Highway subgrade three-dimensional deformation monitoring system based on unmanned aerial vehicle oblique photography

Abstract

The present application relates to the technical field of highway engineering safety monitoring, in particular to a highway subgrade three-dimensional deformation monitoring system based on unmanned aerial vehicle oblique photography, comprising: the unmanned aerial vehicle oblique photography module adopts a five-lens synchronous shooting strategy to achieve full-range coverage of the subgrade; a three-dimensional point cloud generation engine generates a high-precision three-dimensional point cloud model through feature extraction, registration and triangulation strategies; a deformation field analysis platform models the subgrade surface as a parameterized Riemann manifold based on differential geometry theory, captures deformation dynamic characteristics through multi-scale curvature flow analysis, and identifies potential risks using the topological evolution characteristics of deformation singular points; a risk evaluator assesses the risk level based on the singular point density, change rate and topological stability and generates early warning information; the present application achieves millimeter-level accurate monitoring of subgrade deformation, can capture various deformation modes from microcracks to large-scale subsidence, provides early warning of potential landslide risks 30-45 days in advance, and significantly reduces the false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of highway engineering safety monitoring technology, and in particular to a highway subgrade three-dimensional deformation monitoring system based on UAV oblique photography, which is used to achieve high-precision monitoring of the three-dimensional spatial deformation of highway subgrade and early warning of landslide risk. Background Technology

[0002] The roadbed is an important component of the highway structure, and its stability directly affects driving safety and the service life of the highway. Due to factors such as geology, climate, and traffic loads, the roadbed is prone to various deformations, such as settlement, displacement, and landslides, which seriously threaten the safe operation of the highway.

[0003] Traditional methods for monitoring roadbed deformation mainly rely on techniques such as manual leveling, GPS surveying, and ground laser scanning. These methods have the following shortcomings: First, they are inefficient and require a lot of manual operation; second, they have a limited monitoring range, making it difficult to achieve comprehensive monitoring of large areas of the roadbed; third, their monitoring accuracy is greatly limited by environmental conditions; fourth, they can only obtain deformation data at discrete points, making it difficult to comprehensively reflect the overall deformation state of the roadbed; and fifth, they lack in-depth analysis of the deformation mechanism, making it difficult to achieve early warning of landslide risks.

[0004] In recent years, with the development of UAV technology and photogrammetry, 3D reconstruction methods based on UAV oblique photogrammetry have been gradually applied to the field of engineering monitoring. Compared with traditional monitoring methods, UAV oblique photogrammetry has advantages such as wide coverage, flexible operation, and rapid data acquisition. However, existing monitoring systems based on UAV oblique photogrammetry mainly focus on static 3D reconstruction, lack in-depth analysis of time-series deformation data, have insufficient understanding of deformation mechanisms, and limited early warning capabilities.

[0005] Therefore, there is an urgent need for a highway subgrade monitoring system that can achieve high-precision three-dimensional deformation monitoring, in-depth analysis of deformation mechanisms, and early warning capabilities. Summary of the Invention

[0006] The purpose of this invention is to provide a three-dimensional deformation monitoring system for highway subgrade based on UAV oblique photography, aiming to overcome the problems existing in the prior art and achieve high-precision monitoring of highway subgrade deformation, in-depth analysis of deformation mechanism, and early warning of landslide risk.

[0007] This invention proposes a three-dimensional deformation monitoring system for highway subgrade based on UAV oblique photography, comprising:

[0008] The UAV oblique photography module is used to collect multi-angle image data of the roadbed and transmit it to the 3D point cloud generation engine;

[0009] A 3D point cloud generation engine is communicatively connected to the UAV oblique photography module, used to receive the multi-angle image data and generate 3D point cloud data in a time series.

[0010] A deformation field analysis platform, communicatively connected to the 3D point cloud generation engine, is used to receive the 3D point cloud data over the time series and perform deformation analysis based on differential geometry theory, including:

[0011] A manifold modeling unit is used to convert the three-dimensional point cloud data into a parameterized Riemannian manifold model;

[0012] The curvature analysis unit is used to calculate the multi-scale curvature features on the parameterized Riemannian manifold model.

[0013] Topological feature units are used to identify deformation singularities and their topological features on the parameterized Riemannian manifold model.

[0014] The risk assessor is connected in communication with the deformation field analysis platform and is used to predict the risk of roadbed landslides and generate graded early warning information based on the time evolution of the deformation singularities and their topological characteristics.

[0015] Preferably, the UAV oblique photography module includes:

[0016] The four-lens photography unit includes four cameras, with the included angle between adjacent cameras set at 90 degrees, for capturing all-around images around the roadbed.

[0017] The flight control unit is used to control the UAV to fly along a preset route, which is either bow-shaped or zig-shaped.

[0018] The data synchronization unit is used to ensure that the four cameras capture images synchronously and that the overlap between adjacent shooting areas is not less than 90%.

[0019] Preferably, the 3D point cloud generation engine includes:

[0020] The feature extraction module is used to extract feature points from the multi-angle image data;

[0021] The spatial matching module is used to build the KD tree structure and perform feature point matching.

[0022] The mismatch elimination module is used to remove mismatches of feature points using the RANSAC algorithm;

[0023] The point cloud registration module is used to perform initial registration and global fine registration of point cloud data using the iterative nearest point algorithm and the ICP algorithm.

[0024] The 3D model generation module is used to generate dense point cloud models and 3D models through a progressive triangulation strategy.

[0025] Preferably, the manifold modeling unit includes:

[0026] The point cloud preprocessing subunit is used to perform noise reduction, registration, and mesh generation on the three-dimensional point cloud data.

[0027] The parameterized mapping subunit is used to establish the mapping relationship between the 3D point cloud and the parameter domain, and to generate parameterized surfaces;

[0028] The metric tensor computation subunit is used to compute the metric tensor on the parameterized surface, characterizing the intrinsic geometric properties of the manifold.

[0029] The manifold optimization sub-unit is used to optimize the parameterized mapping, reduce distortion, and improve the quality of parameterization.

[0030] Preferably, the curvature analysis unit includes:

[0031] The basic curvature calculation subunit is used to calculate the Gaussian curvature and mean curvature of each point on the parameterized Riemannian manifold model.

[0032] The principal curvature extraction sub-unit is used to calculate the principal curvature and corresponding principal direction at each point;

[0033] Multi-scale decomposition subunits are used to construct a multi-scale representation of curvature using a Gaussian smoothing kernel, decomposing curvature features into large-scale, medium-scale, and small-scale components;

[0034] The curvature flow dynamic analysis sub-unit is used to track the evolution trajectory of curvature extrema over time and construct the surface deformation rate field and acceleration field.

[0035] Preferably, the topological feature unit includes:

[0036] The singularity detection subunit is used to locate singularities on a parameterized Riemannian manifold model based on curvature and curvature gradient features.

[0037] A singularity classification subunit is used to classify the singularities into elliptical, hyperbolic, parabolic, or degenerate types;

[0038] The singularity tracking subunit is used to match and track the same singularities across consecutive time frames and record the lifecycle of the singularities.

[0039] The topology evolution map construction subunit is used to construct deformation topology evolution maps based on singular point trajectories and interaction events.

[0040] Preferably, the risk assessor includes:

[0041] A risk pattern library for storing typical risk topology patterns;

[0042] The topology matching module is used to match the current topology evolution state with the patterns in the risk pattern library;

[0043] The risk level assessment module is used to comprehensively consider singularity density, rate of change, and topological stability to assess the risk level.

[0044] The early warning information generation module is used to generate graded early warning information based on the risk level and indicate the location of high-risk areas.

[0045] Preferably, the deformation field analysis platform further includes an adaptive monitoring and control unit, used for:

[0046] The time interval and spatial resolution of subsequent monitoring are automatically adjusted based on the deformation analysis results.

[0047] Increase monitoring density in high-risk areas to improve early warning sensitivity;

[0048] The optimized flight path and shooting parameters are sent to the UAV tilt photography module.

[0049] Preferably, the deformation field analysis platform further includes a multi-source data fusion unit, used for:

[0050] Receive meteorological data and analyze the impact of rainfall on deformation;

[0051] Receive geological survey data and analyze stratigraphic structural factors;

[0052] Receive traffic load data and assess the impact of dynamic loads;

[0053] Improve the accuracy of deformation analysis and risk warning by leveraging fused multi-source data.

[0054] Preferably, the parameterized Riemannian manifold model in the deformation field analysis platform has the following characteristics:

[0055] It can maintain geometric continuity and achieve a precise mathematical description of the roadbed deformation process;

[0056] It features an adaptive control point density distribution in high curvature regions, with a control point grid density of 10-30 points / square meter.

[0057] Area distortion rate is less than 5%, and angular distortion rate is less than 10°;

[0058] By using the principal curvature direction as the basis of the local coordinate system, deformation analysis is ensured to be performed along the natural deformation direction of the roadbed.

[0059] This invention introduces differential geometry theory to model the roadbed surface as a parameterized Riemannian manifold, realizing the transformation from discrete point cloud data to continuous geometric representation; it captures the dynamic characteristics of roadbed deformation through multi-scale curvature flow analysis; and it achieves early warning of potential landslide risks by identifying the topological features of deformation singularities.

[0060] The beneficial effects of this invention are mainly reflected in:

[0061] 1. Improved monitoring accuracy: Through five-lens oblique photography technology and differential geometric modeling method, millimeter-level accurate monitoring of roadbed deformation was achieved. Compared with traditional methods, the accuracy of detecting small deformations was improved by 50% to 60%.

[0062] 2. Comprehensive monitoring is achieved: The system adopts a bow-shaped or zigzag flight path design and a five-lens synchronous shooting strategy to ensure an overlap of more than 90%, enabling all-round and multi-angle observation of the highway subgrade.

[0063] 3. Enhanced deformation analysis: By introducing differential geometry theory, the system can understand the deformation process from a geometrical perspective, not only monitoring the deformation that has occurred, but also analyzing the rate and acceleration characteristics of the deformation.

[0064] 4. Improved early warning capability: Through topological feature identification and evolution analysis of deformation singularities, the system can identify characteristic topological patterns of landslide precursors, and the landslide warning time can be 30-45 days in advance, reducing the false alarm rate by more than 75%.

[0065] 5. Enhanced system adaptability: The multi-scale analysis framework enables the system to simultaneously capture various deformation modes, from microcracks to large-scale settlement, adapting to different types of subgrade engineering. Attached Figure Description

[0066] Figure 1 This is an overall architecture diagram of the highway subgrade three-dimensional deformation monitoring system based on UAV oblique photography of the present invention;

[0067] Figure 2 This is a schematic diagram of the structure of the UAV oblique photography module of the present invention;

[0068] Figure 3 This is a flowchart illustrating the workflow of the 3D point cloud generation engine of this invention.

[0069] Figure 4 This is a flowchart of the manifold modeling unit in the deformation field analysis platform of the present invention;

[0070] Figure 5 This is a schematic diagram of the multi-scale analysis of the curvature analysis unit in the deformation field analysis platform of the present invention;

[0071] Figure 6This is a schematic diagram illustrating the classification of singular points in the topological feature unit of the deformation field analysis platform of this invention.

[0072] Figure 7 This is a flowchart of the risk level assessment process for the risk assessor of this invention. Detailed Implementation

[0073] Please refer to Figures 1-7 The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention.

[0074] like Figure 1 As shown, the highway subgrade three-dimensional deformation monitoring system based on UAV oblique photography provided by the present invention includes a UAV oblique photography module 1, a three-dimensional point cloud generation engine 2, a deformation field analysis platform 3, and a risk assessor 4.

[0075] The UAV oblique photography module 1 is used to acquire multi-angle image data of the highway subgrade and transmit it to the 3D point cloud generation engine 2. The 3D point cloud generation engine 2 receives the multi-angle image data and generates 3D point cloud data over time. The deformation field analysis platform 3 receives the 3D point cloud data over time and performs deformation analysis based on differential geometry theory. The risk assessor 4 predicts the subgrade landslide risk based on the deformation analysis results and generates graded early warning information.

[0076] The modules communicate and interact with each other via a communication interface. Preferably, the communication interface uses wireless transmission, which facilitates system deployment in field environments.

[0077] like Figure 2 As shown, the UAV oblique photography module 1 includes a four-lens photography unit 11, a flight control unit 12, and a data synchronization unit 13.

[0078] The four-lens photography unit 11 includes four cameras, with the included angle between adjacent cameras set at 90 degrees, for acquiring omnidirectional images around the roadbed. Preferably, in one embodiment of the invention, one of the four cameras is mounted vertically downwards on the bottom of the drone, while the other four cameras face forward, backward, left, and right, respectively, at 90-degree angles to the vertical camera. This arrangement ensures omnidirectional coverage of ground targets and reduces obstruction and shadow areas.

[0079] The flight control unit 12 controls the drone to fly along a preset route, which is either bow-shaped or zig-shaped. Furthermore, the flight control unit 12 also controls the drone's altitude, speed, and attitude to ensure image quality and coverage. In a preferred embodiment of the invention, the drone's flight altitude is set to 150-220 meters, and its flight speed is set to 5-8 meters per second. These parameter settings can improve operational efficiency while ensuring image clarity.

[0080] The data synchronization unit 13 ensures that the four cameras capture images synchronously and that the overlap between adjacent shooting areas is no less than 90%. Preferably, the data synchronization unit 13 employs GPS / INS-assisted synchronization triggering technology, controlling the camera synchronization error to within 1 millisecond. Simultaneously, by adjusting the flight path spacing and shooting frequency, it ensures that the image overlap reaches over 90%, providing sufficient image matching points for subsequent high-precision 3D reconstruction.

[0081] like Figure 3 As shown, the 3D point cloud generation engine 2 includes a feature extraction module 21, a spatial matching module 22, a mismatch elimination module 23, a point cloud registration module 24, and a 3D model generation module 25.

[0082] The feature extraction module 21 is used to extract feature points from multi-angle image data. In one embodiment of the present invention, the feature extraction module 21 uses the SIFT (Scale Invariant Feature Transform) algorithm to extract image feature points. The SIFT algorithm has the characteristics of scale invariance, rotation invariance and insensitivity to illumination changes, and is suitable for processing multi-view images acquired by UAV oblique photography.

[0083] The spatial matching module 22 is used to establish a KD-tree structure and perform feature point matching. A KD-tree is a spatial partitioning tree that effectively reduces the search complexity in feature point matching. In this invention, the spatial matching module 22 first constructs a KD-tree index structure, and then uses a nearest neighbor search algorithm for feature point matching, significantly improving matching efficiency.

[0084] The mismatch elimination module 23 is used to remove mismatches of feature points using the RANSAC (Random Sample Consensus) algorithm. The RANSAC algorithm effectively identifies and eliminates mismatches that do not conform to geometric consistency constraints through iterative random sampling and model fitting, improving the reliability of the matching results. In a preferred embodiment of the invention, the number of RANSAC iterations is set to 1000, and the interior point threshold is set to 3 pixels. These parameter settings ensure both algorithm robustness and high computational efficiency.

[0085] The point cloud registration module 24 is used to perform initial registration and global fine registration of point cloud data using the Iterative Closest Point (ICP) algorithm and a global fine registration algorithm. Preferably, the initial registration uses coarse feature point matching results, and the global fine registration criterion uses an improved ICP algorithm to achieve high-precision registration by minimizing the distance error between point sets. In one embodiment of the present invention, the convergence threshold of the ICP algorithm is set to 0.001 meters, and the maximum number of iterations is set to 50. These parameter settings can control the computational complexity while ensuring registration accuracy.

[0086] The 3D model generation module 25 is used to generate a dense point cloud model and a 3D model through a progressive triangulation strategy. Preferably, the 3D model generation module 25 first generates a dense point cloud based on the registered sparse point cloud using a multi-view stereo matching algorithm, then generates a triangular mesh model using a Poisson surface reconstruction algorithm, and finally performs texture mapping to generate a 3D model with realistic texture. In one embodiment of the present invention, the resolution of the dense point cloud is set to 5 cm, which can meet the accuracy requirements of roadbed deformation monitoring.

[0087] The deformation field analysis platform 3 is the core innovation of this invention. Based on differential geometry theory, it performs in-depth analysis of three-dimensional point cloud data, extracts deformation features, and identifies potential risks. For example... Figure 1 As shown, the deformation field analysis platform 3 includes a manifold modeling unit 31, a curvature analysis unit 32, a topological feature unit 33, an adaptive monitoring and control unit 34, and a multi-source data fusion unit 35.

[0088] like Figure 4 As shown, the manifold modeling unit 31 includes a point cloud preprocessing subunit 311, a parameterized mapping subunit 312, a metric tensor calculation subunit 313, and a manifold optimization subunit 314.

[0089] The point cloud preprocessing subunit 311 is used to perform noise reduction, registration, and mesh generation on the 3D point cloud data. Preferably, noise reduction employs a statistical outlier filtering algorithm, registration employs an iterative nearest-point algorithm, and mesh generation employs an adaptive Delaunay triangulation algorithm. These preprocessing steps can effectively improve the accuracy and efficiency of subsequent analysis. In one embodiment of the present invention, the standard deviation factor of the statistical outlier filtering algorithm is set to 2.5, which can effectively remove noise points while retaining valid information; the minimum angle constraint of the Delaunay triangulation is set to 15 degrees, which can generate a high-quality triangular mesh.

[0090] The parameterization mapping subunit 312 is used to establish the mapping relationship between the 3D point cloud and the parameter domain, generating a parameterized surface. This invention uses a non-uniform rational B-spline (NURBS) surface for parameterization representation. NURBS surfaces possess good local controllability and mathematical properties, making them suitable for representing complex roadbed surfaces. The parameterization mapping process can be represented as follows: .in: Let be a parametric surface, representing the coordinates of a point in three-dimensional space, which are about the parameters. The function; The coordinates are defined in the parameter domain in a two-dimensional parameter space. middle; These are rational basis functions used to control the shape of the surface; For control points, the coordinates of a point in three-dimensional space are used. and They are respectively direction and The number of control points for direction is usually determined based on the complexity of the surface. In this embodiment... , . This represents the summation of the contributions of all control points, i.e., from... arrive ,from arrive The double summation.

[0091] rational basis functions Defined as: ,in: This is a weighting factor that controls the degree of influence of control points on the surface. It is usually set to a positive value, and the default value in this embodiment is 0. ; and They are respectively direction and The B-spline basis functions in the direction are recursively defined polynomial functions; and They are respectively direction and The order of the spline in the direction determines the smoothness of the surface. In this embodiment... , representing a cubic B-spline; This indicates that the contribution of all control point weights is normalized, i.e., from... arrive ,from arrive The double summation.

[0092] The metric tensor computation subunit 313 is used to compute metric tensors on parameterized surfaces, characterizing the intrinsic geometric properties of the manifold. The metric tensor is a fundamental tool for describing the local geometric properties of Riemannian manifolds; it defines metrics such as length, angle, and area on the manifold. The calculation formula is: ,in: To measure a tensor, it is a Matrix; and Represents parameter coordinates The components in, i.e. and Representing parametric surfaces Regarding parameters and The partial derivative of is a vector in three-dimensional space; it represents the dot product operation of vectors.

[0093] Specifically, when hour, when or hour, when hour, The manifold optimization subunit 314 is used to optimize the parameterized mapping, reduce distortion, and improve the parameterization quality.

[0094] Preferably, the manifold optimization sub-unit 314 employs an energy minimization method, achieving parameterized optimization by minimizing the following energy function: .in: The total energy function represents a measure of parameterized quality. It represents the area distortion energy and quantifies the degree of change in area ratio during parameterization. It represents the energy of angular distortion and quantifies the degree of angular preservation during parameterization. Represents the boundary constraint energy, ensuring that the parameterized boundary satisfies specific conditions; , γ and are weighting coefficients that control the relative importance of each energy term, taking values ​​in the range [0,1] and satisfying the following conditions: In a preferred embodiment of the present invention, and This weighting configuration ensures that boundary constraints are satisfied while maintaining area and angle preservation.

[0095] Furthermore, in one embodiment of the present invention, the control point grid density of the parametric surface is set to 10-30 points / square meter, with a base density of 10 points / square meter, which can be adaptively increased to 30 points / square meter in high curvature regions. This density configuration ensures both the accuracy of surface representation and control of computational complexity. Regarding parametric quality, the area distortion rate is controlled within 5%, and the angular distortion rate is controlled within 10 degrees, ensuring that the parametric surface can accurately represent the geometric characteristics of the roadbed surface.

[0096] like Figure 5 As shown, the curvature analysis unit 32 includes a basic curvature calculation subunit 321, a principal curvature extraction subunit 322, a multi-scale decomposition subunit 323, and a curvature flow dynamic analysis subunit 324.

[0097] The basic curvature calculation subunit 321 is used to calculate the Gaussian curvature and mean curvature at various points on the parameterized Riemannian manifold model. Gaussian curvature K and mean curvature H are important indicators describing the local geometric properties of a surface, and their calculation formulas are as follows: , .in: Gaussian curvature, representing the intrinsic curvature of a surface, is expressed in units of 1200 ppm. ; The mean curvature represents the intrinsic curvature of the surface, with units of . ; and The matrices representing the first and second fundamental forms respectively describe the metric and curvature properties of the surface. and Represent matrices respectively and The determinant of; Representation matrix The trace, that is, the sum of the elements on the main diagonal; , , and , , These are the coefficients of the first and second fundamental forms, respectively. These coefficients can be calculated using the partial derivatives of the parametric surface: . .in: and Representing a curved surface Regarding parameters and The first-order partial derivative of is a three-dimensional vector; , and Representing a curved surface The second-order partial derivative; This represents the dot product operation; The unit normal vector representing the surface is a three-dimensional vector of length 1, which can be expressed as a unit normal vector. The calculation yielded, where This represents the cross product operation. It represents the magnitude of the vector.

[0098] Principal curvature extraction sub-unit 322 is used to calculate the principal curvature and corresponding principal direction at each point. Principal curvature and It represents the maximum and minimum values ​​of curvature at a point on a surface along any direction, in units of . The calculation formula is as follows: ,in: Indicates the maximum principal curvature. Indicates the minimum principal curvature. For the mean curvature, Let ± represent the Gaussian curvature, where ± indicates that the two principal curvatures are respectively... and .

[0099] The principal direction corresponds to the principal curvature and can be obtained by solving the following eigenvalue problem: .in: This represents the second fundamental form of the matrix. This represents the first basic form of the matrix. The principal direction vector is the parameter space. Two-dimensional vectors in This represents the corresponding principal curvature value, which can be... or In roadbed deformation monitoring, the principal curvature direction usually coincides with the main deformation direction, and therefore can be used as an important indicator for analyzing deformation patterns.

[0100] Multiscale decomposition subunit 323 is used to construct a multiscale representation of curvature using a Gaussian smoothing kernel, decomposing the curvature features into large-scale, medium-scale, and small-scale components. Multiscale decomposition is based on scale-space theory, through the analysis of the curvature function... and By applying Gaussian kernels of different scales for convolution, curvature representations at multiple scale levels are obtained: . .in: and The scale is represented as Gaussian curvature and mean curvature, in units of respectively and ; and Represents the original Gaussian curvature and the mean curvature function; The standard deviation is expressed as The two-dimensional Gaussian kernel is defined as ,in The standard deviation of the Gaussian kernel is expressed in units of 1000 ppm. ; To represent a two-dimensional convolution operation, it is defined as follows: In a preferred embodiment of the invention, the standard deviation sequence of the Gaussian kernel is set as follows: , respectively corresponding to small scale ( ), mesoscale ( ) and large scale ( The deformation characteristics of the structure. This multi-scale decomposition can simultaneously capture various deformation modes, from microcracks to large-scale settlement.

[0101] The curvature flow dynamic analysis subunit 324 is used to track the evolution trajectory of curvature extrema over time, constructing the surface deformation rate field and acceleration field. Curvature flow is a mathematical model describing the evolution of a surface over time, and its basic form can be expressed as: .

[0102] in: Representing a curved surface Over time The rate of change of is a three-dimensional vector field, with units of . ; It's about the principal curvature. and The function defines the rate of surface evolution; The normal vector of the surface is a unit three-dimensional vector. It represents the product of a scalar and a vector.

[0103] Deformation rate field and acceleration field It can be calculated using the following formula: . .in: The deformation rate field is a three-dimensional vector field with units of . ; This represents the deformation acceleration field, which is a three-dimensional vector field with units of 1. ; Indicates time Parametric surface at time; Indicates the interval between adjacent monitoring times, in units of In practical applications, deformation rate and acceleration are important indicators for judging roadbed stability. In a preferred embodiment of the present invention, the warning value for vertical deformation rate is set to 5 mm / day, the warning value for horizontal deformation rate is set to 3 mm / day, and the warning value for deformation acceleration is set to 0.5 mm / day². When the monitored deformation rate or acceleration exceeds these thresholds, the system will trigger the corresponding early warning mechanism.

[0104] like Figure 6As shown, the topological feature unit 33 includes a singularity detection subunit 331, a singularity classification subunit 332, a singularity tracking subunit 333, and a topological evolution map construction subunit 334.

[0105] The singularity detection subunit 331 is used to locate singularities on a parameterized Riemannian manifold model based on curvature and curvature gradient features. In differential geometry, singularities are special points in a curvature field, representing locations where the surface shape undergoes significant changes. Singularity detection is achieved by analyzing the curvature gradient field. and In this implementation, singular points typically correspond to locations where the gradient magnitude is locally maxima or where the gradient direction undergoes a singular change. .in: and Representing Gaussian curvature respectively and mean curvature The gradient vectors are respectively about the parameters. Two-dimensional vectors, with units of , and ; The coordinates of the singular point are represented by parameters. Represents the magnitude of the vector; and These are the thresholds for Gaussian curvature and mean curvature gradient, respectively, with units of... and ; This represents a logical OR operation, meaning that a point is identified as a singular if either condition is met. In a preferred embodiment of the invention, Set to the negative first power of 0.005 square meters. Setting these thresholds to the negative first power of 0.01 meters effectively identifies key singularities in roadbed deformation while avoiding the misidentification of too many noise points.

[0106] Singularity classification subunit 332 is used to classify singularities into elliptic, hyperbolic, parabolic, or degenerate types. The type of a singularity is closely related to its local curvature structure and can be determined by its principal curvature. and The sign and relative size are used to determine the value:

[0107] Elliptical singularity: and "Approaching" indicates that a part of the body is either concave or convex.

[0108] Hyperbolic singularity: This indicates that the local area is saddle-shaped;

[0109] Parabolic singularity: and or , indicating that a part is cylindrical in shape;

[0110] Degenerate singularities: However, the higher-order derivatives are not zero, indicating that the local area is flat but there are higher-order variations.

[0111] In roadbed deformation monitoring, different types of singularities correspond to different deformation patterns and risk levels. For example, hyperbolic singularities usually indicate stress concentration areas and may be a precursor to potential landslides; parabolic singularities may indicate unidirectional bending and are commonly seen in the early stages of roadbed settlement.

[0112] The singularity tracking subunit 333 is used to match and track identical singularities across consecutive time frames, recording the lifecycle of each singularity. The singularity tracking employs a combination of an improved Kalman filter and a Hungarian algorithm. The Kalman filter predicts the singularity's position at the next time step, while the Hungarian algorithm solves the singularity matching problem. Preferably, the matching process considers multiple features of the singularity, such as its position, type, intensity, and local shape, to improve the accuracy and robustness of the tracking.

[0113] Subunit 334, responsible for constructing the topological evolution graph, is used to build a deformation topological evolution graph based on singular point trajectories and interaction events. (Topological evolution graph) It is a directed graph, where the vertex set is... Denotes singular points and edge sets. This represents the evolutionary relationships between singularities. These relationships include events such as the generation, disappearance, merging, splitting, and type transformation of singularities. By analyzing the topological evolution map, specific evolutionary patterns can be identified, and these patterns are often closely related to changes in roadbed stability.

[0114] In one embodiment of the present invention, the system focuses on the following typical topology evolution modes:

[0115] 1. A sharp increase in singularity density: A significant increase in the number of singularities in a specific region within a short period of time usually indicates that the region is beginning to show signs of instability;

[0116] 2. Continuous appearance of hyperbolic singularities: Hyperbolic singularities continuously appear in a certain region over multiple consecutive time frames, indicating that the stress concentration state persists.

[0117] 3. Singularity type transformation: The transformation of a originally stable elliptical singularity into a hyperbolic or parabolic singularity usually indicates a change in the local deformation pattern;

[0118] 4. Singularity Merging Event: Multiple singularities gradually approach each other and eventually merge into one, which usually indicates an increase in local deformation.

[0119] These topological evolution patterns are significantly correlated with the risk of roadbed landslides and serve as an important basis for the system to conduct risk early warning.

[0120] The adaptive monitoring and control unit 34 is used to automatically adjust the time interval and spatial resolution of subsequent monitoring based on the deformation analysis results, increase the monitoring density in high-risk areas, and send optimized flight paths and shooting parameters to the UAV oblique photography module.

[0121] Preferably, the adaptive monitoring and control unit 34 sets different monitoring strategies according to the risk level:

[0122] 1. Low-risk areas: Standard monitoring interval is 7 days, flight altitude is 200-220 meters, and image overlap is 90%;

[0123] 2. Medium-risk areas: Monitoring intervals shortened to 3-5 days, flight altitude 180-200 meters, image overlap 92%;

[0124] 3. High-risk areas: The monitoring interval is shortened to 1-2 days, the flight altitude is 150-180 meters, the image overlap is 95%, and special local encrypted flight routes are added for this area.

[0125] This adaptive monitoring strategy can optimize resource allocation and improve the overall efficiency and economy of the system while ensuring monitoring effectiveness.

[0126] The multi-source data fusion unit 35 is used to receive multi-source information such as meteorological data, geological survey data and traffic load data, analyze the impact of these factors on roadbed deformation, and improve the accuracy of deformation analysis and risk warning.

[0127] In a preferred embodiment of the present invention, multi-source data fusion employs a fusion method based on Dempster-Shafer evidence theory, which can effectively handle uncertainties and conflicts between data from different sources. The multi-source data is first unified to the same scale through normalization processing: .in: This represents the normalized data, which is dimensionless and takes values ​​in the range [0,1]. This represents raw data, and the unit depends on the specific data type. and These represent the minimum and maximum values ​​of the data, respectively, with the same units as the original data.

[0128] Then, basic probability mass (BPA) is assigned based on the reliability of each data source and the significance of its impact on deformation, and then fused using Dempster's combination rule: .

[0129] in: denoted as the basic probability mass of hypothesis A after fusion, dimensionless, with a value range of [0,1]; This indicates the basic probability quality of the first source of evidence for hypothesis B; This indicates the basic probability quality of the second source of evidence for hypothesis C; The statement that the intersection of hypothesis B and hypothesis C equals hypothesis A is a set operation. This indicates that the intersection of hypothesis B and hypothesis C is an empty set, signifying a conflict between the two sources of evidence. It means that for all satisfying Summing the assumptions over (B, C); It means that for all satisfying The assumption is to sum the values ​​of (B,C).

[0130] Multi-source data fusion can provide a more comprehensive assessment of roadbed condition, reduce misjudgments that may result from a single data source, and improve the accuracy and reliability of risk warnings.

[0131] like Figure 7 As shown, the risk assessor 4 includes a risk pattern library 41, a topology matching module 42, a risk level assessment module 43, and an early warning information generation module 44.

[0132] The risk pattern library 41 is used to store typical risk topology patterns derived from historical monitoring data and expert knowledge. In one embodiment of the invention, the risk pattern library contains multiple typical landslide precursor topology patterns, each associated with a specific risk level and development trend.

[0133] The topology matching module 42 is used to match the current topology evolution state with patterns in the risk pattern library. The matching process is based on graph structure similarity calculation, taking into account the degree of matching of node features and edge relationships. Preferably, the matching adopts a method combining subgraph isomorphism algorithm and graph edit distance, which can identify patterns that match completely, as well as handle cases of partial and approximate matching.

[0134] The risk level assessment module 43 is used to comprehensively consider singularity density, rate of change, and topological stability to assess the risk level. The risk level assessment is based on the following indicators:

[0135] 1. Singularity density: Calculates the number of singularities per unit area. The higher the density, the higher the risk level.

[0136] 2. Rate of Change: Calculates the rate of change of the number, location, and type of singularities. The higher the rate of change, the higher the risk level.

[0137] 3. Topological stability: Evaluate the stability of the topological structure based on structural changes in the topological evolution map.

[0138] In a preferred embodiment of the present invention, the risk level is divided into four levels:

[0139] Low risk: Singularity density <0.05 / m², stable topology;

[0140] Medium risk: Singularity density 0.05-0.1 per square meter, topological structure changes gradually;

[0141] High risk: Singularity density 0.1-0.2 per square meter, with the emergence of new types of singularities;

[0142] Extremely high risk: Singularity density > 0.2 points / square meter, rapid changes in topology.

[0143] The early warning information generation module 44 is used to generate graded early warning information based on risk levels and indicate the location of high-risk areas. The early warning information includes risk level, risk area, development trend, and recommended measures, presented in a visual manner to facilitate quick understanding and decision-making by managers.

[0144] The parameterized Riemannian manifold model in Deformation Field Analysis Platform 3 has the following characteristics:

[0145] 1. It can maintain geometric continuity and achieve an accurate mathematical description of the roadbed deformation process. Through NURBS surface representation, the model can accurately express complex roadbed surface shapes while maintaining continuity.

[0146] 2. An adaptive control point density distribution is employed in high curvature regions, with a control point grid density of 10-30 points / square meter. This adaptive strategy optimizes the utilization of computational resources while maintaining accuracy.

[0147] 3. The area distortion rate is less than 5%, and the angular distortion rate is less than 10°. These parameters ensure the preservation of geometric properties during the parameterization process, enabling the manifold model to accurately reflect the deformation of the actual roadbed surface.

[0148] 4. By using the principal curvature direction as the basis of the local coordinate system, deformation analysis is ensured to be performed along the natural deformation direction of the roadbed. This design enables the system to better capture the intrinsic mechanism of deformation, improving the accuracy and effectiveness of the analysis.

[0149] The typical operating flow of the system of this invention includes the following steps:

[0150] 1. Initialization phase: The UAV oblique photography module 1 takes the first aerial photograph of the monitoring area according to the preset route and collects the image data of the baseline state; the 3D point cloud generation engine 2 processes the image data and generates the baseline 3D point cloud model; the deformation field analysis platform 3 establishes the baseline parameterized Riemann manifold model.

[0151] 2. Regular monitoring phase: The UAV oblique photography module 1 conducts regular aerial photography at planned intervals (usually 7 days); the 3D point cloud generation engine 2 processes the newly acquired image data and generates a 3D point cloud model of the current state; the deformation field analysis platform 3 constructs a parameterized Riemannian manifold model of the current state and compares it with the benchmark model to extract deformation features.

[0152] 3. Risk assessment stage: Deformation field analysis platform 3 analyzes deformation characteristics, identifies singularities and their topological features; risk assessor 4 matches the current topological state with the risk pattern library, assesses the risk level, and generates early warning information.

[0153] 4. Feedback and Adjustment Phase: The adaptive monitoring and control unit 34 adjusts the subsequent monitoring plan based on the risk assessment results, including monitoring intervals, flight path design, and shooting parameters.

[0154] The specific applications of this invention are illustrated below with examples:

[0155] In a highway subgrade monitoring project, the system of this invention was used to conduct three-dimensional deformation monitoring and risk warning for a subgrade section approximately 5 kilometers long. This section is located in a mountainous area with complex geological conditions and has experienced several small landslides in the past.

[0156] First, the drone oblique photography module uses the DJI Motrice 300RTK drone platform, equipped with a five-lens camera system, to conduct aerial photography of the entire monitoring area along a zigzag flight path. The flight altitude is set at 180 meters to ensure that the image overlap reaches more than 92%. After the aerial photography is completed, the image data is transmitted wirelessly to the 3D point cloud generation engine.

[0157] The 3D point cloud generation engine uses the SIFT algorithm to extract image feature points, and performs feature point matching and mismatch elimination using a KD tree structure and the RANSAC algorithm. Then, the ICP algorithm is used for fine registration of the point cloud, and finally, a dense point cloud model is generated through a progressive triangulation strategy, achieving a point cloud density of 25 points / square meter and an accuracy better than 2 centimeters.

[0158] The deformation field analysis platform first preprocesses the 3D point cloud data, including statistical outlier filtering (standard deviation factor set to 2.5) and Delaunay triangulation (minimum angular constraint set to 15 degrees). Then, a parametric Riemannian manifold model is established using NURBS surfaces, with a base point density of 10 points / m², adaptively densified to 25 points / m² in high-curvature regions. The area distortion rate is controlled within 3.2%, and the angular distortion rate within 8.6 degrees.

[0159] In multi-scale curvature analysis, the system used Gaussian kernels with standard deviations of 1 meter, 5 meters, and 20 meters for smoothing, extracting deformation features at large, medium, and small scales. By analyzing the curvature field, the system identified multiple hyperbolic singularities on a roadbed slope, with a singularity density of 0.12 singularities per square meter, showing a gradually increasing trend. Singularity tracking revealed significant changes in the topology of these singularities over three consecutive monitoring sessions; multiple elliptical singularities transformed into hyperbolic ones, and singularity merging events occurred.

[0160] The risk assessor matched the current topological evolution state with patterns in the risk pattern library and found that it had 85% similarity to the precursor patterns of slope slippage. Taking into account the singularity density (0.12 points / m²), the rate of change (30% / week), and topological stability, the system assessed the risk level of the area as high risk, predicting that a landslide may occur within 15-30 days.

[0161] The system generates early warning information, indicating the location, extent, and expected development trend of high-risk areas, and recommends strengthening monitoring and taking protective measures. Simultaneously, the adaptive monitoring and control unit shortens the monitoring interval for this area to 2 days, lowers the flight altitude to 160 meters, increases image overlap to 95%, and adds locally enhanced flight routes specifically targeting this area.

[0162] Subsequent monitoring confirmed that significant signs of landslides did indeed appear in the high-risk area 25 days later. However, thanks to the system's early warning and corresponding protective measures, a large-scale landslide accident was avoided, effectively ensuring the safe operation of the highway.

[0163] As can be seen from this example, the system of the present invention, through the innovative application of differential geometry theory, achieves high-precision monitoring of roadbed deformation and early warning of landslide risk, which has significant practical value and social benefits.

[0164] The embodiments described above are merely illustrative of specific implementations of the present invention, and while the descriptions are detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention.

Claims

1. A three-dimensional deformation monitoring system for highway subgrade based on UAV oblique photography, including: The UAV oblique photography module is used to acquire multi-angle image data of the highway subgrade and transmit it to the 3D point cloud generation engine; A 3D point cloud generation engine, which is communicatively connected to the UAV oblique photography module, is used to receive the multi-angle image data and generate 3D point cloud data over a time series. A deformation field analysis platform, communicatively connected to the 3D point cloud generation engine, is used to receive the 3D point cloud data over the time series and perform deformation analysis based on differential geometry theory. The deformation field analysis platform includes: The manifold modeling unit is used to convert the 3D point cloud data into a parameterized Riemannian manifold model. The parameterized Riemannian manifold model is represented parametrically using a non-uniform rational B-spline NURBS surface, wherein the parameterization mapping process is expressed as follows: ,in Let be a parametric surface, representing the coordinates of a point in three-dimensional space, which are about the parameters. and The function; The coordinates are defined in the parameter domain in a two-dimensional parameter space. middle; These are rational basis functions used to control the shape of the surface; For control points, the coordinates of a point in three-dimensional space are used. and They are respectively direction and Number of control points for direction; The curvature analysis unit is used to calculate the multi-scale curvature features on the parameterized Riemannian manifold model, and the curvature analysis unit calculates the Gaussian curvature. and mean curvature Its calculation formula is: , ,in and Matrices representing the first and second basic forms, respectively. and Represent matrices respectively and determinant, Representation matrix traces, , , and , , These are the coefficients of the first and second basic forms, respectively; the curvature analysis unit convolves the curvature function with Gaussian kernels of different scales to generate curvature representations at multiple scale levels, and the formula for the multi-scale decomposition is... , ,in and The scale is represented as Gaussian curvature and mean curvature and Represents the original Gaussian curvature and the mean curvature function. The standard deviation is expressed as Two-dimensional Gaussian kernel, This represents a two-dimensional convolution operation, where the standard deviation sequence is set as follows: These correspond to deformation characteristics at small, medium, and large scales, respectively. Topological feature units are used to identify deformation singularities and their topological features on the parameterized Riemannian manifold model, the topological feature units being based on curvature gradient fields. and Singularity detection is performed under the following conditions: ,in and Representing Gaussian curvature respectively and mean curvature gradient vector, The parameter coordinates representing singular points, Represents the magnitude of a vector. and These are the thresholds for Gaussian curvature and mean curvature gradient, respectively; OR represents a logical OR operation. The topological feature unit classifies singularities into elliptic, hyperbolic, parabolic, or degenerate types, and the classification judgment is based on the principal curvature. and The sign and relative size of the elliptical singularities satisfy the following: and Hyperbolic singularities satisfy Parabolic singularities satisfy and or Degenerate singularities satisfy The topological feature unit matches and tracks the same singular points across consecutive time frames, records the lifecycle of the singular points, and constructs a deformation topological evolution map based on the singular point trajectories and interaction events. A risk assessor, communicatively connected to the deformation field analysis platform, is used to predict roadbed landslide risk and generate graded early warning information based on the temporal evolution of the deformation singularities and their topological characteristics. The risk assessor comprehensively considers singularity density, rate of change, and topological stability to evaluate the risk level. The risk levels are divided into low risk, medium risk, high risk, and extremely high risk. Low risk corresponds to a singularity density of less than 0.05 points / m² and a stable topological structure; medium risk corresponds to a singularity density of 0.05-0.1 points / m² and a slowly changing topological structure; high risk corresponds to a singularity density of 0.1-0.2 points / m² and the emergence of new singularities; and extremely high risk corresponds to a singularity density of greater than 0.2 points / m² and a rapidly changing topological structure. The manifold modeling unit includes: The point cloud preprocessing subunit is used to perform noise reduction, registration, and mesh generation on the 3D point cloud data. The parameterized mapping sub-unit is used to establish the mapping relationship between the 3D point cloud and the parameter domain, and to generate parameterized surfaces; The metric tensor computation sub-unit is used to compute the metric tensor on the parameterized surface, characterizing the intrinsic geometric properties of the manifold; The manifold optimization sub-unit is used to optimize the parameterized mapping, reduce distortion, and improve the quality of parameterization. The topological feature unit includes: A singularity detection subunit is used to locate singularities on a parameterized Riemannian manifold model based on curvature and curvature gradient features; A singularity classification subunit is used to classify the singularities into elliptical, hyperbolic, parabolic, or degenerate types; The singularity tracking subunit is used to match and track the same singularities across consecutive time frames and record the lifecycle of the singularities; The topology evolution map construction subunit is used to construct deformation topology evolution maps based on singular point trajectories and interaction events.

2. The system according to claim 1, characterized in that... The UAV oblique photography module includes: The four-lens photography unit includes four cameras, with the included angle between adjacent cameras set at 90 degrees, for capturing omnidirectional images around the roadbed. The flight control unit is used to control the UAV to fly along a preset route, which is either bow-shaped or zig-shaped. The data synchronization unit is used to ensure that the four cameras capture images synchronously and that the overlap between adjacent shooting areas is not less than 90%.

3. The system according to claim 1, characterized in that... The 3D point cloud generation engine includes: The feature extraction module is used to extract feature points from the multi-angle image data; The spatial matching module is used to build the KD tree structure and perform feature point matching. The mismatch elimination module is used to remove mismatches of feature points using the RANSAC algorithm. The point cloud registration module is used to perform initial registration and global fine registration of point cloud data using the iterative nearest point algorithm and the ICP algorithm. The 3D model generation module is used to generate dense point cloud models and 3D models through a progressive triangulation strategy.

4. The system according to claim 1, characterized in that... The curvature analysis unit includes: The basic curvature calculation subunit is used to calculate the Gaussian curvature and mean curvature at each point on the parameterized Riemannian manifold model; The principal curvature extraction sub-unit is used to calculate the principal curvature and corresponding principal direction at each point; Multi-scale decomposition subunits are used to construct a multi-scale representation of curvature using a Gaussian smoothing kernel, decomposing curvature features into large-scale, medium-scale, and small-scale components; The curvature flow dynamic analysis sub-unit is used to track the evolution trajectory of curvature extrema over time and construct the surface deformation rate field and acceleration field.

5. The system according to claim 1, characterized in that... The risk assessor includes: A risk pattern library for storing typical risk topology patterns; The topology matching module is used to match the current topology evolution state with patterns in the risk pattern library; The risk level assessment module is used to comprehensively consider singularity density, rate of change, and topological stability to assess the risk level. The early warning information generation module is used to generate graded early warning information based on the risk level and indicate the location of high-risk areas.

6. The system according to claim 1, characterized in that... The deformation field analysis platform also includes an adaptive monitoring and control unit, used for: The time interval and spatial resolution of subsequent monitoring are automatically adjusted based on the deformation analysis results. Increase monitoring density in high-risk areas to improve early warning sensitivity; The optimized flight path and shooting parameters are sent to the UAV tilt photography module.

7. The system according to claim 1, characterized in that... The deformation field analysis platform also includes a multi-source data fusion unit, used for: Receive meteorological data and analyze the impact of rainfall on deformation; Receive geological survey data and analyze stratigraphic structural factors; Receive traffic load data and assess the impact of dynamic loads; Improve the accuracy of deformation analysis and risk warning by leveraging fused multi-source data.

8. The system according to claim 1, characterized in that... The parameterized Riemannian manifold model in the deformation field analysis platform has the following characteristics: It can maintain geometric continuity and achieve a precise mathematical description of the roadbed deformation process; In high curvature regions, there is an adaptive control point density distribution, with a control point grid density of 10-30 points / square meter; Area distortion rate is less than 5%, and angular distortion rate is less than 10°; By using the principal curvature direction as the basis of the local coordinate system, deformation analysis is ensured to be performed along the natural deformation direction of the roadbed.

Images