Foundation pit support deformation real-time monitoring and early warning method based on oblique photography

Abstract

The invention discloses a real-time monitoring and early warning method for deformation of a foundation pit support based on oblique photography, and relates to the technical field of photogrammetry and deformation measurement, high-resolution images of a support structure are acquired in a multi-angle manner through a multi-stage route, and a high-density three-dimensional point cloud reference model is constructed after preprocessing; then periodically acquiring a monitoring point cloud, and realizing high-precision space alignment with the reference model through control point coarse registration and iterative nearest point fine registration; and finally, calculating the three-dimensional coordinate deviation of each point through point cloud matching, extracting the full-surface deformation, and generating a deformation cloud picture and a statistical report. According to the method, the technical spanning from discrete point monitoring to full-field continuous monitoring and from low-dimensional data to true three-dimensional vectorization deformation analysis is realized, and the comprehensiveness and accuracy of deformation monitoring are improved.

Description

Technical Field

[0001] This invention relates to the field of photogrammetry and deformation measurement technology, specifically to a method for real-time monitoring and early warning of foundation pit support deformation based on oblique photography. Background Technology

[0002] Currently, foundation pit measurement and monitoring still rely primarily on traditional methods, resulting in a heavy workload. With the increasing number of deep and large foundation pit projects in cities, higher demands are placed on the accuracy, coverage, and real-time performance of support structure deformation monitoring. In foundation pit safety monitoring projects, especially those located at important urban transportation hubs, the small construction area, the accumulation of equipment and materials, and the constant movement of personnel easily obscure and damage existing monitoring points. This compromises the stability of the results and increases labor intensity. Furthermore, traditional monitoring methods involve point-by-point measurements, resulting in a relatively small number of monitoring points. This makes it difficult to detect deformation in areas without monitoring points, and errors in control points can affect the continuity of results. Existing technologies mainly suffer from the following shortcomings:

[0003] The monitoring points are sparse, making it impossible to obtain the continuous deformation field of the support surface, which easily leads to blind spots;

[0004] Manual high-altitude operations are risky and cannot meet the needs of high-frequency, unmanned monitoring.

[0005] The coordinate benchmarks for measurements in different periods are not consistent, and manual adjustment is required. The registration error often reaches the centimeter level, which cannot meet the requirements of millimeter-level deformation analysis.

[0006] The data processing flow is lengthy, the transformation results are delayed, and there is a lack of real-time early warning capabilities.

[0007] The early warning indicators are too simplistic and fail to differentiate the importance of different components such as retaining piles, walls, and roof slabs, resulting in a high rate of false alarms and missed alarms.

[0008] In recent years, although some studies have introduced ground-based three-dimensional laser scanning or vertical aerial photography into foundation pit monitoring, due to limitations such as fixed viewing angle, insufficient resolution, or low image overlap, it is still difficult to obtain high-resolution all-round data of the support structure facade and roof in a single flight. Moreover, the accuracy and real-time performance of multi-phase point cloud registration still have bottlenecks. Summary of the Invention

[0009] The technical problems to be solved by this invention are incomplete acquisition of deformation information of support structures, large monitoring blind spots, and low monitoring accuracy. The purpose is to provide a real-time monitoring and early warning method for foundation pit support deformation based on oblique photography. By combining photogrammetry, point cloud processing, and deformation analysis, high-precision non-contact full-field monitoring of support structure deformation is achieved. Multi-level flight path and multi-angle oblique photography are used to collect data in one go, improving the completeness of data acquisition and reducing monitoring blind spots. When performing deformation monitoring, the comparison error is reduced and the alignment accuracy is ensured by unifying the point clouds of multiple periods to the same reference coordinate system. The deformation of massive points is intuitively rendered on a three-dimensional model, enabling global and visual rapid location of the maximum deformation area, identification of deformation distribution patterns, and abnormal deformation areas.

[0010] This invention is achieved through the following technical solution:

[0011] This invention provides a method for real-time monitoring and early warning of foundation pit support deformation based on oblique photography, comprising the following specific steps:

[0012] Construct high-resolution image data of the foundation pit support structure from multiple angles and along multiple flight paths;

[0013] The high-resolution image data of the multi-angle foundation pit support structure are preprocessed, and the preprocessed image package is output.

[0014] A high-density 3D point cloud benchmark model is constructed based on the preprocessed image package as an initial reference for deformation analysis;

[0015] Periodically acquire monitoring periodic point cloud data of the support structure, and perform spatial registration using a pre-established benchmark model. The registration includes coarse registration based on control points and fine registration based on the iterative nearest point algorithm.

[0016] Spatial matching is performed between the finely registered monitoring point cloud and the reference point cloud, and the three-dimensional coordinate deviation of each corresponding point is calculated. The deviation includes the offset in the X, Y and Z directions, and the deformation of each point of the support structure is extracted based on the offset.

[0017] Based on the extracted deformation, a deformation cloud map and an offset statistics table are generated.

[0018] Set deformation thresholds and multi-level alarm thresholds, assign corresponding weight coefficients according to different components or areas of the support structure, and monitor the offset based on the weight coefficients, deformation thresholds, and multi-level alarm thresholds.

[0019] Furthermore, the construction of high-resolution image data for the multi-level flight path and multi-angle acquisition of the foundation pit support structure specifically includes:

[0020] Construct high-resolution image data of the foundation pit support structure from multiple angles and along multiple flight paths;

[0021] The collected image data is uniformly numbered according to a preset numbering rule, which includes at least the collection date, flight path level, flight sequence number, camera angle, and serial number.

[0022] The numbered image data is then stored according to a predetermined storage structure.

[0023] Furthermore, the multi-level routes include top-level routes, middle-level routes, bottom-level routes, and near-field routes:

[0024] The top-level flight path is used to collect the overall shape of the foundation pit support structure and its relationship with the surrounding environment;

[0025] The middle-level flight path is used to collect data on the surface condition of the support facade area and the support system.

[0026] The bottom flight path is used to collect data on the stability and precipitation effect of the bottom area of ​​the foundation pit.

[0027] The close-up flight path is used to collect detailed features of key nodes or suspected defect areas.

[0028] Furthermore, when performing multi-level flight path acquisition, it also includes detecting the overlap of images acquired from multiple flight paths;

[0029] Using the collinearity condition equation, the four corner points of each image are projected backwards onto the average flight altitude plane to obtain a quadrilateral in ground coordinates.

[0030] A pairing strategy is defined to perform image pairing for the entire target detection region. The pairing strategy includes: pairing two images before and after each other in the same flight zone, and pairing the nearest images in adjacent flight zones.

[0031] Based on the paired images, calculate the area of ​​the intersection polygon of their ground quadrilaterals;

[0032] Calculate the image overlap by combining the ground coverage area of ​​a single image;

[0033] If the overlap is less than the set threshold, the flight path is adjusted according to the overlap until the overlap of images acquired from multiple flight paths reaches the set threshold.

[0034] Furthermore, the preprocessing of the high-resolution image data of the acquired multi-angle foundation pit support structure specifically includes:

[0035] Based on preset rejection criteria, the numbered image data is automatically filtered and manually reviewed to remove unqualified images. The rejection criteria include at least motion blur, abnormal exposure, severe occlusion, and focus failure.

[0036] In the 3D calibration field, the camera of the acquisition device is calibrated to obtain the camera's orientation elements and lens distortion parameters;

[0037] The distortion parameters are used to perform batch distortion removal processing on the filtered image data;

[0038] Color equalization is performed on the distortion-reduced image to make the color difference between different images in the same area less than a set threshold.

[0039] Construct a multi-level image pyramid, wherein the pyramid includes at least the original resolution level, the 1 / 2 resolution level, the 1 / 4 resolution level, and the 1 / 8 resolution level;

[0040] The preprocessed image data is subjected to quality checks, and the checks include at least the number of images, the degree of distortion correction, color consistency, pyramid integrity, and metadata accuracy.

[0041] After passing the inspection, the output is a preprocessed image package containing the corrected image, pyramid image, camera parameter file, quality report, and data list.

[0042] Furthermore, the construction of a high-density 3D point cloud benchmark model based on the preprocessed image package, as an initial reference for deformation analysis, specifically includes:

[0043] The image data of the preprocessed image package is subjected to bundle adjustment, and each image is abstracted as a beam of light from the projection center to the image point.

[0044] Adjustment is performed using the collinearity condition equation as a constraint to obtain the exterior orientation elements of each image and the ground coordinates corresponding to all connection points;

[0045] Projecting the images onto each image based on the ground coordinates of all connection points, the approximate depth range of each stereo pair is obtained.

[0046] Rotational epipolar resampling is performed on each stereo image pair to obtain horizontal epipolar image pairs;

[0047] A semi-global matching algorithm is used to perform a one-dimensional search along the epipolar direction of the horizontal epipolar image pair to calculate the disparity map;

[0048] The disparity map of each stereo image pair is back-projected into the object space, and a local depth map is obtained by forward intersection.

[0049] Perform a multi-view consistency check on each pixel and output the checked depth map;

[0050] The depth map is converted into a 3D point cloud, all point clouds generated from the images are merged, and noise points are removed using a statistical outlier removal algorithm or radius filtering, resulting in a final high-density 3D point cloud benchmark model.

[0051] Furthermore, the coarse registration based on control points and the fine registration based on the iterative nearest point algorithm specifically include:

[0052] Generate a depth image of the current period's image data using a multi-view stereo vision method;

[0053] An initial dense point cloud model is reconstructed based on the depth image;

[0054] Introduce calibration points into the point cloud. These calibration points are the points with the same name corresponding to the same location in different point clouds in the real world.

[0055] Using the known spatial information of the calibration points, the spatial information of each point cloud is unified into the basic coordinate system to achieve preliminary point cloud registration based on point-to-point matching;

[0056] A modified iterative nearest-point algorithm is used to perform secondary fine registration on the point cloud after the initial point cloud registration;

[0057] The fine registration includes:

[0058] Given a monitoring point cloud and a reference point cloud, the optimal rigid body transformation matrix is ​​solved iteratively to align the two point clouds.

[0059] Spatial matching is performed between the finely registered monitoring point cloud and the reference point cloud, and the three-dimensional coordinate deviation of each corresponding point is calculated. The deviation includes the offset in the X, Y and Z directions, and the deformation of each point of the support structure is extracted based on the offset.

[0060] Based on the extracted deformation, a deformation cloud map and an offset statistics table are generated.

[0061] Furthermore, the fine registration further includes:

[0062] At least four control points are set up in the monitoring area, and the corresponding control points are identified in the baseline model and the point cloud of the monitoring period, respectively.

[0063] Based on the control points, the rotation matrix and translation vector are calculated, and the point cloud of the monitoring period is transformed to the reference coordinate system;

[0064] The iterative nearest point algorithm is used to perform fine registration between the transformed monitoring point cloud and the benchmark model, minimizing the Euclidean distance error between the two point clouds.

[0065] Furthermore, the generation of deformation cloud maps and offset statistics tables based on the extracted deformation amounts specifically includes:

[0066] Identify deformation amount:

[0067] For retaining piles or retaining walls, analyze the convergence deviation in the X and Y directions to identify inward or outward deformation.

[0068] For the top or bottom slab, analyze the settlement or uplift deviation in the Z direction, and extract the maximum settlement and its distribution area;

[0069] A deformable cloud map is generated based on the offset, and the magnitude of the offset is represented by color coding in the deformable cloud map.

[0070] Output an offset statistics table, which includes the maximum offset, average offset, and standard deviation of offset for each region and component.

[0071] Furthermore, the alarm monitoring specifically includes:

[0072] Based on the current building structure status and engineering experience of the monitored support structure, deformation thresholds and multi-level alarm thresholds are set.

[0073] Assign corresponding weight coefficients according to the different components or areas of the support structure;

[0074] Based on the three-dimensional offset of each point and the weighting coefficient, a weighted deformation index reflecting the degree of overall or local deformation is obtained through weighted calculation.

[0075] The weighted deformation index is compared with a preset deformation threshold. When the deformation threshold is exceeded, an alarm of the corresponding level is automatically triggered.

[0076] The current period's point cloud data, after being verified and confirmed following the triggering of an alarm, will be updated to a new monitoring benchmark model for comparison and deviation calculation in subsequent periods.

[0077] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0078] Through multi-angle, high-resolution image acquisition and 3D reconstruction, high-density spatial information of the entire visible surface of the support structure can be obtained. Through multi-layer flight path acquisition, comprehensive monitoring of all parts such as the support facade, support beams, and cap beams is achieved without blind spots. For the first time, engineers can grasp the deformation field of the structure as a whole, which significantly improves the comprehensiveness and reliability of risk identification.

[0079] Through high-precision spatial registration and point cloud matching, the precise offset of each surface point in the three orthogonal directions XYZ can be directly calculated and output, that is, the complete three-dimensional displacement vector. This can accurately quantify the overall settlement and horizontal slip of the structure and clearly reveal the complex deformation mode.

[0080] Using drones or ground mobile platforms for non-contact image acquisition, operators do not need to make close contact with the monitoring target, which greatly ensures personnel safety and minimizes interference with the construction site. The data acquisition process is fast and flexible, and a single operation can cover a large foundation pit. It is suitable for high-frequency, periodic automated monitoring in complex, dangerous or busy construction conditions.

[0081] A two-level strategy is adopted, combining coarse registration based on stable control points and fine registration based on the iterative nearest point algorithm. Coarse registration first stably incorporates data from different periods into a unified engineering coordinate system, solving the problem of large-scale benchmark unification. Fine registration eliminates subtle system deviations caused by differences in acquisition angle and illumination, reduces human intervention, and ensures high repeatability accuracy for comparative analysis between data from different periods. Attached Figure Description

[0082] To more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly described below. It should be understood that the following drawings only show some embodiments of the present invention and should not be considered as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort. In the drawings:

[0083] Figure 1 This is the monitoring process in an embodiment of the present invention. Detailed Implementation

[0084] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments and accompanying drawings. The illustrative embodiments and descriptions of the present invention are only used to explain the present invention and are not intended to limit the present invention.

[0085] As one possible implementation method, such as Figure 1 As shown, this embodiment provides a real-time monitoring and early warning method for foundation pit support deformation based on oblique photography. High-resolution images of the support structure are acquired from multiple angles and via multiple flight paths. After preprocessing, a high-density three-dimensional point cloud benchmark model is constructed. Then, monitoring point clouds are periodically acquired, and high-precision spatial alignment with the benchmark model is achieved through coarse registration of control points and fine registration of the nearest point iteratively. Finally, the three-dimensional coordinate deviation of each point is calculated through point cloud matching, the deformation of the entire surface is extracted, and deformation cloud maps and statistical reports are generated. This embodiment achieves a technological leap from discrete point monitoring to continuous full-field monitoring, and from low-dimensional data to true three-dimensional vectorized deformation analysis, improving the comprehensiveness and accuracy of deformation monitoring.

[0086] The specific implementation method of this embodiment is as follows:

[0087] Step 1: Construct high-resolution image data of the foundation pit support structure from multiple angles and along multiple flight paths;

[0088] High-resolution image data of the foundation pit support structure is constructed using multi-level flight paths and multi-angle acquisition. The acquired image data is uniformly numbered according to a preset numbering rule, which includes at least the acquisition date, flight path level, sortie number, camera angle, and sequence number. The numbered image data is then stored according to a predetermined storage structure. This numbering rule, which includes information such as date, level, sortie number, and angle, ensures that each image has a clear spatiotemporal and logical label. When defects are discovered, their location (which level, which side, which sortie number) and acquisition time can be quickly pinpointed, facilitating accountability and change comparison.

[0089] The multi-level flight path consists of four levels: top-level, middle-level, bottom-level, and close-up. This four-level flight path design enables full-scale data acquisition from the macroscopic overall structure to the microscopic details, ensuring that every important part of the foundation pit (top, walls, bottom, and nodes) is systematically recorded. Especially for vertical structures such as support facades, adjusting the camera's acquisition angle (e.g., oblique photography) avoids the loss of lateral information caused by single vertical shooting, allowing for more complete capture of features such as cracks, deformation, and leakage.

[0090] The top-level flight path is used to collect data on the overall morphology of the foundation pit support structure and its relationship with the surrounding environment.

[0091] The mid-level flight path is used to collect data on the surface condition of the support facade area and the support system.

[0092] The bottom flight path is used to collect data on the stability of the bottom area of ​​the foundation pit and the effect of precipitation.

[0093] Close-up flight paths are used to collect detailed features of key nodes or suspected defect areas.

[0094] In some possible implementations, the acquisition parameters for the fourth-level flight path can be as shown in Table 1.

[0095]

[0096] When performing Level 4 flight path acquisition, the overlap of multiple flight path acquisition images is also detected.

[0097] Using the collinearity condition equation, the four corner points of each image are projected backwards onto the average flight altitude plane to obtain a quadrilateral in ground coordinates.

[0098] A pairing strategy is defined to perform image pairing for the entire target detection region. The pairing strategy includes: pairing two images before and after each other in the same flight zone, and pairing the nearest images in adjacent flight zones.

[0099] Based on the paired images, calculate the area of ​​the intersection polygon of their ground quadrilaterals;

[0100] Calculate the image overlap by combining the ground coverage area of ​​a single image;

[0101] Overlap threshold settings: Forward overlap (images before and after the same flight path) ≥ 80%, Lateral overlap ≥ 70%.

[0102] If the overlap is less than the set threshold, the flight path is adjusted according to the overlap until the overlap of images acquired from multiple flight paths reaches the set threshold.

[0103] Step 2: Preprocess the high-resolution image data of the multi-angle foundation pit support structure and output the preprocessed image package;

[0104] Based on preset rejection criteria, the numbered image data is automatically screened and manually reviewed to remove unqualified images. The rejection criteria include at least motion blur, abnormal exposure, severe occlusion, and focus failure.

[0105] In the 3D calibration field, the camera of the acquisition device is calibrated to obtain the camera's orientation elements and lens distortion parameters;

[0106] Batch distortion removal processing is performed on the filtered image data using distortion parameters;

[0107] Color equalization is performed on the distortion-reduced image to make the color difference between different images in the same area less than a set threshold.

[0108] Construct a multi-level image pyramid, which includes at least the original resolution level, the 1 / 2 resolution level, the 1 / 4 resolution level, and the 1 / 8 resolution level;

[0109] The preprocessed image data is subjected to quality checks, and the checks include at least the number of images, the degree of distortion correction, color consistency, pyramid integrity, and metadata accuracy.

[0110] After passing the inspection, the output is a preprocessed image package containing the corrected image, pyramid image, camera parameter file, quality report, and data list.

[0111] By combining automatic screening with manual review, 3D calibration, distortion correction, global color equalization, and multi-level pyramid processing, the original high-resolution image is quickly transformed into a clean, color-difference-free, multi-scale preprocessing package with strict quality certification, thereby improving the accuracy of subsequent deformation monitoring steps.

[0112] Step 3: Construct a high-density 3D point cloud benchmark model based on the preprocessed image package as the initial reference for deformation analysis;

[0113] A high-precision, high-density, and noise-free 3D point cloud model is constructed as the geometric benchmark for subsequent deformation monitoring. The accuracy of this benchmark model directly determines the reliability of the entire deformation analysis system.

[0114] Bundle network adjustment is performed on the image data of the preprocessed image package. Each image is abstracted as a ray of light from the projection center to the image point. Bundle network adjustment incorporates all images, all image feature points, and camera intrinsic parameters into a unified mathematical model. Using the collinearity condition equation (i.e., the object point, the camera projection center, and its image point on the image are collinear) as the core constraint, the optimal solution is obtained through least squares iterative solving, yielding the exterior orientation elements of each image and the ground coordinates corresponding to all connection points.

[0115] Projecting the images onto each image based on the ground coordinates of all connection points, the approximate depth range of each stereo pair is obtained.

[0116] Rotational epipolar resampling is performed on each stereo image pair to obtain horizontal epipolar image pairs. Within each stereo image pair, the original two-dimensional search—finding the corresponding point across the entire image—is transformed into a one-dimensional search along the epipolar lines. For a point in the left image, its corresponding point in the right image must lie on a specific straight line (epidial line). Using the relative orientation elements of the stereo image pairs, rotational epipolar resampling can be performed to generate horizontal epipolar image pairs. In the resampled image, the epipolar lines are precisely aligned with the image rows (or columns). This reduces computational complexity and improves the efficiency of dense matching.

[0117] A semi-global matching algorithm is used to perform a one-dimensional search along the epipolar direction of the horizontal epipolar image pair. That is, one-dimensional dynamic programming is performed on each pixel along multiple paths (usually 8 or 16 directions), the matching cost is aggregated, and then the disparity value that minimizes the aggregation cost is selected.

[0118] The disparity map of each stereo image pair is projected backwards into the object space. Using disparity, camera intrinsic and extrinsic parameters, and the principle of triangulation, each pixel in the disparity map is converted into a three-dimensional point in the object space, forming a depth map of a single image.

[0119] For each pixel, a multi-view consistency test is performed. For a disparity point P generated from images A and B, it is reprojected onto the third, fourth, and Nth images. If the pixel color / feature of the projected position of point P is consistent with the height of the original point on these additional images, then point P is considered reliable; otherwise, it may be a mismatch point and is discarded.

[0120] Based on the multi-view consistency test results, the depth map corresponding to each original image after removal is output;

[0121] The depth map is converted into a 3D point cloud, all point clouds generated from the images are merged, and noise points are removed using a statistical outlier removal algorithm or radius filtering, resulting in a final high-density 3D point cloud benchmark model.

[0122] Step 4: Periodically acquire the monitoring periodic point cloud data of the support structure, and perform spatial registration using the pre-established benchmark model. The registration includes coarse registration based on control points and fine registration based on the iterative nearest point algorithm.

[0123] To ensure the consistency and comparability of data sources, the same processing flow minimizes systematic errors introduced by algorithm differences, ensuring that point clouds from different periods maintain consistent density and accuracy characteristics, laying the foundation for subsequent accurate comparisons. In each subsequent monitoring period, using the same interior and exterior orientation elements as when establishing the baseline model, the support structure is photographed from the same set of fixed camera stations. The same dense matching and depth map reconstruction process as in step three is employed to generate the initial dense point cloud for the current period.

[0124] Introducing calibration points into the point cloud: At least four high-precision, high-contrast artificial markers are placed on the surface of the support structure or in the surrounding stable area. These calibration points are the corresponding points of the same location in the real world in different point clouds. The precise three-dimensional coordinates of these markers were accurately determined when the baseline model was established.

[0125] In the reference point cloud and the monitoring point cloud of each cycle, these identical marker points are accurately identified automatically or semi-automatically (based on patterns, spatial relationships, etc.). Using the known spatial information of the calibration points, the spatial information of each point cloud is unified into the basic coordinate system to achieve preliminary coarse registration of the point cloud based on point-to-point matching. Based on the coarse registration based on control points, a modified iterative nearest point algorithm is used to perform secondary fine registration of the point cloud after the preliminary registration.

[0126] Fine registration includes: given a monitoring point cloud and a reference point cloud, for each point in the monitoring point cloud, finding the point in the reference point cloud with the closest Euclidean distance to form a corresponding point pair; for the effective point pairs, using singular value decomposition (SVD) or quaternion methods, calculating the optimal rigid body transformation matrix (rotation matrix R and translation vector T) to minimize the sum of squared distances between corresponding point pairs. Even if the two point clouds are aligned, ICP can still maximize the alignment of the undeformed region and the stable region when the structure has undergone slight deformation, thereby separating the true deformation signal.

[0127] After fine registration, the two point clouds are in the same coordinate system and are optimally aligned. Spatial matching is performed between the finely registered monitoring point cloud and the reference point cloud, and the three-dimensional coordinate deviation of each corresponding point is calculated. The deviation includes the offset in the X, Y and Z directions. Based on the offset, the deformation of each point of the support structure is extracted. Based on the extracted deformation, a deformation cloud map and an offset statistics table are generated.

[0128] Fine registration further includes:

[0129] At least four control points are set up in the monitoring area, and the corresponding control points are identified in the baseline model and the point cloud of the monitoring period, respectively.

[0130] Based on the control points, the rotation matrix and translation vector are calculated to transform the point cloud of the monitoring period to the reference coordinate system;

[0131] The iterative nearest point algorithm is used to perform fine registration between the transformed monitoring point cloud and the benchmark model, minimizing the Euclidean distance error between the two point clouds.

[0132] Spatial matching is performed between the finely registered monitoring point cloud and the reference point cloud. The three-dimensional coordinate deviation of each corresponding point is calculated, including the offset in the X, Y, and Z directions. Based on the offset, the deformation of each point of the support structure is extracted.

[0133] In the above, periodic point cloud generation ensures the temporal consistency of the data source, making data from different periods comparable, which is a prerequisite for reliable change detection. The introduction of control points and coarse registration solves the problems of absolute position and scale, providing high-quality initial values ​​for fine registration, significantly improving the efficiency and success rate of registration, and suppressing error accumulation. Improved ICP fine registration achieves ultra-high precision spatial alignment between point clouds. Coordinate deviation calculation enables fully automatic, full-field, and quantitative measurement of deformation.

[0134] Step 5: Based on the extracted deformation, generate a deformation cloud map and an offset statistics table;

[0135] Identify deformation amount:

[0136] For retaining piles or retaining walls, analyze the convergence deviation in the X and Y directions (horizontal displacement in the plane perpendicular to the pile / wall axis), calculate the horizontal displacement of all points in the area to identify inward squeezing or outward expansion deformation, directly assess the lateral stability and earth pressure resistance of the support system, and determine whether the support structure is unstable.

[0137] For the top or bottom slab, analyze the settlement or uplift deviation (i.e., vertical displacement) in the Z direction, calculate the Z-direction offset of all points in the area, with a decrease in Z value indicating settlement and an increase in Z value indicating uplift. Statistically calculate the maximum settlement / uplift and the average settlement rate, delineate the main distribution areas of settlement / uplift, and assess the impact of the excavation on the structure itself and the surrounding ground surface.

[0138] Deformation cloud maps are generated based on offsets. The magnitude of the offset can be represented by a continuously gradient rainbow color scheme or a red-yellow-green monochrome gradient color scheme.

[0139] Output an offset statistics table, which includes the maximum offset, average offset, and standard deviation of offset for each region and component.

[0140] Step 6: Set deformation threshold and multi-level alarm threshold. Assign corresponding weight coefficients according to different components or areas of the support structure. Monitor the offset based on the weight coefficients, deformation threshold and multi-level alarm threshold.

[0141] Based on the allowable deformation values ​​(such as horizontal displacement and settlement control values) of various support structures (piles, diaphragm walls, anchors, etc.) in national standards and industry specifications (such as the "Technical Specification for Monitoring of Building Foundation Pit Engineering"), the stress and deformation of the structure under different loads are simulated through structural mechanics simulation, the theoretical safety limit value is calculated, and then the deformation threshold and multi-level alarm threshold are set.

[0142] Assign corresponding weight coefficients according to the different components or areas of the support structure;

[0143] The weighting coefficients are allocated based on the following criteria: Components that play a crucial role in overall stability (such as corner supports and support points) have higher weights. According to mechanical analysis, areas of stress concentration or maximum bending moment (such as the middle of the pile or the end of the support) have higher weights. Areas that have historically experienced anomalies or have been reinforced have higher weights. Quantification methods: The Analytic Hierarchy Process (AHP) or expert scoring can be used to convert the above qualitative criteria into specific weight values ​​(e.g., the sum of weights is 1).

[0144] Based on the three-dimensional offset and weight coefficient of each point, a weighted deformation index reflecting the degree of overall or local deformation is obtained through weighted calculation. The weighted deformation index is compared with a preset deformation threshold. When the deformation threshold is exceeded, an alarm of the corresponding level is automatically triggered. The current period point cloud data after the alarm is triggered and verified is updated into a new monitoring benchmark model for comparison and deviation calculation in subsequent periods.

[0145] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A tilt photography-based real-time monitoring and early warning method for deformation of a foundation pit support, characterized in that, The method comprises the following specific steps: Constructing high-resolution image data of the foundation pit support structure collected by multi-level flight lines and multi-angle; Pretreating the collected high-resolution image data of the foundation pit support structure of multi-angle, and outputting pretreated image packages; Constructing a high-density three-dimensional point cloud reference model based on the pretreated image packages, as an initial reference for deformation analysis; Periodically acquiring monitoring period point cloud data of the support structure, and performing spatial registration by using the previously established reference model, wherein the registration comprises coarse registration based on control points and fine registration based on an iterative closest point algorithm; Performing spatial matching on the monitoring point cloud and the reference point cloud after fine registration, calculating three-dimensional coordinate deviations of corresponding points, wherein the deviations comprise offset amounts in X, Y and Z directions, and extracting deformation amounts of points of the support structure based on the offset amounts; Generating a deformation cloud chart and an offset amount statistical table based on the extracted deformation amounts; Setting a deformation threshold and multi-level alarm thresholds, assigning corresponding weight coefficients to different components or regions of the support structure, and performing alarm monitoring on the offset amounts based on the weight coefficients, the deformation threshold and the multi-level alarm thresholds.

2. The tilt photography-based real-time monitoring and early warning method for foundation pit support deformation according to claim 1, characterized in that, The method of constructing high-resolution image data of the foundation pit support structure collected by multi-level flight lines and multi-angle specifically comprises: Constructing high-resolution image data of the foundation pit support structure collected by multi-level flight lines and multi-angle; Uniformly numbering the collected image data according to a preset numbering rule, wherein the numbering rule at least comprises collection date, flight line level, flight order number, camera angle and serial number; And storing the numbered image data according to a predetermined storage structure.

3. The tilt photography-based real-time monitoring and early warning method for foundation pit support deformation according to claim 2, characterized in that, The multi-level flight lines comprise top-level flight lines, middle-level flight lines, bottom-level flight lines and close-range flight lines: The top-level flight lines are used to collect the overall shape of the foundation pit support structure and the relationship with the surrounding environment; The middle-level flight lines are used to collect the surface conditions of the support facade region and the support system; The bottom-level flight lines are used to collect the stability of the foundation pit bottom region and the precipitation effect; The close-range flight lines are used to collect the detailed features of key nodes or suspected defect regions.

4. The tilt photography-based real-time monitoring and early warning method for foundation pit support deformation according to claim 2, characterized in that, When performing multi-level flight line collection, it also comprises detecting the overlap degree of the multi-flight line collected images; Using a collinear condition equation, the four corner points of each image are inversely projected onto the average flight height plane to obtain a quadrilateral on the ground; Defining a pairing strategy to pair images in the entire target detection region, wherein the pairing strategy comprises pairing the front and rear two images of the same flight strip and pairing the nearest images of adjacent flight strips; Based on the paired images, the intersection polygon area of the ground quadrilaterals is calculated; Combining the ground coverage area of a single image, the image overlap degree is calculated; If the overlap degree is less than a set threshold, the flight lines are adjusted according to the overlap degree until the image overlap degrees collected by the multiple flight directions reach the set threshold.

5. The tilt photography-based real-time monitoring and early warning method for foundation pit support deformation according to claim 1, characterized in that, The pretreatment of the collected high-resolution image data of the foundation pit support structure of multi-angle specifically comprises: Based on a preset rejection standard, the numbered image data is automatically screened and manually reviewed, and unqualified images are rejected, wherein the rejection standard at least comprises motion blur, exposure abnormality, severe occlusion and focus failure; In a three-dimensional calibration field, the camera of the collection device is calibrated to obtain camera interior orientation elements and lens distortion parameters; Batch distortion removal processing is performed on the screened image data by using the distortion parameters; Color equalization is performed on the de-distorted images, so that color difference between different images of the same region is less than a set threshold value; A multi-level image pyramid is constructed, which includes at least a raw resolution level, a 1 / 2 resolution level, a 1 / 4 resolution level and a 1 / 8 resolution level; Quality inspection is performed on the pre-processed image data, and the inspection items include at least image quantity, distortion correction completion, color consistency, pyramid integrity and metadata accuracy; After the inspection, a pre-processed image package is output, which includes corrected images, pyramid images, camera parameter files, quality reports and data lists.

6. The tilt photography-based real-time monitoring and early warning method for foundation pit support deformation according to claim 1, characterized in that, The pre-processed image package is used to construct a high-density three-dimensional point cloud reference model as an initial reference for deformation analysis, which specifically includes: The image data of the pre-processed image package is subjected to bundle block adjustment, and each image is abstracted as a bundle of light rays from the projection center to the image point; Adjustment is performed with the collinearity condition equation as a constraint to obtain the exterior orientation elements of each image and the ground coordinates corresponding to all connected points; According to the ground coordinates corresponding to all connected points, projection is performed on each image to obtain the approximate depth range of each stereo image pair; Rotary epipolar resampling is performed on each stereo image pair to obtain a pair of horizontal epipolar images; A half-global matching algorithm is used to perform one-dimensional search along the epipolar direction of the pair of horizontal epipolar images to calculate a disparity map; The disparity map of each stereo image pair is inversely projected into the object space to obtain a local depth map through forward intersection; Multi-view consistency test is performed on each pixel to output a verified depth map; The depth map is converted into a three-dimensional point cloud, all image-generated point clouds are merged, and a statistical outlier removal algorithm or radius filtering is used to remove noise points, and finally a high-density three-dimensional point cloud reference model is output.

7. The tilt photography-based real-time monitoring and early warning method for foundation pit support deformation according to claim 1, characterized in that, The coarse registration based on control points and the fine registration based on the iterative closest point algorithm specifically include: A multi-view stereo vision method is used to generate a depth image of the current period image data; An initial dense point cloud model is reconstructed according to the depth image; Calibration points are introduced into the point cloud, which are the same-named points corresponding to the same position in different point clouds in the real world; The known spatial information of the calibration points is used to unify the spatial information of each point cloud to a base coordinate system to realize preliminary point cloud registration based on point-to-point matching; A modified iterative closest point algorithm is used to perform secondary fine registration on the point cloud after preliminary point cloud registration. The fine registration includes: Given a monitoring point cloud and a reference point cloud, the best rigid transformation matrix is solved iteratively to align the two point clouds; The monitoring point cloud and the reference point cloud after fine registration are subjected to spatial matching to calculate the three-dimensional coordinate deviation of each corresponding point, which includes the offset in X, Y and Z directions, and the deformation of each point of the support structure is extracted based on the offset; Based on the extracted deformation, a deformation cloud chart and an offset statistical table are generated.

8. The tilt photography-based real-time monitoring and early warning method for foundation pit support deformation according to claim 7, characterized in that, The fine registration further includes: At least four control points are arranged in the monitoring area, and corresponding control points are identified in the reference model and the monitoring period point cloud; A rotation matrix and a translation vector are calculated based on the control points to convert the monitoring period point cloud to the reference coordinate system; The converted monitoring point cloud and the reference model are finely registered by using an iterative closest point algorithm to minimize the Euclidean distance error between the two point clouds.

9. The tilt photography-based real-time monitoring and early warning method for foundation pit support deformation according to claim 1, characterized in that, The deformation cloud map and the offset statistical table are generated based on the extracted deformation, specifically including: Identifying the deformation: For support piles or support walls, analyze the X and Y direction convergence deviation, and identify the internal extrusion or external expansion deformation; For the roof or floor, analyze the Z direction settlement or uplift deviation, extract the maximum settlement and distribution area; Based on the offset, generate a deformation cloud map, which represents the offset size by color coding; Output the offset statistical table, which includes the maximum offset, average offset, and offset standard deviation of each region and component.

10. The tilt photography-based real-time monitoring and early warning method for foundation pit support deformation according to claim 1, characterized in that, The alarm monitoring specifically includes: Based on the building structure status of the monitored support structure and engineering experience, set the deformation threshold and multi-level alarm threshold; According to the different components or regions of the support structure, assign corresponding weight coefficients; Based on the three-dimensional offset of each point and the weight coefficient, a weighted deformation index reflecting the overall or local deformation degree is calculated by weighting; Compare the weighted deformation index with the preset deformation threshold, and when it exceeds the deformation threshold, automatically trigger the corresponding level of alarm; The current period point cloud data verified and confirmed after triggering the alarm is updated as a new monitoring reference model for comparison and deviation calculation in the subsequent period.

Images