A non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy

Abstract

This invention discloses a non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy. First, calculation center points are selected at certain intervals in the original three-dimensional point cloud to form preliminary point cloud blocks. Next, the average domain vector algorithm is used to register the point cloud blocks with point cloud data from another period, interpreting the current deformation data. Subsequently, the spacing between the calculation center points is reduced to obtain finer-grained calculation points. Distance weights are used to take the initial deformation of neighboring calculation points as a reference for the current calculation point, and the registration step is repeated, with the calculation results superimposed. Calculation of the current region is stopped when the spacing is reduced to a specified threshold or the deformation is less than a set threshold. During the calculation process, key parameters such as block radius, registration search distance, and outlier removal are automatically determined through a trained neural network model combined with Bayesian optimization. This method automates the calculation process, achieving the acquisition of the overall displacement and deformation field with millimeter-level accuracy.

Description

Technical Field

[0001] This invention belongs to the field of three-dimensional point cloud technology, specifically relating to a non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level precision. Background Technology

[0002] Large-scale 3D point cloud datasets of the Earth's surface can be obtained through 3D laser scanners, airborne LiDAR, and UAV aerial photography. Based on two sets of 3D point cloud datasets of the Earth's surface, deformation analysis of the target area can be performed. Traditional 3D point cloud analysis methods include the shortest distance method, the vertical distance method, and the M3C2 method.

[0003] The vertical distance method obtains the displacement change in the vertical direction of a point as the deformation amount at that point. The shortest distance method searches for the nearest point to the target point in another point cloud and uses the distance between them as the deformation amount at the target point. The advantages of the vertical distance and shortest distance methods are their simplicity and short computation time, allowing for a quick assessment of the approximate deformation of a landslide. However, the vertical distance method cannot reflect the deformation along the horizontal direction, while the shortest distance method has a higher degree of randomness, and the nearest point found by the algorithm may be far from the actual corresponding point before and after the landslide change. The M3C2 method can obtain the deformation amount of the calculation point along the normal direction by fitting a plane, and its applicability and accuracy are better than the shortest distance method. However, this method cannot reflect the displacement of the calculation point along the slope direction, and in some landslides with large sliding amounts along the slope, the calculation results of this method will have a large error.

[0004] The average domain vector algorithm is a point cloud deformation calculation method based on the ICP algorithm, a point cloud registration algorithm. Surface features before and after deformation are often similar; ICP registration allows the surface before deformation to be registered to the surface after deformation. Using the calculation data during point cloud registration, the overall transformation matrix can be obtained, thus revealing the deformation information of the point cloud. The average domain vector algorithm has high accuracy in processing locally uniformly deformed point clouds, but it has several limitations in large-area deformation analysis. For example, the deformation analysis results are difficult to refine. If the local analysis area is too large, the boundaries of the deformed area become unclear, and the calculated deformation value differs significantly from the actual deformation value. If the local analysis area is too small, the algorithm is prone to getting trapped in local optima, resulting in incorrect deformation values. Furthermore, this algorithm relies on manual intervention, which is time-consuming and labor-intensive. The analysis area needs to be manually defined, making it difficult to perform deformation calculations over large areas. Additionally, the calculation parameters need to be manually adjusted; unreasonable parameters can lead to large errors, often requiring multiple trials to obtain suitable parameters.

[0005] The ICP registration algorithm heavily relies on the features contained in the point cloud during the registration process. When the point cloud contains enough features, the registration result is often better; conversely, it is prone to getting trapped in local optima during registration. However, in actual 3D point cloud data, although large areas of point cloud contain more features, the internal deformation is uneven, making it impossible to achieve fine deformation calculation. Summary of the Invention

[0006] To address the technical problems existing in the background art, this invention aims to provide a non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy. To utilize sufficient features from a large-area point cloud, a sufficiently large point cloud block is defined during initial registration. Preliminary deformation information is obtained through registration. Then, the point cloud block is gradually reduced in size, and the deformation value obtained in the previous block registration is assigned to it according to its weight. This allows the point cloud block to gradually move closer to its correct matching position, preventing the algorithm from getting trapped in local optima due to insufficient features by directly selecting the smallest point cloud block. After the calculation results reach the target accuracy, the calculation is completed. The final transformation matrix of the point cloud block is obtained from the transformation matrix obtained in each registration step, thereby further interpreting the detailed overall deformation field of the analysis area. The entire step-by-step block registration process is implemented programmatically, reducing workload. Furthermore, since the step-by-step registration calculation involves many calculation parameters, a neural network model is trained using a large amount of step-by-step registration experimental data, and the optimal parameters are predicted using a Bayesian optimization algorithm. This significantly reduces the cost of human trial and error and improves the accuracy of deformation interpretation results.

[0007] To solve the technical problem, the technical solution of the present invention is as follows:

[0008] A non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy, the method comprising:

[0009] S1: Acquire two phases of 3D deformation point cloud data of the land surface for deformation interpretation and extract overall features;

[0010] S2: Based on the extracted overall features of the point cloud and the trained neural network model, the Bayesian optimization algorithm is used to find the optimal initial calculation point spacing, minimum point cloud search radius, maximum corresponding point search distance, point cloud containment threshold, and root mean square error threshold.

[0011] S3: Based on the initial calculation point spacing, obtain the calculation points, divide the initial point cloud into blocks, use the average domain vector algorithm for preliminary registration, obtain the initial transformation matrix, and parse out the translation vector and rotation matrix;

[0012] S4: Reduce the spacing between calculation points, obtain new calculation points, and use the inverse distance weighting method to perform a weighted average of the rotation vector and translation vector to obtain the initial transformation parameters of the new calculation points;

[0013] S5: Rotate and translate the point cloud blocks according to the transformation parameters of the new calculation points to the new positions, register them again, obtain the new transformation matrix, and interpret the new translation vector and rotation matrix;

[0014] S6: If the displacement of the calculation point is less than the specified threshold, the calculation of its region is considered complete. Gradually reduce the spacing between calculation points until the spacing between calculation points reaches the minimum value, then the overall calculation is considered complete.

[0015] Furthermore, step S2 includes:

[0016] The data used for training includes point cloud features, computational parameters, RMSE of points corresponding to the registration results, and the proportion of successfully matched blocks.

[0017] The loss function takes into account the root mean square error of the registration results and the proportion of successfully matched blocks. Its calculation formula is given by the following formula:

[0018] loss = αRMSE + βP + γ

[0019] Where: MSE represents the average root mean square error of all block matching results; P represents the percentage of successfully matched blocks; α, β, and γ are hyperparameters;

[0020] The average root mean square error (RMSE) of all block matching results is calculated using the following formula:

[0021]

[0022] Where, n S RMSE represents the total number of successfully registered blocks. i This represents the root mean square error of the corresponding point pair after the i-th registration, and it is calculated as follows:

[0023]

[0024] Where, n i Let be the number of points contained in the i-th block. Let j be the coordinates of the j-th point in the i-th block. Let J be the coordinates of the matching point of the j-th point in the i-th block in another point cloud.

[0025] The percentage P of successfully matched blocks is calculated as follows:

[0026]

[0027] Where n S This represents the total number of successfully registered blocks, where n represents the total number of blocks.

[0028] Furthermore, step S3 includes:

[0029] The initial calculation points are obtained based on the initial calculation point spacing, and the initial point cloud blocks are obtained based on the point cloud search radius. The initial point cloud blocks are registered with another point cloud using the average domain vector algorithm based on the determined calculation parameters, and the transformation matrix M0 of the point cloud blocks in the first registration is obtained.

[0030]

[0031] Simultaneously, the transformation matrix is ​​interpreted to obtain the translation vector of the current calculation point. and rotation matrix

[0032] Furthermore, step S4 includes:

[0033] Reduce the spacing between calculation points to obtain new calculation points. For each new calculation point, find its five nearest previous calculation points and apply the rotation matrix. Convert to a rotation vector expressed using axis angles. The initial translation of the calculation point is obtained by weighting its translation vector and rotation vector (expressed in terms of axis angle) using the inverse distance weighting method. and the initial rotation vector expressed in terms of axis angle

[0034]

[0035] Among them, w i The weights corresponding to the i most recent computation points are calculated as follows:

[0036]

[0037] Where, d i The distance from the i-th nearest calculation point to this calculation point;

[0038] Initial rotation vector Convert to initial rotation matrix

[0039] Furthermore, step S5 includes:

[0040] Initial rotation vector Convert to initial rotation matrix The point cloud block A corresponding to the calculated point is rotated and translated to a new position A. * ,

[0041]

[0042] The point cloud block is registered again using the average domain vector algorithm to obtain a new transformation matrix M2. The new translation vector and rotation matrix are then interpreted and superimposed with the translation vector and rotation matrix obtained in the previous step to obtain the current total deformation.

[0043] Furthermore, step S6 includes:

[0044] Repeat steps S4 and S5. When the displacement of a certain calculation point is less than a specified threshold after a certain registration, assuming the current calculation point spacing is d, the calculation is considered complete within d / 2 of the current calculation point. When the calculation point spacing is reduced next time, the current region will not take a new calculation point, but will directly use the calculation points from the previous round. At the same time, when the calculation point spacing is reduced to the specified threshold, the calculation is considered complete after the current round ends.

[0045] Furthermore, after step S6, the method further includes:

[0046] Based on all the calculation points and their corresponding transformation matrices in the last round of calculations, the displacement vectors corresponding to the calculation points are deciphered. The initial displacement vector of the current calculation round Superimpose the values ​​to obtain the total displacement vector at the calculation point.

[0047]

[0048] This allows us to obtain the overall displacement and deformation field of the computational region.

[0049] A non-contact monitoring system for three-dimensional overall deformation of landslides with millimeter-level precision, the system being applied to the method described in any one of the above-mentioned methods, the system comprising:

[0050] Point cloud data acquisition and feature extraction module: acquires two phases of 3D deformation point cloud data of the land surface for deformation interpretation and extracts overall features;

[0051] Parameter optimization module: Based on the extracted overall features of the point cloud and the trained neural network model, the Bayesian optimization algorithm is used to find the optimal initial calculation point spacing, minimum point cloud search radius, maximum corresponding point search distance, point cloud containment threshold, and root mean square error threshold.

[0052] Initial calculation and registration module: Based on the initial calculation point spacing, the calculation points are obtained, and the initial point cloud is divided into blocks. The average domain vector algorithm is used for preliminary registration to obtain the initial transformation matrix, and the translation vector and rotation matrix are parsed out.

[0053] Refine the calculation points and weighted average module: reduce the spacing between calculation points, obtain new calculation points, and use the inverse distance weighting method to perform a weighted average of the rotation vector and translation vector to obtain the initial transformation parameters of the new calculation points;

[0054] Point cloud block transformation and re-registration module: Rotates and translates the point cloud blocks to new positions according to the transformation parameters of the new calculation points, re-registers them, obtains the new transformation matrix, and interprets the new translation vector and rotation matrix;

[0055] Convergence Judgment and Calculation Control Module: If the displacement of a calculation point is less than a specified threshold, the calculation of its region is considered complete. The spacing between calculation points is gradually reduced until the spacing between calculation points reaches the minimum value, at which point the overall calculation is considered complete.

[0056] Final Deformation Calculation Module: Based on all the calculation points and their corresponding transformation matrices in the last round of calculation, the displacement vector corresponding to the calculation point is interpreted and superimposed with the initial displacement vector of the current calculation round to obtain the total displacement vector of the calculation point; thus, the overall displacement deformation field of the calculation area can be obtained.

[0057] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level precision, as described in any one of the above-mentioned methods.

[0058] A computer-readable storage medium storing a computer program that, when executed by a processor, implements a non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level precision, as described above.

[0059] Compared with the prior art, the advantages of the present invention are as follows:

[0060] 1. High-precision 3D deformation interpretation

[0061] By gradually reducing the distance between the calculation center points, a coarse-to-fine deformation interpretation is achieved, thereby improving the accuracy of the final interpretation result.

[0062] The average domain vector algorithm is used for point cloud block registration to reduce the impact of local noise and ensure the stability and accuracy of point cloud matching.

[0063] By using iterative optimization and deformation superposition, the continuity of the point cloud is fully considered, resulting in smoother and more reasonable calculation results.

[0064] 2. Adaptive parameter optimization improves computational efficiency.

[0065] By combining neural network models with Bayesian optimization algorithms, key parameters (such as search radius, point threshold, root mean square error threshold, etc.) are automatically optimized, avoiding manual experience-based settings and improving the intelligence level of the calculation.

[0066] Through training with a large amount of experimental data, the calculation parameters can be dynamically adjusted according to different point cloud characteristics (density, size, deformation, terrain type, etc.), making the algorithm applicable to various terrains and data acquisition methods.

[0067] 3. The scientific method for initializing calculation points improves stability.

[0068] The inverse distance weighting (IDW) method is used to estimate the initial deformation of new calculation points, so that the calculation points have good smoothness and continuity during the iteration process and avoid the influence of local outliers.

[0069] By using weighted interpolation of deformation information from neighboring points, noise interference during registration is effectively reduced, and the stability of deformation interpretation is improved.

[0070] 4. The calculation termination mechanism is reasonable, reducing redundant calculations.

[0071] Set dual termination conditions:

[0072] To avoid wasting computing resources, the calculation point spacing is reduced to the minimum threshold.

[0073] When the change in deformation is less than a set threshold, the calculation is terminated early to improve calculation efficiency.

[0074] By adopting a region-adaptive termination strategy, regions that have already converged are no longer recalculated, further reducing the computational burden and improving the overall interpretation efficiency.

[0075] 5. The deformation field is fully constructed and suitable for a variety of applications.

[0076] By reconstructing the overall displacement and deformation field of the Earth's surface using the deformation data of the final calculation points, it can be used in fields such as geological disaster monitoring and engineering deformation monitoring.

[0077] It is applicable to various point cloud data sources (such as LiDAR, photogrammetry, UAV point clouds, etc.) and has strong generalization ability. Attached Figure Description

[0078] Figure 1 This is a schematic diagram of the step-by-step calculation process of the present invention;

[0079] Figure 2 This is a schematic diagram of the network training steps of the present invention;

[0080] Figure 3 This is a diagram showing the interpretation results of the deformation field of the quarry slope. Detailed Implementation

[0081] The specific implementation of the present invention is described below with reference to embodiments:

[0082] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0083] Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0084] Example 1:

[0085] like Figure 1-2 As shown, a non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy is described, and the specific technical solution is as follows;

[0086] (1) Prepare two phases of 3D deformation point cloud data for deformation interpretation. The areas covered by the two phases of point cloud data should correspond.

[0087] (2) Obtain the overall characteristics of the point cloud, including the average density, smoothness, size, surface type, data acquisition method, and average deformation estimate.

[0088] (3) Based on the overall features of the point cloud obtained in (2) and the trained neural network model, the Bayesian optimization algorithm is used to find the optimal initial calculation point spacing, minimum point cloud search radius, maximum corresponding point search distance, point cloud point count threshold, root mean square error threshold, and other key parameters required for subsequent calculations. The data used for training includes point cloud features (average density of the point cloud, point cloud size, magnitude of deformation, terrain type, point cloud acquisition method, and point cloud smoothness), calculation parameters (minimum point cloud search radius, maximum corresponding point search distance, point cloud point count threshold, root mean square error threshold), and the RMSE of the corresponding points in the registration result and the proportion of successfully matched blocks. The loss function mainly considers the root mean square error of the registration result and the proportion of successfully matched blocks, and its calculation formula is given by the following formula:

[0089] loss = αRMSE + βP + γ

[0090] Where: MSE represents the average root mean square error of all block matching results; P represents the percentage of successfully matched blocks; α, β, and γ are hyperparameters;

[0091] The average root mean square error (RMSE) of all block matching results is calculated using the following formula:

[0092]

[0093] Where, n S RMSE represents the total number of successfully registered blocks. i This represents the root mean square error of the corresponding point pair after the i-th registration, and it is calculated as follows:

[0094]

[0095] Where, n i Let be the number of points contained in the i-th block. Let j be the coordinates of the j-th point in the i-th block. Let be the coordinates of the matching point of the j-th point in the i-th block in another point cloud.

[0096] The percentage P of successfully matched blocks is calculated as follows:

[0097]

[0098] Where n S This represents the total number of successfully registered blocks, where n represents the total number of blocks.

[0099] (4) Obtain the initial calculation point based on the initial calculation point spacing, obtain the initial point cloud block based on the point cloud search radius, and use the calculation parameters determined in (3) to register the initial point cloud block with another period point cloud using the average domain vector algorithm to obtain the transformation matrix M0 of the point cloud block in the first registration.

[0100]

[0101] Simultaneously, the transformation matrix is ​​interpreted to obtain the translation vector of the current calculation point. and rotation matrix

[0102] (5) Reduce the spacing between calculation points to obtain new calculation points. For each new calculation point, find the five nearest previous calculation points around it and rotate the matrix. Convert to a rotation vector expressed using axis angles. The initial translation of the calculation point is obtained by weighting its translation vector and rotation vector (expressed in terms of axis angle) using the inverse distance weighting method. and the initial rotation vector expressed in terms of axis angle

[0103]

[0104] Among them, w i The weights corresponding to the i most recent computation points are calculated as follows:

[0105]

[0106] Where, d i Let be the distance from the i-th nearest calculation point to this calculation point.

[0107] Initial rotation vector Convert to initial rotation matrix

[0108] (6) Initial rotation vector Convert to initial rotation matrix The point cloud block A corresponding to the calculated point is rotated and translated to a new position A. * ,

[0109]

[0110] The point cloud block is registered again using the average domain vector algorithm to obtain a new transformation matrix M2. The new translation vector and rotation matrix are then interpreted and superimposed with the translation vector and rotation matrix obtained in the previous step to obtain the current total deformation.

[0111] (7) Repeat steps (5) and (6). When the displacement of a certain calculation point is less than the specified threshold after a certain registration, assuming the current calculation point spacing is d, the calculation is considered complete within d / 2 of the current calculation point. When the calculation point spacing is reduced next time, the current region will not take a new calculation point, but will directly use the calculation points from the previous round. At the same time, when the calculation point spacing is reduced to the specified threshold, the calculation is considered complete after the current round ends.

[0112] (8) Based on all the calculation points and their corresponding transformation matrices in the last round of calculation, the displacement vectors corresponding to the calculation points are deciphered. The initial displacement vector of the current calculation round Superimpose the values ​​to obtain the total displacement vector at the calculation point.

[0113]

[0114] This allows us to obtain the overall displacement and deformation field of the computational region.

[0115] Example 2:

[0116] This embodiment presents a non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level precision, using a landslide analysis case in Liangshan Prefecture, Sichuan Province as an example:

[0117] (1) Two phases of three-dimensional laser scanning point cloud data of a landslide in Liangshan Prefecture, Sichuan Province, were obtained by processing the two phases of three-dimensional point cloud data of the deformation analysis area using RiSCAN Pro software.

[0118] (2) Statistically analyze the average density, point cloud size, preliminary estimated average deformation, point cloud acquisition method (3D laser scanning), point cloud smoothness (obtained through PointNet), and terrain type (mountainous terrain with vegetation). Convert the above features into corresponding data and input them into the trained neural network model. Obtain recommended calculation parameters through Bayesian optimization: minimum block radius (0.8m), maximum corresponding point search distance (2.2m), minimum number of points in the block (63), and RMSE threshold (0.19m).

[0119] (3) Pass the above parameters into the step-by-step registration program for step-by-step calculation. The specific process is as follows:

[0120] Initially, the calculation point spacing was set to 1 / 5 of the horizontal dimension of the point cloud. The block search deformation was set to the larger of 1 / 2 of the calculation point spacing and the minimum block radius (0.8m). Blocks with fewer than 63 points were considered invalid data. During registration, data with an RMSE greater than 0.19m were considered incorrect registrations and discarded as interference. The maximum search radius during ICP registration was set to 2.2m. After the final round of block division, deformation data from 238,404 calculation points were obtained, thus generating the overall deformation field within the region.

[0121] (4) To verify the quality of the deformation field results obtained in step (3) of this method, the actual displacement was obtained by manually calculating the coordinate difference of obvious feature points in the original point cloud data. At the same time, the deformation calculation results were obtained by using the shortest distance method and the M3C2 method and compared, as shown in Table 1:

[0122] Table 1 - Comparison of displacement calculation results at multiple points using different methods (unit: m)

[0123]

[0124] As can be seen, the displacement calculated by this method is generally closer to the actual displacement, with a smaller difference, proving that the result is more accurate and better.

[0125] (5) To further verify the accuracy of this method, a quarry slope in Aba Prefecture, Sichuan Province was selected for analysis. The overall deformation field of the analysis area was generated according to the above steps. The deformation results of a part of the stable area were selected for analysis in the deformation interpretation results. The results showed that the deformation interpretation results in the stable area were all within 1cm, proving that the method achieved mm-level accuracy.

[0126] Example 3:

[0127] This embodiment provides a terminal device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment can be used in the operation of a non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level precision, including the following steps:

[0128] First, computational center points are selected at certain intervals in the original 3D point cloud to form preliminary point cloud blocks. Next, the average domain vector algorithm is used to register the point cloud blocks with point cloud data from another period, interpreting the current deformation data. Then, the spacing between the computational center points is reduced to obtain finer-grained computational points. Distance weights are used to take the initial deformation of neighboring computational points as a reference for the current computational point, and the registration step is repeated, with the calculation results superimposed. The calculation of the current region is stopped when the spacing is reduced to a specified threshold or the deformation is less than a set threshold. During the calculation process, key parameters such as block radius, registration search distance, and outlier removal are automatically determined through a trained neural network model combined with Bayesian optimization.

[0129] Example 4:

[0130] This embodiment provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a terminal device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the terminal device and extended storage media supported by the terminal device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device.

[0131] One or more instructions stored in a computer-readable storage medium can be loaded and executed by a processor to implement the corresponding steps of the non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy in the above embodiments; one or more instructions in the computer-readable storage medium are loaded and executed by the processor in the following steps:

[0132] First, computational center points are selected at certain intervals in the original 3D point cloud to form preliminary point cloud blocks. Next, the average domain vector algorithm is used to register the point cloud blocks with point cloud data from another period, interpreting the current deformation data. Then, the spacing between the computational center points is reduced to obtain finer-grained computational points. Distance weights are used to take the initial deformation of neighboring computational points as a reference for the current computational point, and the registration step is repeated, with the calculation results superimposed. The calculation of the current region is stopped when the spacing is reduced to a specified threshold or the deformation is less than a set threshold. During the calculation process, key parameters such as block radius, registration search distance, and outlier removal are automatically determined through a trained neural network model combined with Bayesian optimization.

[0133] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0134] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0135] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0136] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0137] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

[0138] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. A non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy, characterized in that, The method includes: S1: Acquire two phases of 3D deformation point cloud data of the land surface for deformation interpretation, and extract the overall features of the point cloud, including: the average density, smoothness, size, land surface type, data acquisition method and average deformation estimate of the point cloud. S2: Based on the extracted overall features of the point cloud and the trained neural network model, the Bayesian optimization algorithm is used to find the optimal initial calculation point spacing, minimum point cloud search radius, maximum corresponding point search distance, point cloud containment threshold, and root mean square error threshold. S3: Based on the initial calculation point spacing, obtain the calculation points, divide the initial point cloud into blocks, use the average domain vector algorithm for preliminary registration, obtain the initial transformation matrix, and parse out the translation vector and rotation matrix; S4: Reduce the distance between calculation points, obtain new calculation points, and use the inverse distance weighting method to perform weighted averages on the rotation vector and translation vector respectively to obtain the initial rotation vector and translation amount of the new calculation point; S5: Rotate and translate the point cloud blocks according to the transformation parameters of the new calculation points to the new positions, register them again, obtain the new transformation matrix, and interpret the new translation vector and rotation matrix; S6: Repeat steps S4 and S5. When the displacement of a certain calculation point is less than a specified threshold after a certain registration, assuming the current calculation point spacing is d, the calculation is considered complete within d / 2 of the current calculation point. When the calculation point spacing is reduced next time, the current region will not take new calculation points, but will directly use the calculation points from the previous round. At the same time, when the calculation point spacing is reduced to the specified threshold, the calculation is considered complete after the current round ends. After step S6, the method further includes: Based on all the calculation points and their corresponding transformation matrices in the last round of calculations, the displacement vectors corresponding to the calculation points are deciphered. The initial displacement vector of the current calculation round Superimpose the values ​​to obtain the total displacement vector at the calculation point. This allows us to obtain the overall displacement and deformation field of the computational region.

2. The non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy according to claim 1, characterized in that, Step S2 includes: The data used for training includes point cloud features, computational parameters, RMSE of points corresponding to the registration results, and the proportion of successfully matched blocks. The loss function takes into account the root mean square error of the registration results and the proportion of successfully matched blocks. Its calculation formula is given by the following formula: loss = αRMSE + βP + γ Where: RMSE represents the average root mean square error of all block matching results; P represents the percentage of successfully matched blocks; α, β, and γ are hyperparameters; The average root mean square error (RMSE) of all block matching results is calculated using the following formula: Where, n S RMSE represents the total number of successfully registered blocks. i This represents the root mean square error of the corresponding point pair after the i-th registration, and it is calculated as follows: Where, n i Let be the number of points contained in the i-th block. Let j be the coordinates of the j-th point in the i-th block. Let J be the coordinates of the matching point of the j-th point in the i-th block in another point cloud. The percentage P of successfully matched blocks is calculated as follows: Where n S This represents the total number of successfully registered blocks, where n represents the total number of blocks.

3. The non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy according to claim 1, characterized in that, Step S3 includes: The initial calculation points are obtained based on the initial calculation point spacing, and the initial point cloud blocks are obtained based on the point cloud search radius. The initial point cloud blocks are registered with another point cloud using the average domain vector algorithm based on the determined calculation parameters, and the transformation matrix M0 of the point cloud blocks in the first registration is obtained. Simultaneously, the transformation matrix is ​​interpreted to obtain the translation vector of the current calculation point. and rotation matrix 4. The non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy according to claim 1, characterized in that, Step S4 includes: Reduce the spacing between calculation points to obtain new calculation points. For each new calculation point, find its five nearest previous calculation points and apply the rotation matrix. Convert to a rotation vector expressed using axis angles. The initial translation of the calculation point is obtained by weighting its translation vector and rotation vector (expressed in terms of axis angle) using the inverse distance weighting method. and the initial rotation vector expressed in terms of axis angle Among them, w i The weights corresponding to the i most recent computation points are calculated as follows: Where, d i The distance from the i-th nearest calculation point to this calculation point; Initial rotation vector Convert to initial rotation matrix 5. The non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level accuracy according to claim 1, characterized in that, Step S5 includes: Initial rotation vector Convert to initial rotation matrix The point cloud block A corresponding to the calculated point is rotated and translated to a new position A. * , The point cloud block is registered again using the average domain vector algorithm to obtain a new transformation matrix M2. The new translation vector and rotation matrix are then interpreted and superimposed with the translation vector and rotation matrix obtained in the previous step to obtain the current total deformation.

6. A non-contact monitoring system for three-dimensional overall deformation of landslides with millimeter-level precision, characterized in that, The system is applied to the method according to any one of claims 1-5, the system comprising: Point cloud data acquisition and feature extraction module: Acquire two phases of 3D deformation point cloud data of the land surface for deformation interpretation, and extract the overall features of the point cloud, including: the average density, smoothness, size, land surface type, data acquisition method and average deformation estimate of the point cloud; Parameter optimization module: Based on the extracted overall features of the point cloud and the trained neural network model, the Bayesian optimization algorithm is used to find the optimal initial calculation point spacing, minimum point cloud search radius, maximum corresponding point search distance, point cloud containment threshold, and root mean square error threshold. Initial calculation and registration module: Based on the initial calculation point spacing, the calculation points are obtained, and the initial point cloud is divided into blocks. The average domain vector algorithm is used for preliminary registration to obtain the initial transformation matrix, and the translation vector and rotation matrix are parsed out. Refine the calculation point and weighted average module: reduce the spacing between calculation points, obtain new calculation points, and use the inverse distance weighting method to perform weighted averages on the rotation vector and translation vector respectively to obtain the initial rotation vector and translation amount of the new calculation point; Point cloud block transformation and re-registration module: Rotates and translates the point cloud blocks to new positions according to the transformation parameters of the new calculation points, re-registers them, obtains the new transformation matrix, and interprets the new translation vector and rotation matrix; Convergence Judgment and Calculation Control Module: When the displacement of a calculation point is less than a specified threshold after a registration, assuming the current calculation point spacing is d, the calculation is considered complete within d / 2 of the current calculation point. When the calculation point spacing is reduced next time, the current region will not take new calculation points, but will directly use the calculation points from the previous round. At the same time, when the calculation point spacing is reduced to the specified threshold, the calculation is considered complete after the current round ends. Final Deformation Calculation Module: Based on all calculation points and their corresponding transformation matrices in the last round of calculation, the displacement vector corresponding to the calculation point is interpreted and superimposed with the initial displacement vector of the current calculation round to obtain the total displacement vector of the calculation point; thus, the overall displacement deformation field of the calculation region is obtained.

7. A computer device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level precision, as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements a non-contact monitoring method for three-dimensional overall deformation of landslides with millimeter-level precision, as described in any one of claims 1 to 5.

Images