ContextCapture Viewer, high-precision 3D mesh models of steep slope areas measured at different times in multiple periods were quickly compared. ContextCapture Viewer has built-in functions for calculating 3D distance, height difference, surface area, and volume, enabling efficient and rapid measurement of information on unstable rocks within steep slope areas. Assuming two high-precision 3D mesh models M1 and M2 of the same steep slope area measured at different times, the 3D distance measurement function in ContextCaptureViewer was directly used to measure the movement distance of the same rock in M1 and M2 as dis. M1→M2 When dis M1→M2 Distance greater than the safe threshold for the movement of dangerous rocks maxIf so, the rock is considered to be a dangerous rock mass, posing a risk of collapse and landslide.

[0069] Beneficial effects of this invention:

[0070] 1. This invention utilizes a combination of UAV oblique photography and laser point cloud technology to measure data in the same steep slope area. Compared with oblique photography alone, it can greatly improve the accuracy of monitoring data. Compared with laser point cloud technology alone, it can avoid monitoring errors caused by a single measurement method, further improving the accuracy of data monitoring.

[0071] 2. This invention addresses the registration of aerial triangulation dense point clouds and laser point clouds by combining coarse and fine registration methods, which can better improve the registration accuracy of point cloud models and provide good data support for subsequent mesh model reconstruction.

[0072] 3. This invention introduces a rapid model detection and comparison method, which can detect dangerous rock information using a reconstructed high-precision grid model of steep slopes. By comparing the algorithm with the point cloud model, the detection speed is greatly improved when the accuracy is similar, thus realizing efficient monitoring of dangerous rocks in steep slope areas.

[0073] 4. This invention utilizes UAV oblique photography and laser point cloud technology to measure data in the same steep slope area, and performs voxel filtering on the laser point cloud data to remove invalid points and reduce the amount of point cloud data.

[0074] 5. This invention is suitable for flexible monitoring of dangerous rocks in complex and steep slope terrain conditions that are inaccessible to humans, thereby achieving non-contact, convenient, flexible, high-precision, and large-area monitoring of dangerous rocks on steep slopes. Attached Figure Description

[0075] Figure 1 This is a flowchart illustrating a method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology. Detailed Implementation

[0076] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0077] Example 1: A method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology, the process is as follows: Figure 1 As shown. Includes the following steps:

[0078] Step 1: Use UAV oblique photogrammetry technology to collect oblique photogrammetric data of dangerous rocks in the steep slope area. Then, import the collected oblique photogrammetric data into ContextCapture, check the integrity of the images, and remove substandard oblique photogrammetric images. Finally, submit aerial triangulation to obtain an aerial triangulation dense point cloud model of the steep slope area. In this embodiment, aerial triangulation is referred to as aerial triangulation.

[0079] Step 2: Use a drone equipped with a laser point cloud acquisition device to collect laser point cloud data of the steep slope area. During the point cloud data acquisition process, various instrument limitations, environmental fluctuations, and the unique properties of the scanned object inevitably introduce various noise points. These noise points not only unnecessarily increase the size of the point cloud dataset but also significantly impact the accuracy of subsequent 3D modeling and information extraction. Therefore, effectively removing these noise points is crucial.

[0080] For laser point cloud data, this invention mainly uses voxel filtering processing, and the specific steps are as follows:

[0081] (1) Construct a spatial voxel mesh. Assume the laser point cloud data is M = {m1, m2, ..., m}. n}, where m1, m2, ..., m n For each individual point in the point cloud data, calculate the minimum cubic boundary of the point cloud data M using the following formula:

[0082]

[0083] In the formula: x max y max z max Let x, y, and z represent the maximum values ​​of the point cloud data M in the x, y, and z directions, respectively; x min y min z min Let M represent the minimum values ​​of point cloud data M in the x, y, and z directions, respectively; N x N y N z represents the number of grid cells in the x, y, and z directions, respectively; l represents the grid size.

[0084] (2) Calculate the voxel index of the point cloud data. After constructing the spatial voxel grid of the point cloud, the voxel index of each point m in the point cloud data M is calculated. i =(x i ,y i ,z i The formula for calculating the voxel index is as follows:

[0085]

[0086] In the formula: i is a one-dimensional code, through which the voxel grid containing each point in the point cloud data can be directly retrieved; i x i y i z These represent the voxel index values ​​in the x, y, and z directions, respectively; N x N y N z represents the number of grid cells in the x, y, and z directions, respectively; l represents the grid size.

[0087] (3) Calculate the centroid of each voxel grid in the point cloud data. After determining the correspondence between each point in the point cloud data and the voxel grid through the voxel index value, calculate the centroid of each voxel grid. Then, use the point in the voxel grid that is closest to the centroid as the centroid of the voxel grid. The formula for calculating the centroid of the voxel grid is as follows:

[0088]

[0089] In the formula: k is the number of points contained in the voxel grid; G i Denotes the centroid of the i-th voxel mesh; m i This represents a point located in the i-th voxel grid.

[0090] (4) The centroid of each voxel grid is used to replace the points contained in the voxel grid, thereby completing the filtering process of the point cloud data.

[0091] Step 3: Use a point cloud registration algorithm to register the aerial triangulation dense point cloud model and the processed laser point cloud data. The purpose of point cloud registration is to register point clouds from different data sources to the same coordinate system, thereby forming a more complete and accurate point cloud data.

[0092] First, a coarse registration process is performed for point cloud registration. This invention utilizes a random sampling consensus algorithm for coarse registration of point cloud data. This algorithm calculates the transformation model by randomly sampling the point cloud data, then checks each transformation model, and repeats the above process until the optimal transformation model is selected. The specific process is as follows:

[0093] (1) Assume that the aerial triangulation dense point cloud data is the source point cloud P, and the laser point cloud data is the target point cloud Q. First, randomly select 4 corresponding points in the source point cloud P and the target point cloud Q to form four pairs of points. Then, calculate the rigid body transformation matrix H using the least squares method. Next, calculate the new point cloud set P′ of the source point cloud P under the rigid body transformation matrix H using the following formula:

[0094] P′=R c P+t c

[0095] In the formula: R c t is the rotation matrix; cP' represents the translation amount; P' denotes the new point cloud set of the source point cloud P under the rigid body transformation matrix H.

[0096] (2) Calculate the distance threshold between corresponding points of P′ and the target point cloud Q. The calculation formula is as follows:

[0097] D = d 2 (p′ i ,Hp i i = 1, 2, 3, 4

[0098] In the formula: D is the distance threshold; d represents the distance between corresponding points of P′ and the target point cloud Q; H represents the transformation matrix of the corresponding points; p′ i This represents the new points in H corresponding to the four selected points in the source point cloud P; p i This represents four corresponding points randomly selected from the source point cloud.

[0099] After determining the distance threshold D between corresponding points of P′ and the target point cloud Q, it is then determined whether the distance between the four pairs of points between P′ and the target point cloud Q is greater than the distance threshold D. If it is greater, the pair of points is considered not to belong to the interior point set I; if it is less, the pair of points is considered to belong to the interior point set I. Then, the registration score S is calculated using the following formula:

[0100]

[0101] In the formula: S is the registration score between the source point cloud P and the target point cloud Q; card(I) represents the cardinality of the interior point set I; card(P) represents the cardinality of the source point cloud P; card(Q) represents the cardinality of the target point cloud Q.

[0102] (3) Set the registration score threshold S min Repeat steps (1) and (2) until the calculated registration score S reaches the set registration score threshold S. min The loop ends, and the source point cloud P after coarse registration is obtained. c .

[0103] After coarse registration, the dense aerial triangulation point cloud data and the laser point cloud data have been initially aligned. The next step is fine registration to further improve the registration accuracy of the two point cloud datasets. This invention utilizes an iterative nearest-point algorithm that considers normals for fine registration. The iterative nearest-point algorithm primarily uses the minimum distance between corresponding points as the objective function. This invention, by incorporating normal information into the iterative nearest-point algorithm, can better improve the registration effect. The specific steps are as follows:

[0104] (a) First, find the source point cloud P after coarse registration. c The point closest to each point in the target point cloud Q is taken as its corresponding point.

[0105] (b) Calculate the source point cloud P c The optimal translation and rotation parameters for registration between corresponding points in the target point cloud Q are defined by the following objective optimization function:

[0106]

[0107] In the formula: q i Represents a point in the target point cloud Q; This represents the source point cloud P after coarse registration. c The point in; η i For q i arrive The projection onto the surface normal; t f R is the translation vector; f F(t) is the rotation matrix; f ,R f ) represents the optimal rigid body transformation between the corresponding point and the surface after the minimum transformation; F i (t f ,R f ) represents the rigid transformation in the i-th iteration.

[0108] Next, calculate the source point cloud P. c The centroid of the target point cloud Q is calculated using the following formula:

[0109]

[0110] In the formula: p cz Source point cloud P c The center of mass; q z Let n be the centroid of the target point cloud Q; and let n and m be the source point cloud P. c The number of points and the number of points in the target point cloud Q; Source point cloud P c The point in; q c Let Q be the point in the target point cloud.

[0111] Find P c After finding the center of mass of Q, then for P c Decentralizing Q is achieved using the following formula:

[0112]

[0113] In the formula: Represents the source point cloud P c Decentralized point cloud data; p represents the decentralized point cloud data of the target point cloud Q; cz Source point cloud P c The center of mass; q z The centroid of the target point cloud Q; q iRepresents a point in the target point cloud Q; Represents the source point cloud P c The point in the middle.

[0114] Assuming the transformation matrix is ​​G, its calculation formula is as follows:

[0115]

[0116] In the formula: G is the transformation matrix; Represents the source point cloud P c Decentralized point cloud data; This represents the decentralized point cloud data of the target point cloud Q.

[0117] Finally, the translation vector t is calculated. f and rotation matrix R f The calculation formula is as follows:

[0118]

[0119] In the formula: R f Represents the rotation matrix; t f p represents the translation vector; cz Source point cloud P c The center of mass; q z Let Q be the centroid of the target point cloud; A and B represent the orthogonal matrices of the transformation matrix G after SVD decomposition.

[0120] (c) Set the maximum number of iterations N max And the error threshold E, then repeat steps (a) and (b) iteratively, and calculate the error between every two iterations, as shown in the following formula:

[0121] ε=F k (t f ,R f )-F k-1 (t f ,R f )

[0122] In the formula: ε represents the iteration error; F k (t f ,R f ), F k-1 (t f ,R f ) represent the optimal rigid body transformations between the corresponding points and surfaces after the k-th and (k-1)-th minimum transformations, respectively.

[0123] When the iteration error ε is less than or equal to E or the number of iterations reaches N max The iteration is completed, and a high-precision point cloud model of the steep slope area is obtained after fine registration of the aerial triangulation dense point cloud and the laser point cloud.

[0124] Step 4: Based on the high-precision point cloud model of the steep slope area, 3D mesh reconstruction is performed using ContextCpature. First, in the software's spatial frame options, the coordinate system of the dangerous rock monitoring area model is input, and the model's cutting mode is set to regular planar mesh. Using the clipping box function, the output model's range is adjusted according to the dangerous rock monitoring area. Then, the output model type is selected as a 3D mesh model, the model output format is .OBJ format, texture compression is set to 100%, and adaptive tree is used for the level of detail. Finally, a high-precision 3D mesh model of the steep slope area is generated.

[0125] Step 5: Using the mesh model measurement tool ContextCapture Viewer, quickly compare the high-precision 3D mesh models of steep slope areas measured at different times. ContextCapture Viewer has built-in functions such as 3D distance, height difference, surface and volume calculations, enabling efficient and rapid measurement of information on unstable rocks in steep slope areas. Assuming two high-precision 3D mesh models M1 and M2 of the same steep slope area measured at different times, generally, the 3D distance measurement function in ContextCapture Viewer can be directly used to measure the movement distance of the same rock in M1 and M2 as dis. M1→M2 When dis M1→M2 Distance greater than the safe threshold for the movement of dangerous rocks max If so, the rock is considered to be a dangerous rock mass, posing a risk of collapse and landslide.

[0126] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A high and steep slope dangerous rock monitoring method based on point cloud-tilt photography technology, characterized in that: It includes the following steps: Step 1: Use UAV oblique photography technology to collect oblique photography data of dangerous rocks in the steep slope area, and use ContextCapture to build an aerial triangulation dense point cloud model of the steep slope area. Step 2: Use a drone equipped with a laser point cloud acquisition device to collect laser point cloud data in the steep slope area. At the same time, perform voxel filtering on the laser point cloud data to remove invalid points and reduce the amount of point cloud data. Step 3: Use a point cloud registration algorithm to register the aerial triangulation dense point cloud model and the laser point cloud data to obtain a high-precision point cloud model of the steep slope area after the fusion of oblique photography and point cloud data; Step 4: Based on the high-precision point cloud model of the steep slope area, use ContextCpature to reconstruct the three-dimensional mesh to obtain a high-precision three-dimensional mesh model of the steep slope area. Step 5: Use grid model measurement tools or point cloud comparison algorithms to compare measurement models of different periods in the high and steep slope area, thereby achieving effective monitoring of dangerous rocks on high and steep slopes; The specific process of step three is as follows: Step 3.1, assuming the empty three dense point cloud data as the source point cloud , and the laser point cloud data as the target point cloud ; first, randomly select four corresponding points in the source point cloud and the target point cloud respectively to form four groups of point pairs, and then obtain the rigid body transformation matrix by the least square method, and then calculate the new point cloud set of the source point cloud under the rigid body transformation matrix using the following formula : ； wherein: is a rotation matrix; is a translation; denotes the source point cloud the new point cloud set under the rigid body transformation matrix ; Step 3.2, calculating and the distance decision threshold between the corresponding points of the target point cloud The formula is as follows: ； In the formula: The distance determination threshold; express and target point cloud The distance between corresponding points; This represents the transformation matrix at the corresponding point; Represents source point cloud The four corresponding points selected in The new point below; This represents four corresponding points randomly selected from the source point cloud; Find and target point cloud Distance threshold between corresponding points Then, a judgment is made. and target point cloud Whether the distance between the four pairs of points is greater than the distance judgment threshold If the value is greater than 1, then the pair of points is considered not to belong to the set of interior points. If the difference is less than 1, then the pair of points is considered to belong to the set of interior points. Then calculate the registration score. The calculation formula is as follows: ； In the formula: Source Point Cloud and target point cloud Registration scores between them; Represents the set of interior points The cardinality; Represents source point cloud The cardinality; Represents the target point cloud The cardinality; Step 3.3: Set the registration score threshold Repeat steps 3.1 and 3.2 until the registration score is calculated. Reaching the set registration score threshold The loop then ends, yielding the source point cloud after coarse registration. ; Step 3.4: After completing the coarse registration, the dense aerial triangulation cloud data and the laser point cloud data have been initially aligned. Next, fine registration will be performed to further improve the registration accuracy of the two point cloud data. The fine registration will be performed using the iterative nearest point algorithm that takes into account the normal. The iterative nearest point algorithm uses the minimum distance between corresponding points as the objective function and takes into account the normal information on the basis of the iterative nearest point algorithm. Step 3.4 specifically includes the following: (a) First, find the source point cloud after coarse registration. Each point in the target point cloud The point closest to it in the interval is taken as its corresponding point; (b) Calculate the source point cloud and target point cloud The optimal translation and rotation parameters for registration between corresponding points are determined by the following objective function: ； In the formula: Represents the target point cloud The point in the middle; This represents the source point cloud after coarse registration. The point in the middle; for arrive The projection onto the surface normal; It is a translation vector; It is a rotation matrix; This represents the optimal rigid body transformation between the corresponding point and the surface after the minimum transformation. Indicates the first Rigid transformation in the next iteration; Next, calculate the source point cloud. and target point cloud The centroid is calculated using the following formula: ； In the formula: Source Point Cloud The center of mass; For target point cloud The center of mass; , Source cloud Point count and target point cloud The number of points; Source Point Cloud The point in the middle; For target point cloud The point in the middle; Find and After the center of mass, for and Decentralization is achieved, and the calculation formula is as follows: ； In the formula: Represents source point cloud Decentralized point cloud data; Represents the target point cloud Decentralized point cloud data; Source Point Cloud The center of mass; For target point cloud The center of mass; Represents the target point cloud The point in the middle; Represents source point cloud The point in the middle; Assume the transformation matrix is The calculation formula is as follows: ； In the formula: The transformation matrix; Represents source point cloud Decentralized point cloud data; Represents the target point cloud Decentralized point cloud data; Finally, the translation vector was calculated. and rotation matrix The calculation formula is as follows: ； In the formula: Represents the rotation matrix; Represents the translation vector; Source Point Cloud The center of mass; For target point cloud The center of mass; and Represents the transformation matrix Orthogonal matrix after SVD decomposition; (c) Set the maximum number of iterations and error threshold Then repeat steps (a) and (b) iteratively, and calculate the error between every two iterations using the following formula: ； In the formula: Indicates iteration error; , These represent the optimal rigid body transformations between the corresponding points and surfaces after the k-th and (k-1)-th minimum transformations, respectively. When iteration error Less than or equal to Or the number of iterations reaches The iteration is completed, and a high-precision point cloud model of the steep slope area is obtained after fine registration of the aerial triangulation dense point cloud and the laser point cloud.

2. The method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology according to claim 1, characterized in that: The specific process of step one is as follows: UAV oblique photography technology was used to collect oblique photography data of dangerous rocks in the steep slope area. The collected oblique photography data was then imported into ContextCapture. Next, the integrity of the images was checked and substandard oblique photography images were removed. Finally, aerial triangulation was performed to obtain an aerial triangulation dense point cloud model of the steep slope area.

3. The method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology according to claim 1, characterized in that: In step two, the specific steps for voxel filtering of the laser point cloud data are as follows: Step 2.1: Construct a spatial voxel mesh; Step 2.2: Calculate the voxel index of the point cloud data; Step 2.3: Calculate the centroid of each voxel mesh in the point cloud data; Step 2.4: Replace the points contained in each voxel grid with the centroid of each voxel grid to complete the filtering process of the point cloud data.

4. The method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology according to claim 3, characterized in that: The specific process of step 2.1 is as follows: Assuming the laser point cloud data is ,in Calculate the point cloud data for each individual point in the point cloud data. The minimum cubic boundary value is calculated using the following formula: ； In the formula: , , Representing point cloud data respectively exist x , y and z The maximum value in the direction; , , Representing point cloud data respectively exist x , y and z Minimum value in direction; , , They are respectively x , y and z Number of grid cells in each direction; This is the grid size.

5. The method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology according to claim 3, characterized in that: The specific process of step 2.2 is as follows: After constructing the spatial voxel mesh of the point cloud, the point cloud data various points in the middle The formula for calculating the voxel index is as follows: ； In the formula: For one-dimensional encoding, through You can directly retrieve the voxel grid containing each point in the point cloud data; , , They represent x , y and z The voxel index value of the direction; , , They are respectively x , y and z Number of grid cells in each direction; This is the grid size.

6. The method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology according to claim 3, characterized in that: The specific process of step 2.3 is as follows: After determining the correspondence between each point in the point cloud data and the voxel grid using voxel index values, the centroid of each voxel grid is calculated. Then, the point in the voxel grid closest to the centroid is used as the centroid of the voxel grid. The formula for calculating the centroid of the voxel grid is as follows: ； In the formula: It is the number of points contained in the voxel grid; Indicates the first The centroid of the individual element mesh; Indicates that it is located at the th Points in an individual element grid.

7. The method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology according to claim 1, characterized in that: The specific process of step four is as follows: Based on a high-precision point cloud model of a steep slope area, 3D mesh reconstruction is performed using ContextCpature. First, in the software's spatial frame options, the coordinate system of the dangerous rock monitoring area model is input, and the model's cutting mode is set to regular planar mesh. Using the clipping box function, the range of the output model is adjusted according to the dangerous rock monitoring area. Then, the output model type is selected as a 3D mesh model, the model output format is .OBJ format, texture compression is set to 100%, and adaptive tree is used for the level of detail. Finally, a high-precision 3D mesh model of the steep slope area is generated, resulting in a high-precision 3D mesh model of the steep slope area.

8. The method for monitoring dangerous rocks on steep slopes based on point cloud-oblique photography technology according to claim 1, characterized in that: The specific process of step five is as follows: Using the mesh model measurement tool ContextCapture Viewer, high-precision 3D mesh models of steep slope areas measured at different times in multiple periods are quickly compared. ContextCapture Viewer has built-in functions for calculating 3D distance, height difference, surface area, and volume, enabling efficient and rapid measurement of information on unstable rocks within steep slope areas. This is assuming two high-precision 3D mesh models of the same steep slope area measured at different times are used. and The 3D ranging function in ContextCaptureViewer was used directly to measure the distance. and The distance the same rock moves is ,when Distance greater than the safe threshold for the movement of dangerous rocks If so, the rock is considered to be a dangerous rock mass, posing a risk of collapse and landslide.

Images