Construction settlement visual measurement method and system based on deep learning compensation

Abstract

The application discloses a structure settlement visual measurement method and system based on deep learning compensation, and relates to the technical field of building settlement measurement. The method comprises the following steps: according to a mobile detection platform, driving a detection component to collect a structure region and determining structure data; constructing a settlement measurement component by spatial decomposition and implicit causal analysis of settlement dimensions, motion dimensions and interference dimensions; performing background domain segmentation on the structure data, executing subspace decomposition feature recognition and implicit causal coupling analysis, and determining settlement data pairs; performing directional projection and three-dimensional mapping through a three-dimensional projector, realizing three-dimensional distance measurement, and generating and updating a settlement curve display. The technical problems of the existing structure settlement measurement, such as the difficulty in effectively correcting the spatial displacement measurement error caused by attitude change and environmental interference, and the insufficient distance measurement accuracy, are solved, the technical effect of realizing high-precision non-contact distance measurement and dynamic compensation of structure settlement displacement is achieved, and the stability of spatial distance measurement and the environmental adaptability are improved.

Description

Technical Field

[0001] This invention relates to the field of building settlement measurement technology, specifically to a visual measurement method and system for structure settlement based on deep learning compensation. Background Technology

[0002] With the acceleration of urbanization, the construction scale of underground spaces and large-scale civil engineering structures is constantly expanding, making settlement monitoring of structures a crucial link in ensuring structural safety and engineering quality. Traditional settlement measurement mainly relies on contact sensors or manual leveling methods, such as levels, total stations, and settlement sensors. While these methods offer a certain level of accuracy, they suffer from problems such as complex deployment, high cost, poor real-time performance, and susceptibility to environmental interference, making it difficult to meet the demands of modern engineering for large-scale, continuous, and high-precision monitoring.

[0003] In recent years, spatial geometric ranging technology based on non-contact measurement principles has developed rapidly. Utilizing computer imaging or optical projection principles, non-contact ranging of structural surfaces can be achieved, providing a new approach to settlement monitoring. However, due to factors such as changes in the motion posture of the detection platform, ambient light interference, and background variations, existing spatial ranging methods are still prone to problems such as the accumulation of ranging errors and unstable positioning, resulting in insufficient accuracy of settlement curves. Summary of the Invention

[0004] This application provides a visual measurement method and system for structure settlement based on deep learning compensation, which solves the technical problems of insufficient accuracy and difficulty in effectively correcting spatial displacement ranging errors caused by posture changes and environmental interference in existing structure settlement measurements.

[0005] The first aspect of this application provides a visual measurement method for structure settlement based on deep learning compensation, the method comprising:

[0006] Based on the mobile detection platform, the driving detection component collects data on the structure area to determine the structure data. A settlement measurement component is constructed using spatial decomposition of the structure based on settlement, motion, and disturbance dimensions, and analysis based on implicit causal relationships. By segmenting the structure data into a background domain, structured subspace decomposition feature recognition based on the settlement measurement component and coupled analysis based on implicit causal relationships are performed to determine settlement data pairs, where each set of settlement data pairs represents the settlement distance at any location. Using a 3D projector, the settlement data pairs are written in 2D and mapped in 3D based on a directional view projection plane to perform 3D distance measurement, determine 3D settlement data, update the settlement curve, and display it on the interface.

[0007] A second aspect of this application provides a visual measurement system for structure settlement based on deep learning compensation, the system comprising:

[0008] Data Acquisition Module: Based on the mobile detection platform, the detection component is driven to collect data on the structure area and determine the structure data; Measurement Component Construction Module: Based on the spatial decomposition of the structure based on settlement, motion, and disturbance dimensions and the analysis based on implicit causality, a settlement measurement component is constructed; Feature Analysis Module: By segmenting the structure data into a background domain, a structured subspace decomposition feature recognition based on the settlement measurement component and a coupled analysis based on implicit causality are performed to determine settlement data pairs, where a set of settlement data pairs represents the settlement distance at any location; 3D Measurement Module: Through a 3D projector, the settlement data pairs are written in 2D and mapped in 3D based on a directional view projection plane, 3D distance measurement is performed, 3D settlement data is determined, updated to the settlement curve, and displayed on the interface.

[0009] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0010] First, a mobile detection platform drives the detection components to collect raw data about the structure area non-contactly. Then, spatial information is decomposed and analyzed from three dimensions: settlement change, platform movement, and external disturbances. A settlement measurement model is constructed based on implicit causal relationships to identify and compensate for distance measurement errors caused by movement or environmental factors. Next, the collected structure data undergoes background segmentation and feature extraction. Through structured subspace decomposition and causal coupling analysis, data pairs reflecting the settlement distances at different locations of the structure are obtained. Finally, a 3D projector is used to project this settlement data in 2D and map it in 3D from a directional perspective, achieving high-precision 3D distance measurement, generating settlement curves, and updating and displaying them in real time. This enables dynamic monitoring and visualization of the structure's settlement status. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 A schematic diagram of the visual measurement method for structure settlement based on deep learning compensation provided in the embodiments of this application.

[0013] Figure 2 A schematic diagram of the structure of a visual measurement system for building settlement based on deep learning compensation provided in an embodiment of this application.

[0014] Figure labeling: Data acquisition module 11, measurement component construction module 12, feature analysis module 13, 3D measurement module 14. Detailed Implementation

[0015] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0016] Example 1, as Figure 1 As shown, this application provides a visual measurement method for structure settlement based on deep learning compensation, the method including:

[0017] The mobile inspection platform drives the inspection components to collect data on the structure area and determine the structure data.

[0018] In this embodiment, a mobile detection platform is first used as the data acquisition carrier. This mobile detection platform can be an unmanned vehicle, a tracked vehicle, or other structure with stable driving and attitude control capabilities. Subsequently, a detection component mounted on the platform is driven. This component typically includes a high-resolution imaging device, a distance measurement unit, and an attitude sensing unit, used for continuous, panoramic data acquisition of the structure area during platform movement. As the detection platform moves along the structure monitoring path, the detection component automatically acquires spatial information of the structure surface at a preset sampling frequency and viewing angle, and simultaneously records the platform's pose parameters, sampling time, and ambient lighting conditions, forming a structure dataset corresponding to the structure area. This structure dataset contains both the spatial geometric features of the structural surface and retains dynamic parameters related to platform movement and environmental changes, providing complete basic data support for subsequent settlement decomposition analysis and distance calculation.

[0019] A settlement measurement component is constructed by decomposing the structure based on settlement, motion, and disturbance dimensions and by analyzing implicit causal relationships.

[0020] In one embodiment, to achieve high-precision identification and ranging compensation for structure settlement, three analysis dimensions are first defined: settlement dimension, motion dimension, and disturbance dimension. The settlement dimension describes the vertical and horizontal displacement characteristics of the structure itself caused by foundation deformation or load changes. The motion dimension characterizes the attitude changes and spatial displacements of the detection platform and camera components during data acquisition, including six degrees of freedom motion characteristics such as translation, rotation, and tilt. The disturbance dimension quantifies external noise introduced by changes in illumination, jitter errors, air disturbances, and other non-structural environmental influences. By decomposing and analyzing these three dimensions, the original structure data can be mapped to a structured subspace, allowing settlement, motion, and disturbance characteristics to be expressed in independent spaces. Subsequently, using these dimensional data as input causes and changes in data characteristics as effects, implicit causal relationships are established. Then, a deep learning network is used to jointly train and optimize the weights of the established implicit causal relationships, forming a settlement measurement component that can adaptively identify and separate multi-dimensional influencing factors. In subsequent analysis, this settlement measurement component can automatically correct the ranging errors caused by the movement of the detection platform and environmental interference, achieving high-precision modeling and measurement of the actual settlement displacement of the structure.

[0021] Furthermore, the spatial decomposition of structures based on settlement, motion, and disturbance dimensions includes:

[0022] A settlement factor is defined, wherein the settlement factor encodes the pure deformation of the structure in two-dimensional space; a motion factor is defined, wherein the motion factor encodes the six-degree-of-freedom rigid body motion of the detection component, defined by the homography transformation law of the moving spatial transformation; an interference factor is defined, wherein the interference factor encodes light and shadow changes and jitter interference, defined by extracting stable features from the background data, wherein the jitter interference is the numerical relationship between the jitter of the detection component and the displacement of the structure; based on the settlement factor, motion factor, and interference factor, the structure space is decomposed to generate a structured subspace.

[0023] Preferably, when decomposing the structure's space, a settlement factor is first defined to compare the spatial geometric data of the structure collected at different times. The displacement and deformation characteristics in the two-dimensional plane are extracted according to the variable name corresponding to the settlement factor. This settlement factor only reflects pure structural deformation information caused by foundation deformation or structural stress changes, and does not include changes caused by the movement of the detection equipment or external interference, thus accurately encoding the settlement state of the structure. Secondly, a motion factor is defined. The detection component operates on a mobile detection platform, and its position and orientation undergo six-degree-of-freedom rigid body motion over time. To accurately describe the impact of this motion on the measurement data, a spatial transformation law model is established based on the homography transformation principle. The orientation changes, viewing angle shifts, and imaging displacements of the detection component are uniformly encoded as motion factors, so that these dynamic disturbances can be decoupled and compensated in subsequent analysis. Subsequently, in the detection environment, changes in ambient lighting, background jitter, and air disturbances can all interfere with the data. By performing inter-frame difference analysis on the background image, regions that remain relatively stable over time and do not change with the deformation of structures, such as distant static backgrounds or fixed reference objects, are identified for subsequent interference feature extraction. Then, based on edge, corner, or texture feature point extraction algorithms, such as SIFT and ORB, feature detection and description operations are performed on the identified stable regions to obtain a set of key feature points that are time-stable and less affected by lighting changes. These points are used to characterize the inherent appearance features of the background, providing a foundation for modeling light and shadow changes. Then, for the background features at different sampling times, the changes in brightness, contrast, and chromaticity are calculated. By constructing a lighting change model based on grayscale distribution regression or convolutional neural networks, the changes in light and shadow are quantified into lighting interference coefficients, which are used to encode the impact of lighting changes under different ambient light conditions on the measurement results. During movement, the detection component generates minute vibrations. Attitude sensors, such as an IMU, acquire the component's angular velocity, acceleration, and displacement information in real time. These collected attitude change data are then synchronized with the displacement response data of the structure area through time synchronization and numerical regression to establish a functional relationship model between the detection component's vibration and the structure's surface displacement. This model outputs a vibration interference value, characterizing the impact of motion disturbance on spatial ranging. By normalizing the light and shadow interference coefficient and the vibration interference value, an interference factor is defined. This factor quantifies the degree of influence of light and shadow changes and camera shake. Finally, using the settlement factor, motion factor, and interference factor as spatial decomposition conditions, the structure space is spatially decomposed to generate a structured subspace model containing settlement, motion, and interference subspaces. This model separates and independently models different influencing sources, providing a foundation for subsequent causal analysis, deep learning compensation, and high-precision 3D ranging.

[0024] Furthermore, based on the analysis of implicit causal relationships, a settlement measurement component is constructed, including:

[0025] A first implicit causal relationship is established with interference factors and motion factors as causes and changes in data characteristics as effects; a second implicit causal relationship is established with settlement factors and motion factors as causes and changes in data characteristics as effects; after decomposing the structure space, deep learning based on the first and second implicit causal relationships is performed as a settlement measurement component.

[0026] Optionally, firstly, for the interference factor, the interference factor and motion factor are taken as causes, and the pixel values ​​representing data features are taken as effects. A latent causal model is constructed using multivariate regression analysis or a nonlinear mapping network (such as a causal network based on LSTM or Transformer). During construction, the implicit response law of the interference changing with motion is learned by minimizing the feature prediction error. After construction, this latent causal model is defined as the first latent causal relationship. This first latent causal relationship can reveal the potential impact of interference factors such as changes in light and shadow, jitter and blur on the ranging data under different shooting angles and different motion speeds. Secondly, for the settlement factor, the settlement factor and motion factor are taken as causes, and the pixel values ​​representing data features are taken as effects. Similarly, a latent causal model is constructed using multivariate regression analysis or a nonlinear mapping network to obtain the second latent causal relationship. This second latent causal relationship is used to analyze the characteristic response of the structure's pure settlement deformation under different actual motion angles to distinguish the difference between the pseudo displacement caused by motion and the actual settlement of the structure. Then, the first and second latent causal relationships are input into a deep learning framework for joint training. During training, the parameter weights of the two types of relationship models are optimized by minimizing the causal response error function, i.e., the mean square error, thereby obtaining a settlement measurement component with adaptive resolution capability. This settlement measurement component can automatically identify and separate interference effects from real deformation in complex dynamic environments, realize high-precision settlement measurement and error compensation based on pixel-level data features, and provide core algorithm support for three-dimensional spatial ranging and settlement curve construction.

[0027] By segmenting the background domain of the structure data, performing structured subspace decomposition feature recognition based on the settlement measurement component and coupled analysis based on implicit causal relationships, settlement data pairs are determined, wherein a set of settlement data pairs represents the settlement distance at any location.

[0028] In one embodiment, after acquiring structure data, the data is fuzzily segmented into a background domain based on the segmentation boundary. This separates the background area from the target structure area, eliminating interference from non-structural objects and the environmental background, thereby obtaining segmented structure data containing only structure information. Subsequently, the obtained segmented structure data is input into the settlement measurement component. Through its internal structured subspace, the corresponding characteristics of settlement factors, motion factors, and interference factors are identified. Through implicit causal relationship coupling analysis, a set of one-to-one settlement data pairs is formed. Each set of settlement data pairs consists of the coordinate changes of the same location at different monitoring times, directly representing the settlement distance and direction at that location, providing an accurate data foundation for subsequent three-dimensional ranging and settlement curve updates.

[0029] Furthermore, by segmenting the structure data into a background domain, a coupled analysis based on structured subspace decomposition feature recognition and implicit causal relationships based on the settlement measurement component is performed, including:

[0030] The structure data is transmitted back and background domain fuzzy segmentation is performed to determine the segmented structure data based on the segmentation boundary calibration; the segmented structure data is transmitted to the settlement measurement component for settlement distance analysis.

[0031] Optionally, after the mobile detection platform completes data acquisition of the structure area, it transmits the acquired structure data back in real time, providing an accurate data source for subsequent segmentation and ranging analysis. Then, Gaussian filtering and local histogram equalization algorithms are used to suppress noise and normalize brightness in the transmitted structure data, reducing uneven illumination and noise interference, providing a balanced image feature foundation for subsequent segmentation. Next, the spatial gradient, gray-level variance, and texture similarity of pixels are calculated, and the fuzzy boundary of the target area of ​​the structure is determined using fuzzy C-means. Subsequently, edge enhancement and morphological operations, including dilation, erosion, and contour smoothing, are used to accurately calibrate the fuzzy boundary. Combined with the detection platform's pose parameters and camera intrinsic parameter model, spatial back-projection calculations are performed to ensure the geometric accuracy of the calibrated segmentation boundary position in real space. Afterward, based on the determined segmentation boundary position, the structure data is segmented, generating segmented structure data containing only the target area of ​​the structure. This segmented structure data retains the spatial geometric information and texture features of the structural surface, eliminating the influence of redundant background data. Finally, the obtained segmented structure data is input into the settlement measurement component for settlement distance analysis, and high-precision settlement distance results are output, providing key support for the formation of settlement data pairs and the construction of three-dimensional settlement curves.

[0032] Furthermore, settlement distance analysis includes:

[0033] Based on the structured subspace, factor feature identification is performed on the segmented structure data in parallel to determine the three-dimensional data sequence; for the three-dimensional data sequence, factor coupling analysis based on a first implicit causal relationship is performed on the interference data sequence and the motion data sequence to determine the first motion error; factor coupling analysis based on a second implicit causal relationship is performed on the motion data sequence and the settlement data sequence to determine the second settlement data; based on the first motion error, the second settlement data is compensated to determine the settlement data pair, wherein the settlement data pair is a pair of position mapping points before and after settlement.

[0034] Optionally, after receiving the segmented structure data, the settlement measurement component synchronizes this data to its internal structured subspace. The structured subspace then performs factor feature identification on the segmented structure data. Feature parameters corresponding to settlement factors, motion factors, and interference factors are extracted from the settlement subspace, motion subspace, and interference subspace, respectively. These parameters are then organized into a set of three-dimensional data sequences based on time series and spatial coordinates. This sequence reflects the characteristic changes of the same structure area under different perspectives and motion modes in actual motion, providing a foundation for subsequent causal coupling analysis. Subsequently, factor coupling analysis based on the first implicit causal relationship is performed on the interference data sequence and motion data sequence within the three-dimensional data sequence. That is, the degree of influence of interference factors under motion change conditions is identified through the implicit causal model corresponding to the first implicit causal relationship, thereby calculating the first motion error. This first motion error reflects the offset effect of external factors such as detection component vibration and illumination changes on data features under different perspectives and motion states. Simultaneously, factor coupling analysis based on a second implicit causal relationship is performed on the motion data sequence and the settlement data sequence. Specifically, the second settlement data of the structure under different motion perspectives is calculated using the implicit causal model corresponding to the second implicit causal relationship. This second settlement data represents the actual settlement deformation of the structure after eliminating some motion effects. Finally, based on the obtained first motion error, dynamic compensation is performed on the second settlement data. The compensation process involves multiplying the first motion error by a preset weight and then subtracting this product from the second settlement data. This corrects the residual deviation caused by motion coupling, resulting in a high-precision, corrected settlement result. The compensated settlement results are then organized into settlement data pairs, i.e., pairs of spatial coordinate mapping points at the same location before and after settlement. These settlement data pairs directly reflect the actual settlement distance and direction of the structure and are the core foundational data for subsequent three-dimensional distance measurement calculations and settlement curve generation.

[0035] Using a 3D projector, the settlement data is written in two dimensions and mapped in three dimensions based on the directional view projection plane. Three-dimensional distance measurement is performed to determine the three-dimensional settlement data, which is then updated to the settlement curve and displayed on the interface.

[0036] In one embodiment, after determining the settlement data pairs, a three-dimensional spatial coordinate system corresponding to the actual scene is established based on the attitude parameters of the detection platform and the geometric range of the monitoring area of ​​the structure, and a three-dimensional projector is constructed accordingly. Subsequently, the obtained settlement data pairs are written into the three-dimensional projector, performing two-dimensional writing and three-dimensional mapping based on the directional view projection plane. This ensures that the relative displacement information of each point maintains continuity and proportional consistency on the projection plane, thereby restoring the two-dimensional projection coordinates to three-dimensional spatial coordinates, forming three-dimensional settlement data for each monitoring point. This three-dimensional settlement data not only includes settlement depth but also reflects horizontal displacement and tilt trends, comprehensively describing the true settlement state of the structure. Finally, the three-dimensional settlement data is imported, and settlement curves are automatically generated based on the time series. These curves are then drawn and updated using a standard coordinate system, realizing the dynamic evolution record of settlement over time. The updated settlement curves and spatial settlement distribution are visualized through the interface display unit in the form of three-dimensional graphics, isosurfaces, or animations, allowing monitoring personnel to intuitively view the settlement change trend and abnormal locations of the structure, thus providing accurate and visual ranging results for engineering safety assessment and structural health monitoring.

[0037] Furthermore, the settlement data pair is subjected to two-dimensional writing and three-dimensional mapping based on the directional view projection plane, and three-dimensional ranging is performed, including:

[0038] For the structure area, a three-dimensional space is constructed. A regional mapping space is built by subtracting the two-dimensional projection space from the three-dimensional space. The projection perspective of the two-dimensional projection space is dynamic. A three-dimensional projector is constructed based on the regional mapping space to perform three-dimensional distance calculation based on the settlement data pairs and determine the three-dimensional settlement data.

[0039] Optionally, based on the structure area data collected by the detection platform, including depth information, viewing angle parameters, and spatial boundary information, a three-dimensional coordinate space corresponding to the actual scene is first established. This three-dimensional coordinate space uses the bottom of the structure or a stable reference point as the origin, defining X, Y, and Z three-dimensional spatial axes to represent the spatial position, direction, and height of the structure surface. Subsequently, based on the imaging parameters of the detection component, such as focal length, principal point position, attitude angle, and current viewing angle, the position and orientation of the camera coordinate system in three-dimensional space are determined. Then, with the optical center of the detection component as the projection center, a projection plane parallel to the imaging plane is established in this attitude direction, the position of which is determined by the focal length. Afterward, by calculating the perspective projection of the three-dimensional spatial point onto this projection plane, the coordinates of the spatial point are converted into corresponding two-dimensional pixel coordinates, thus forming a two-dimensional projection space. Since the mobile detection platform continuously changes position and angle during measurement, the system calculates and updates the normal vector and viewing direction of the projection plane in real time, making the projection angle of the two-dimensional projection space dynamically variable. This enables a stable mapping relationship at different detection positions, ensuring geometric consistency between frames of data. Subsequently, a dynamic homography spatial mapping function based on a perspective projection model is used to establish a correspondence between 3D coordinate points and 2D projected coordinates, forming a regional mapping space between 3D space and 2D projected space. Based on this regional mapping space, a 3D projector is constructed. This 3D projector can automatically calculate the perspective transformation between 3D coordinates and 2D pixel coordinates according to the attitude parameters and calibration matrix of the detection components, realizing unified projection and inversion between different spatial coordinate systems. Then, the settlement data pairs are input into the 3D projector, and the data is restored from 2D projected coordinates to 3D spatial coordinates through the mapping relationship. For each pair of points, the Euclidean distance and direction vector in the 3D coordinate system before and after settlement are calculated to obtain the 3D settlement displacement of the measuring point. Finally, the 3D settlement results of all measuring points are aggregated to generate a comprehensive 3D settlement dataset. This 3D settlement dataset not only contains the settlement depth of each monitoring point, but also retains lateral displacement and spatial distribution information, providing a complete and accurate 3D ranging basis for subsequent settlement curve plotting, anomaly identification, and structure stability assessment.

[0040] Furthermore, performing three-dimensional distance calculations based on the settlement data pairs includes:

[0041] Based on the motion data sequence in the three-dimensional data sequence, a two-dimensional projection viewpoint is determined; based on the two-dimensional projection viewpoint, a directional viewpoint projection is performed on the three-dimensional space to determine the target projection space; the settlement data is written into the target projection space, and a spatial mapping and three-dimensional transformation calculation based on the three-dimensional space and the target projection space is performed to determine the three-dimensional settlement data.

[0042] Optionally, to overcome spatial misalignment and ranging errors caused by the lack of depth information in 2D image measurement, the motion data sequence in the ternary data sequence is transformed into a rotation matrix and translation vector using Euler angles or quaternions. Then, the normal direction of the projection plane is calculated based on the pinhole camera model, and this direction is used as the 2D projection viewpoint, i.e., the observation direction of the camera relative to the 3D space of the structure and the orientation of the projection plane during acquisition. Subsequently, based on the determined 2D projection viewpoint, a projection plane corresponding to the current camera viewpoint is selected in 3D space, ensuring its normal vector aligns with the camera's optical axis. The 3D spatial points are then mapped to the 2D plane using a perspective projection matrix or homography mapping function, forming a target projection space. This target projection space maintains the same geometric relationship as the actual acquired image and serves as an intermediate bridge between 2D settlement information and 3D space reconstruction. Afterward, the acquired settlement data pairs are written into the target projection space. Using the established mapping relationship between the 3D space and the target projection space, the 2D plane coordinates are restored to 3D coordinates through back projection and coordinate transformation, achieving the accurate positioning of the settlement points in 3D space. Finally, the position of each pair of settlement points in three-dimensional space is transformed and calculated, that is, the Euclidean distance and direction vector of the corresponding points before and after settlement are calculated, so as to obtain the real three-dimensional settlement data at that position, providing a reliable data foundation for subsequent settlement curve updates and visualization.

[0043] Furthermore, updating the settlement curve and displaying it on the interface includes:

[0044] A standard coordinate system based on the structure area is constructed; a first coordinate node is generated based on the three-dimensional settlement data and written into the standard coordinate system as a settlement curve; the settlement curve is updated with time-series settlement measurements and visualized on a visual display interface.

[0045] Preferably, a unified spatial reference coordinate system is first established based on the actual spatial location of the structure and the geometric range of the monitoring area. Typically, the stable base or invariant reference point of the structure is used as the origin, defining the X-axis as the horizontal direction, the Y-axis as the longitudinal extension direction, and the Z-axis as the vertical height direction. This standard coordinate system serves as a unified reference framework for all settlement measurement data, ensuring spatial consistency and comparability of data acquired at different measurement times and from different perspectives. Subsequently, the three-dimensional settlement data obtained through three-dimensional projection and spatial calculation is imported into the standard coordinate system. For each monitoring point, its spatial location and corresponding settlement displacement in the three-dimensional coordinate system are extracted, generating the first coordinate node. Then, based on the spatial distribution and sequence relationship of the first coordinate nodes, interpolation or fitting algorithms are used to generate a continuous settlement curve. This settlement curve reflects the overall settlement morphology of the structure's surface or key parts in the standard coordinate system, providing an intuitive data basis for subsequent trend analysis. As monitoring continues, the system collects new structure settlement data at different time intervals. Each time new data is generated, it is compared and merged with historical settlement curves to update the data. Through time-series overlay, a set of settlement curves reflecting the structure's changes over time is formed. This process dynamically reflects settlement rate, settlement trend, and abnormal displacement points, enabling continuous tracking of structural deformation. Finally, the updated settlement curves are displayed using 3D rendering, contour lines, or cross-sectional views. The interface provides multi-view interactive browsing, allowing monitoring personnel to view the settlement depth, rate of change, and spatial distribution trend at different locations in real time. When abnormal nodes or sudden local settlement changes occur, they can be highlighted or alarms can be displayed on the interface, thus achieving an intuitive, dynamic, and intelligent display of the structure's settlement process.

[0046] Furthermore, after updating to the settlement curve and displaying it on the interface, the following steps are included:

[0047] Based on the construction progress of the underground space, the expected settlement curve is determined; by comparing the time-series mapping of the expected settlement curve with the actual settlement curve, abnormal settlement nodes are located; based on the abnormal settlement nodes, abnormal settlement alarms and feedback monitoring of underground space construction are executed.

[0048] Preferably, firstly, based on the construction progress and technological planning of the underground space, combined with historical settlement data and engineering geological conditions, an expected settlement curve is established. This expected settlement curve reflects the ideal settlement trend of the structure or surrounding soil during the construction process and will serve as a reference standard for subsequent monitoring, helping to identify abnormal situations that deviate from the normal settlement trajectory during construction. Then, the real-time constructed settlement curve and the expected settlement curve are compared in a time-series mapping, that is, the actual settlement data and the expected settlement data are compared in chronological order to ensure consistency between the two in the time dimension. Through comparison, the difference between the actual settlement trend and the expected settlement trend can be assessed. When the rate of change or displacement amplitude of the actual settlement data significantly exceeds the allowable range of the expected settlement, abnormal settlement nodes will be automatically marked. Subsequently, based on the identified abnormal settlement nodes, an abnormal settlement alarm mechanism is triggered, immediately notifying relevant personnel through the monitoring platform, mobile devices, or email. The alarm content typically includes the abnormal settlement location, settlement amplitude, and deviation value corresponding to the abnormal settlement node. Furthermore, based on alarm information and a pre-set reference solution library, the system automatically generates feedback and monitoring measures for abnormal settlement. These measures may include suggestions such as reinforcing the area, adjusting construction plans, and adding monitoring points. Related monitoring reports are also generated for the construction company's reference. This monitoring and early warning mechanism not only enables efficient settlement data analysis and anomaly detection but also strengthens dynamic risk management in underground space construction, improving project safety and construction efficiency.

[0049] In summary, the embodiments of this application have at least the following technical effects:

[0050] First, a mobile detection platform drives the detection component to collect data on the structure area, determining the structure data. Next, a settlement measurement component is constructed using spatial decomposition of the structure based on settlement, motion, and disturbance dimensions, and analysis based on implicit causal relationships. Then, by segmenting the structure data into a background domain, a structured subspace decomposition feature recognition based on the settlement measurement component and a coupled analysis based on implicit causal relationships are performed to determine settlement data pairs, where each pair represents the settlement distance at any location. Finally, using a 3D projector, the settlement data pairs are written in 2D and mapped in 3D based on a directional viewpoint projection plane to perform 3D ranging, determining the 3D settlement data, updating the settlement curve, and displaying it on the interface. This method solves the technical problems of insufficient accuracy and difficulty in effectively correcting spatial displacement ranging errors caused by attitude changes and environmental interference in existing structure settlement measurements. It achieves high-precision non-contact ranging and dynamic compensation for structure settlement displacement, improving spatial ranging stability and environmental adaptability.

[0051] Example 2 is based on the same inventive concept as the visual measurement method for structure settlement based on deep learning compensation in the previous examples, such as... Figure 2 As shown, this application provides a visual measurement system for structure settlement based on deep learning compensation. The system includes:

[0052] Data acquisition module 11: Based on the mobile detection platform, the detection component is driven to collect data on the structure area and determine the structure data; Measurement component construction module 12: Based on the spatial decomposition of the structure based on settlement dimension, motion dimension and disturbance dimension and the analysis based on implicit causal relationship, a settlement measurement component is constructed; Feature analysis module 13: By segmenting the structure data into a background domain, the structured subspace decomposition feature recognition based on the settlement measurement component and the coupling analysis based on implicit causal relationship are performed to determine settlement data pairs, wherein a set of settlement data pairs represents the settlement distance at any location; 3D measurement module 14: Through a 3D projector, the settlement data pairs are written in 2D and mapped in 3D based on the directional view projection plane, 3D distance measurement is performed, 3D settlement data is determined, updated to the settlement curve and displayed on the interface.

[0053] Furthermore, the measurement component construction module 12 performs the following method:

[0054] A settlement factor is defined, wherein the settlement factor encodes the pure deformation of the structure in two-dimensional space; a motion factor is defined, wherein the motion factor encodes the six-degree-of-freedom rigid body motion of the detection component, defined by the homography transformation law of the moving spatial transformation; an interference factor is defined, wherein the interference factor encodes light and shadow changes and jitter interference, defined by extracting stable features from the background data, wherein the jitter interference is the numerical relationship between the jitter of the detection component and the displacement of the structure; based on the settlement factor, motion factor, and interference factor, the structure space is decomposed to generate a structured subspace.

[0055] Furthermore, the measurement component construction module 12 is used to perform the following methods:

[0056] A first implicit causal relationship is established with interference factors and motion factors as causes and changes in data characteristics as effects; a second implicit causal relationship is established with settlement factors and motion factors as causes and changes in data characteristics as effects; after decomposing the structure space, deep learning based on the first and second implicit causal relationships is performed as a settlement measurement component.

[0057] Furthermore, the feature analysis module 13 is used to perform the following methods:

[0058] The structure data is transmitted back and background domain fuzzy segmentation is performed to determine the segmented structure data based on the segmentation boundary calibration; the segmented structure data is transmitted to the settlement measurement component for settlement distance analysis.

[0059] Furthermore, the feature analysis module 13 is used to perform the following methods:

[0060] Based on the structured subspace, factor feature identification is performed on the segmented structure data in parallel to determine the three-dimensional data sequence; for the three-dimensional data sequence, factor coupling analysis based on a first implicit causal relationship is performed on the interference data sequence and the motion data sequence to determine the first motion error; factor coupling analysis based on a second implicit causal relationship is performed on the motion data sequence and the settlement data sequence to determine the second settlement data; based on the first motion error, the second settlement data is compensated to determine the settlement data pair, wherein the settlement data pair is a pair of position mapping points before and after settlement.

[0061] Furthermore, the three-dimensional measurement module 14 is used to perform the following methods:

[0062] For the structure area, a three-dimensional space is constructed. A regional mapping space is built by subtracting the two-dimensional projection space from the three-dimensional space. The projection perspective of the two-dimensional projection space is dynamic. A three-dimensional projector is constructed based on the regional mapping space to perform three-dimensional distance calculation based on the settlement data pairs and determine the three-dimensional settlement data.

[0063] Furthermore, the three-dimensional measurement module 14 is used to perform the following methods:

[0064] Based on the motion data sequence in the three-dimensional data sequence, a two-dimensional projection viewpoint is determined; based on the two-dimensional projection viewpoint, a directional viewpoint projection is performed on the three-dimensional space to determine the target projection space; the settlement data is written into the target projection space, and a spatial mapping and three-dimensional transformation calculation based on the three-dimensional space and the target projection space is performed to determine the three-dimensional settlement data.

[0065] Furthermore, the three-dimensional measurement module 14 is used to perform the following methods:

[0066] A standard coordinate system based on the structure area is constructed; a first coordinate node is generated based on the three-dimensional settlement data and written into the standard coordinate system as a settlement curve; the settlement curve is updated with time-series settlement measurements and visualized on a visual display interface.

[0067] Furthermore, the three-dimensional measurement module 14 is used to perform the following methods:

[0068] Based on the construction progress of the underground space, the expected settlement curve is determined; by comparing the time-series mapping of the expected settlement curve with the actual settlement curve, abnormal settlement nodes are located; based on the abnormal settlement nodes, abnormal settlement alarms and feedback monitoring of underground space construction are executed.

[0069] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for visual measurement of settlement of a structure based on deep learning compensation, characterized in that, The method comprises: According to the mobile detection platform, drive the detection component to collect the structure area, determine the structure data; With the structure space decomposition based on the settlement dimension, the motion dimension and the interference dimension and the analysis based on the implicit causal relationship, build the settlement measurement component; By background domain segmentation of the structure data, execute the structured subspace decomposition feature recognition based on the settlement measurement component and the coupling analysis based on the implicit causal relationship, determine the settlement data pair, wherein a set of settlement data pairs represent the settlement distance of any position; Through the three-dimensional projector, execute the two-dimensional writing under the directional visual angle projection plane and the three-dimensional mapping of the settlement data pair, carry out three-dimensional ranging, determine the three-dimensional settlement data, update to the settlement curve and perform interface display; Wherein, the structure space decomposition based on the settlement dimension, the motion dimension and the interference dimension comprises: Define the settlement factor, wherein the settlement factor encodes the pure deformation of the structure in two-dimensional space; Define the motion factor, wherein the motion factor encodes the six-degree-of-freedom rigid body motion of the detection component, and the mobile space transformation law is defined by homographic transformation; Define the interference factor, wherein the interference factor encodes the light and shadow change and the jitter interference, the interference factor is defined by extracting the stable features in the background data, and the jitter interference is the digital relationship between the detection component jitter and the structure displacement; According to the settlement factor, the motion factor and the interference factor, the structure space is decomposed to generate a structured subspace.

2. The deep learning compensation-based construction settlement visual measurement method according to claim 1, wherein, The analysis based on the implicit causal relationship builds the settlement measurement component, comprising: Take the interference factor and the motion factor as the cause, and the data feature change as the effect, to establish the first implicit causal relationship; Take the settlement factor and the motion factor as the cause, and the data feature change as the effect, to establish the second implicit causal relationship; After decomposing the structure space, execute the deep learning based on the first implicit causal relationship and the second implicit causal relationship as the settlement measurement component. 3.The deep learning compensation-based structure settlement visual measurement method of claim 1, wherein, By background domain segmentation of the structure data, execute the structured subspace decomposition feature recognition based on the settlement measurement component and the coupling analysis based on the implicit causal relationship, comprising: Return the structure data and perform background domain fuzzy segmentation to determine the segmented structure data based on the segmentation boundary calibration; Transfer the segmented structure data to the settlement measurement component for settlement ranging analysis. 4.The deep learning compensation-based structure settlement visual measurement method of claim 3, wherein, The settlement ranging analysis comprises: According to the structured subspace, perform factor feature recognition on the segmented structure data in parallel to determine a three-element data sequence; For the three-element data sequence, perform factor coupling analysis based on the first implicit causal relationship on the interference data sequence and the motion data sequence to determine the first motion error; Perform factor coupling analysis based on the second implicit causal relationship on the motion data sequence and the settlement data sequence to determine the second settlement data; According to the first motion error, compensate the second settlement data to determine the settlement data pair, wherein the settlement data pair is the position mapping point pair before and after settlement. 5.The deep learning compensation-based structure settlement visual measurement method of claim 4, wherein, The two-dimensional writing under the directional visual angle projection plane and the three-dimensional mapping of the settlement data pair for three-dimensional ranging comprises: For the structure area, a three-dimensional space is constructed, a region mapping space is built with the three-dimensional space-two-dimensional projection space, wherein the projection visual angle of the two-dimensional projection space is dynamic; Based on the region mapping space, a three-dimensional projector is constructed, and three-dimensional distance measurement based on the settlement data pair is performed to determine three-dimensional settlement data. 6.The deep learning compensation-based structure settlement visual measurement method of claim 5, wherein, Performing three-dimensional distance measurement based on the settlement data pair includes: According to the motion data sequence in the ternary data sequence, a two-dimensional projection visual angle is determined; According to the two-dimensional projection visual angle, a directional visual angle projection is performed on the three-dimensional space to determine a target projection space; The settlement data pair is written into the target projection space, and spatial mapping and three-dimensional conversion measurement based on the three-dimensional space and the target projection space are performed to determine three-dimensional settlement data. 7.The deep learning compensation-based structure settlement visual measurement method of claim 1, wherein, Updating to the settlement curve and performing interface display includes: Constructing a standard coordinate system based on the structure area; Generating a first coordinate node based on the three-dimensional settlement data and writing it into the standard coordinate system as a settlement curve; With time-series settlement measurement, the settlement curve is updated and visualized on the visual display interface. 8.The deep learning compensation-based structure settlement visual measurement method of claim 1, wherein, After updating to the settlement curve and performing interface display, it includes: According to the underground space construction process, an expected settlement curve is determined; By performing time-series mapping comparison between the expected settlement curve and the settlement curve, an abnormal settlement node is located; According to the abnormal settlement node, abnormal settlement warning and feedback supervision of underground space construction are performed.

9. A structure settlement visual measurement system based on deep learning compensation, characterized in that, The system for implementing the deep learning compensation-based structure settlement visual measurement method of any one of claims 1-8 includes: A data acquisition module: according to the mobile detection platform, the detection component drives the detection component to collect the structure area to determine the structure data; A measurement component construction module: based on the settlement dimension, motion dimension and interference dimension, the structure space is decomposed and analyzed based on the implicit causal relationship to construct the settlement measurement component; A feature analysis module: by background domain segmentation of the structure data, structured subspace decomposition feature recognition based on the settlement measurement component and coupling analysis based on the implicit causal relationship are performed to determine the settlement data pair, wherein a group of settlement data pairs represent the settlement distance of any position; A three-dimensional measurement module: by the three-dimensional projector, the settlement data pair is written into the two-dimensional under the directional visual angle projection plane and three-dimensional mapping, three-dimensional ranging is performed, three-dimensional settlement data is determined, and the settlement curve is updated and interface display is performed.

Images