A visual terminal array-based foundation pit deformation real-time perception and intelligent inference method

Abstract

The application discloses a kind of based on visual terminal array's foundation pit deformation real-time perception and intelligent inference method, adopts multi-mode visual terminal array to realize foundation pit and surrounding environment image acquisition, in combination with the process of image processing, data extraction and calibration correction, extract foundation pit deformation data sequence, construct foundation pit global continuous deformation field;The physical constraint of soft soil feature is embedded in LSTM+Transformer model, and an intelligent inference framework is formed;In combination with the coupling deformation analysis of foundation pit and protective object, the data access channel of traditional mode measuring point is reserved, and the spatial field and time development trend of foundation pit deformation are comprehensively analyzed and judged.The advantage of the present application is suitable for urban renewal background soft soil site foundation pit, based on machine learning time series prediction model, realize the intelligent inference of the coupling deformation of foundation pit and surrounding protective object, predict the development of foundation pit deformation, give the opinion comprehensively to assist foundation pit safety decision, improve the intelligent monitoring and risk control level of foundation pit engineering.

Description

Technical Field

[0001] This invention belongs to the technical field of monitoring and safety early warning of deep foundation pit engineering in soft soil in geotechnical engineering, and specifically relates to a method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array. Background Technology

[0002] With the transformation of my country's new urbanization towards intensive development, urban renewal has become an important model for the development of urban central areas. As a crucial link in the urban renewal process, a large number of projects involving the operation, maintenance, renovation, and upgrading of old buildings and structures have emerged, leading to unprecedentedly complex environmental conditions for related foundation pit engineering. These projects are mostly located in the core areas of built-up areas, surrounded by densely distributed buildings, municipal pipelines, traffic arteries, and subway tunnels, placing more stringent demands on design and construction, especially on the deformation control of surrounding environmental protection objects. Therefore, the ability to timely and accurately grasp, process, and analyze deformation monitoring data of deep soft soil foundation pits, and to make forward-looking judgments on the development of foundation pit deformation, has become an important part and mainstream trend in foundation pit construction and monitoring. Existing technologies often do not address the context of urban renewal, particularly the complex geological conditions of deep soft soil layers in the eastern coastal areas of my country. Considering the applicability of foundation pits in soft soil areas, it is difficult to meet the requirements for real-time performance, accuracy, and site space constraints.

[0003] Traditional foundation pit deformation monitoring mainly relies on manual measurement and basic sensing equipment. While its technical system has limitations in accuracy, efficiency, and adaptability, it remains a common method for small and medium-sized projects. Data acquisition is discrete and discontinuous, requiring significant manual labor and complex operation. Overall frequency and efficiency are low, and time lag exists between different measuring points, making it difficult to meet real-time early warning requirements. Furthermore, measurement accuracy is easily affected by environmental conditions, human error, and weather factors. When using traditional sensor networks for foundation pit deformation monitoring, the information acquisition process can be considered a "point-based" monitoring method. Based on the layout of measuring points, only local information from the sensor installation locations can be obtained, making it difficult to comprehensively and completely reflect the overall deformation field of the foundation pit retaining structure and surrounding environment.

[0004] Currently, close-range photography is being considered as a measurement method. However, existing technologies only use a single or a small number of cameras for periodic shooting, and then analyze deformation through post-processing software. This results in problems such as processing lag, lack of real-time feedback, and low automation. Furthermore, monocular vision has limitations in depth information measurement. In practice, the ability to analyze foundation pit deformation may be insufficient. There is often a lack of in-depth processing and reprocessing of the obtained data, or a focus solely on past and present deformation data, neglecting to predict the development trend of foundation pit deformation. This generally makes it difficult to meet the higher safety management requirements of modern smart construction sites. Some technologies possess preliminary predictive capabilities for foundation pit deformation, but their data transmission process may be redundant. The models used in in-depth processing are often difficult to adapt to the temporal characteristics of foundation pit deformation, resulting in weak long-term prediction capabilities and insufficient physical constraints. These are areas that this application needs to focus on improving. Summary of the Invention

[0005] The technical problem to be solved by this invention is to provide a real-time perception and intelligent reasoning method for foundation pit deformation based on a visual terminal array. The method ensures the effectiveness, accuracy and overall reliability of the measurement by using measurement principles and array layout. It achieves better perception of the foundation pit deformation state with a certain degree of globality, continuity and reliability. By using a time-series prediction model based on machine learning and other methods, it realizes intelligent reasoning on the basic situation of foundation pit deformation, predicts the development of foundation pit deformation, and provides comprehensive opinions to assist foundation pit safety decision-making.

[0006] To address the aforementioned technical challenges, this invention provides a real-time perception and intelligent reasoning method for foundation pit deformation based on a visual terminal array. Designed for the unique scenarios of confined sites in urban renewal and deep foundation pits in soft soil, it employs a comprehensive array-style arrangement to form a multi-core, anti-occlusion visual terminal array for full-domain image acquisition. Through image processing, multi-view 3D calculation, and spatial interpolation, a continuous deformation field across the entire foundation pit is constructed, and monitoring calibration points are integrated to form multi-source constraints. Soft soil characteristics are embedded as physical constraints into an LSTM+Transformer model to construct a non-purely data-driven intelligent reasoning framework. Combining the deformation analysis mechanism of the foundation pit and the protected object, a comprehensive assessment of the overall spatial field and temporal development trend of foundation pit deformation is performed, achieving high-precision, high-robustness, and high engineering applicability for deformation prediction and safety decision-making. This effectively empowers engineering practices in the field of geotechnical engineering for deformation monitoring, perception, prediction, and control of deep foundation pits in soft soil under the background of urban renewal, including but not limited to the following steps: S1. Based on the basic situation of the soft soil deep foundation pit project under urban renewal conditions, clarify the monitoring scope of the foundation pit, form a preliminary monitoring plan for foundation pit deformation, determine the layout network of visual perception system measuring points, and determine the accuracy and frequency elements of the measurement. S2. Using close-range photography technology, a visual perception system is established for the measurement points. Considering the characteristics of the urban renewal site being small, surrounded by dense buildings and limited visibility, a multi-mode visual terminal array is selected. Through multi-camera cross-coverage and viewing angle compensation, anti-occlusion perception of non-visual areas is achieved, and image information collection of the foundation pit and surrounding sensitive targets is completed. S3. Establish a data flow transmission system of "acquisition-processing-analysis"; S4. Based on image information, combined with the interference of complex environments such as narrow sites in urban renewal, soft soil foundation pits, and multi-array deployment, preprocessing methods are used to enhance the processed images; through multi-view spatiotemporal alignment and multi-machine cross-observation constraints, high-precision three-dimensional displacement vectors of foundation pit retaining, ground surface and visible sensitive objects are calculated to form a reliable displacement time series data sequence. S5. Combining the array layout of measuring points with the environmental characteristics of the urban renewal foundation pit, and based on the three-dimensional displacement time series data calculated from multiple perspectives, a continuous deformation field covering the foundation pit, surrounding surface and sensitive targets is constructed. Displacement contour lines and strain field distribution are generated through spatial interpolation and deformation continuity constraints, and calibration points are connected to verify the spatial rationality of the deformation field. The analysis focuses on the retaining structure, support system and key parts of the foundation pit corner, and simultaneously supports the extraction of differential settlement, tilt rate and additional deformation characteristics of surrounding buildings and underground pipelines. This realizes the coupled analysis of the deformation of the foundation pit body and the response of the surrounding environment, and provides a basis for the assessment of the safety status of the soft soil foundation pit. S6. Based on the comprehensive foundation pit excavation process and soft soil characteristics, a multi-source time series dataset is constructed based on displacement time series and global deformation field. An LSTM+Transformer intelligent inference framework integrating soft soil physical constraints is established. The concepts of deformation continuity, rate limit or creep law are embedded into the model learning and calculation in the form of formulas. The deformation time series trend of foundation pit and surrounding buildings and pipelines is uniformly judged, and the long-term deformation of soft soil foundation pit is accurately predicted. S7. Based on deformation field analysis, time series prediction results and data integration, conclusions are made, coupled deformation safety assessment of foundation pit and protected object is carried out, the overall safety status of soft soil foundation pit in urban renewal sensitive environment is comprehensively evaluated, deformation development trend is predicted and graded early warning information and auxiliary decision-making suggestions are output, forming a complete closed loop including perception, calculation, prediction and decision-making.

[0007] In step S1, the basic information of the soft soil deep foundation pit project under urban renewal conditions includes the length, width and excavation depth of the foundation pit, soil stratification, soil physical and mechanical properties and construction procedures; at the same time, it includes the requirements of the surrounding environmental characteristics of the soft soil deep foundation pit project on the monitoring range, accuracy and frequency of foundation pit deformation, so as to clarify the basic elements of the specific layout of the foundation pit measuring point network, the accuracy of the measurement process and the measurement frequency, so as to improve the layout scheme of the visual perception system, including specific design and equipment selection.

[0008] In step S2, a multi-mode visual terminal array is used to address the problems of limited space, restricted visibility, and difficulty in setting up large-scale supports in urban renewal projects. Multi-mode point location surveys are conducted in stable areas or stable structures outside the foundation pit, including capping beams and walers. These surveys include support-type, wall-mounted, and hanging types to address the problem of limited construction space. Multi-camera cross-coverage ensures that there are no blind spots in key areas and achieves anti-occlusion perception in non-visual areas. Furthermore, key areas are cross-observed by at least two cameras to ensure visual perception effectiveness. Establish a measurement control network for monitoring foundation pit settlement. In an absolutely stable area, use a total station high-precision instrument, combined with high-level leveling points, to set up and measure the three-dimensional coordinates of a set of control points as the spatial reference for the entire visual measurement system. Based on the equipment list, site map, network topology diagram, and accuracy and frequency requirements, determine the hardware deployment and install intelligent vision terminal nodes, including integrated industrial cameras and edge computing units; To ensure visual recognition effectiveness, targets are set up on the monitoring targets, such as using high-contrast circular coded markers. At the same time, when deploying the visual terminal array, the corresponding sensing system is correctly calibrated to ensure the accuracy of the results. That is, using the deployed control points, the entire visual terminal array is calibrated with high precision through the bundle adjustment algorithm to accurately obtain the intrinsic and extrinsic parameter matrices of each camera. By integrating the information from intelligent vision terminal nodes to form hardware synchronization or high-precision software synchronization, continuous image sequences can be acquired at the foundation pit site.

[0009] The bundle adjustment algorithm involved in the calibration process aims to minimize the reprojection error between the actual image and the theoretical projection point position. The core formulas related to the coordinate relationships are as follows: ; ; Where matrix K is the intrinsic parameter matrix; f x f y The equivalent focal length of the camera in the x and y directions; u0 and v0 are the principal point coordinates, i.e., the intersection of the camera's optical axis and the image plane; s is the axis tilt factor, which is 0; λ is the scale factor, i.e., the Z coordinate in the camera coordinate system; (uv 1) T (RT) is the homogeneous coordinate of the image pixel coordinates; (X) is the extrinsic parameter matrix of the camera; w Y w Z w 1) T These are homogeneous coordinates of the world coordinate system.

[0010] The data flow transmission in step S3, "acquisition-processing-analysis," establishes a cloud-edge collaborative data transmission link. This includes: edge processing within each visual terminal, running a lightweight deep learning model, performing target recognition and tracking, and extracting sub-pixel-level image coordinates and feature descriptions; the edge nodes upload the result data to the cloud platform via a wireless network, forming a comprehensive cloud-edge collaboration. In subsequent steps, the cloud is responsible for massive data storage, large-scale computing, and intelligent analysis, while the edge is responsible for front-end perception and real-time processing, resulting in efficient collaboration.

[0011] In step S4, preprocessing techniques are used to enhance the processed images and extract real-time displacement information of the foundation pit deformation. This involves combining multi-view 3D displacement information calculation, specifically including: For the characteristics of urban renewal sites such as confined spaces with occlusion, complex lighting, and strong interference from dust, rain, and fog, adaptive denoising, contrast enhancement, low-light compensation, and anti-occlusion optimization are performed on the images uploaded by the visual terminal array to improve the stability of image features in complex environments; data captured by different cameras at the same time are spatiotemporally aligned, and effective data integration is performed for the same physical point in different images; based on the parameters calibrated in step S2, high-precision 3D coordinates of each target / feature point in the world coordinate system are calculated using multi-view geometry and forward intersection principles; the real-time calculated high-precision 3D coordinates are compared with the initial reference coordinates to obtain the 3D displacement vector (ΔX, ΔY, ΔZ) of each monitoring point, forming a displacement time-series database to support the construction of the global deformation field and subsequent coupled inference.

[0012] Image denoising employs a bilateral filtering method, considering spatial weights, range weights, and weighted averaging. The core formulas are as follows: ; Where d is the spatial Euclidean distance between neighboring pixels, σ s The standard deviation of the spatial Gaussian kernel. Grayscale value σ represents the grayscale difference between neighboring pixels. r The standard deviation of the Gaussian kernel in the range, and the joint weighting coefficients. Spatial weight Sum range weight The product of the two values ​​outputs the grayscale value. The image contrast enhancement is obtained by weighted averaging of gray values ​​with joint weighting coefficients. The image contrast enhancement utilizes adaptive histogram equalization, which takes into account block division, histogram processing, gray-level mapping, and interpolation fusion steps, and supplements the CLAHE method to limit the contrast processing of the image.

[0013] Multi-view geometry studies the mathematical relationships between cameras and spatial points and image points from different perspectives. The core is to establish spatial constraints through camera intrinsic and extrinsic parameters and image point information to provide a basis for subsequent coordinate calculation. Its core formula has been described in step S2. After completing camera calibration, obtaining the camera's intrinsic and extrinsic parameter matrix data set and image point matching, forward intersection processing is performed. After considering image point normalization and establishing the projection ray equation, the coordinates of spatial points are solved using least squares.

[0014] In step S5, the overall deformation field of the foundation pit is analyzed, focusing on the coupled deformation of key local areas and the surrounding environment. Specifically, this includes: generating a global displacement field covering the entire monitoring area using spatial interpolation algorithms from the three-dimensional displacement data of discrete monitoring points, and visually expressing the overall spatial deformation status of the foundation pit using displacement contour maps and strain field distribution maps; based on the global deformation field, focusing on the detailed displacement vector analysis of the retaining structure, support system, and key parts of the foundation pit's external corners; and automatically extracting the relative deformation and additional strain along the building line, differential settlement of adjacent points, building tilt rate, and pipeline axis direction, taking into account the dense distribution of buildings and underground pipelines around the urban renewal foundation pit, and using the true values ​​of traditional measuring points as anchor points to verify the rationality of the global deformation field; and forming coupled deformation characteristics of the foundation pit and the protected object by uniformly expressing and correlating the deformation of the foundation pit itself with the deformation of the surrounding environment, solving the problem that traditional monitoring only focuses on the foundation pit itself and cannot comprehensively evaluate the linkage deformation of the soft soil foundation pit's surrounding environment, thus improving the completeness and accuracy of safety assessment in complex and sensitive scenarios. Among them, the specific forms of anchor points whose true values ​​are traditional measuring points include, but are not limited to, the corresponding measurement results of total stations, inclinometers, and settlement magnetic rings.

[0015] In step S6, the intelligent inference framework includes: constructing a time-series dataset, primarily integrating the displacement time-series data generated in step S4, and combining it with other information to form a multi-source time-series dataset; using a prediction model LSTM+Transformer incorporating urban soft soil foundation pit characteristics to form a deep learning model, training it on historical data, learning the deformation development pattern over time, and establishing a prediction model; inputting the latest displacement data into the prediction model to predict the deformation development trend in the next few hours to days, achieving real-time foundation pit deformation trend prediction and achieving high-precision prediction results; the core formula for the LSTM model gating mechanism is as follows: ; Where: f t i t Here, W represents the output of the forget gate and the input gate, b represents the bias term, and σ represents the sigmoid function. This indicates vector concatenation.

[0016] The comprehensive prediction model under the physical constraints of soft soil mainly considers a combination of LSTM and Transformer models. The LSTM model, or Long Short-Term Memory network, accurately captures short-term fluctuations and long-term trends through its gating mechanism, such as instantaneous displacement and cumulative settlement of points. Therefore, it possesses characteristics suitable for predicting foundation pit deformation and has a high degree of matching with corresponding mechanical properties. Multi-layer LSTMs are used to extract features at different time scales; shallow layers capture short-term patterns, while deeper layers understand long-term patterns. The Transformer model overcomes temporal distance limitations through its self-attention mechanism, globally capturing cross-time-period correlations of information. It is suitable for predicting long-term trends in foundation pits, and its parallel computing characteristics are also adapted to the efficient processing of massive monitoring data. The encoder processes historical data and extracts features, while the decoder generates future predictions based on these features. Considering that LSTM can process predictions over several days with proper design, and that Transformer performs worse than LSTM in small sample situations, model selection should be based on data characteristics rather than fixed time thresholds.

[0017] To address the rheological, creep, consolidation hysteresis, and deformation rate sensitivity of soft soil, physical constraints on soft soil deformation are embedded into the LSTM+Transformer prediction model. These constraints include rheological time-series decay laws, consolidation deformation trends, displacement continuity constraints, and deformation rate limits. The core formulas involved include: ; Among them: S, S e S v S p For total deformation, elastic deformation, viscoelastic deformation, and plastic flow, τ is the creep time factor.

[0018] In step S7, the comprehensive reference opinions for assisting in foundation pit safety decision-making are provided, including: establishing a coupled safety judgment system for foundation pits and protected objects based on the full-domain deformation field constructed in step S5, the intelligent inference deformation prediction results embedded in the physical constraints of soft soil in step S6, and the consistency judgment conclusions of visual perception data and traditional calibration points; automatically calculating and extracting sensitive indicators such as differential settlement of surrounding buildings, tilt rate, and additional strain of underground pipelines, and incorporating foundation pit retaining deformation, building safety, and pipeline safety into a unified system for joint analysis; and outputting graded early warning information according to risk level in combination with the strict control requirements of the surrounding environment of soft soil foundation pits in the context of urban renewal, and providing targeted construction control, support adjustment, and surrounding protection auxiliary decision-making suggestions, so as to achieve full-cycle, highly reliable, and systematic safety perception and intelligent decision-making for deep soft soil foundation pits in urban renewal.

[0019] The one or more technical solutions proposed in this invention have at least the following beneficial effects: 1) In response to the current urban renewal context, deep foundation pit excavation projects in soft soil areas exhibit strict requirements for settlement control of environmental protection objects during construction, posing challenges to the application practice of timely, accurate acquisition, processing, and analysis of foundation pit deformation monitoring data. This invention utilizes an intelligent reasoning framework based on data processing machine learning to empower engineering practices of foundation pit deformation monitoring, perception, prediction, and control in the field of geotechnical engineering. 2) In view of the current situation in the field of traditional foundation pit monitoring and safety early warning, the collection and processing of foundation pit deformation information is complicated, localized, and has a low degree of automation. This invention proposes a method for extracting foundation pit deformation data by referring to a visual terminal array deployed under a visual perception system. The measurement principle and array layout ensure the effectiveness, accuracy and overall reliability of the measurement. 3) Addressing the contradiction between the advanced technology of close-range photography for information extraction and its insufficient application in engineering practice, and considering the difficulty in ensuring the accuracy of displacement information extraction from a single-viewpoint perspective, as well as the fact that traditional monitoring results are point-based rather than area-based full-field monitoring, this paper proposes a system of visual perception units deployed in an array to form a displacement time-series database with spatiotemporal alignment. This enables a better perception of the deformation state of the foundation pit with a certain degree of globality, continuity, and reliability. Furthermore, considering the potential limitation of construction space in the background of urban renewal foundation pits, a multi-mode visual terminal matrix layout method is proposed. 4) Addressing the relatively weak deformation prediction capabilities of traditional foundation pit deformation monitoring practices, this paper utilizes time-series prediction models based on machine learning to achieve intelligent reasoning about the basic deformation situation of foundation pits, predict the development of deformation, and provide comprehensive opinions to assist in foundation pit safety decisions. By embedding physical constraints incorporating soft soil characteristics into an LSTM+Transformer model, an intelligent reasoning framework is formed. This framework differs from traditional purely data-driven artificial intelligence models, offering greater reliability and persuasiveness, and avoiding a purely black-box prediction mode. Simultaneously, the paper considers combining data from traditional monitoring methods with true value calibration, improving the authenticity, accuracy, and reliability of the overall deformation field and temporal development trend prediction results. Coupled deformation safety assessment of the foundation pit and protected objects is conducted. Unlike traditional techniques that only focus on the foundation pit displacement data itself, this approach comprehensively considers the displacement and deformation of surrounding buildings and structures, especially underground pipelines, making it more suitable for the complex environment and stringent environmental protection requirements of soft soil foundation pit construction in the context of urban renewal. Attached Figure Description

[0020] The accompanying drawings, which form part of this application, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a schematic diagram of the overall architecture of the system of the present invention; Figure 2 This is a schematic diagram of typical construction elements in the context of urban renewal, which is the focus of this invention. Figure 3 This is a typical deployment diagram of the visual terminal array around the foundation pit in this invention; Figure 4 This is a schematic diagram illustrating the principle of multi-view three-dimensional displacement calculation in this invention; Figure 5 This is a flowchart illustrating the intelligent reasoning process for spatiotemporal data of foundation pit deformation in this invention. Explanation of the labels in the diagram: 1—Reference point of displacement measurement control network; 2—Control point of displacement measurement control network; 3—Route connecting benchmark points and control points; 4—Layout of displacement measuring points around the foundation pit; 5—Layout of displacement measuring points for foundation pit retaining structure; 6—Route for connecting measuring points; 7—Visual terminal array base station; 8—Industrial camera; 9—Machine vision micro-motion device; 10—Tripod, fixing and adjustment device; 11—Observation range of the visual terminal array base station; 12—Real-time sensing displacement information transmission; 13—Information integration base station; 14—Image information collection and preprocessing; 15—Image information extraction and analysis; 16—Intelligent reasoning module for foundation pit deformation; 17—Data output and feedback. Detailed Implementation

[0021] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings. The executing entity of a specific embodiment of the present invention is a computing service device with data processing, network communication, and program execution functions.

[0022] Figure 2 This invention focuses on typical construction elements in the context of urban renewal. The left side of the figure lists seven key concepts that are of concern and involved in urban renewal projects, such as building environmental impact analysis. The right side illustrates the specific project objects that are to be updated or protected in six urban renewal projects.

[0023] Figure 3 This is a typical deployment diagram of the visual terminal array around the foundation pit in this invention. The diagram shows the basic components of the visual terminal array and its typical layout around the foundation pit, including the displacement measurement control point network, the visual terminal array base station, the connection route and mode, and the data transmission and processing process; wherein: The measurement control reference unit includes displacement measurement control network reference points 1, displacement measurement control network control points 2, and reference point and control point connection route 3. The displacement measurement control network reference points 1 are located in an absolutely stable area outside the foundation pit, far from construction disturbances. Together with the displacement measurement control network control points 2, which are uniformly distributed on the stable ground outside the foundation pit, they form the whole-field measurement control reference unit. The reference points and control points connection route 3 forms a control point network, used to establish a unified world coordinate system and provide a high-precision spatial reference for vision system calibration and displacement calculation. The sensing and monitoring deployment unit includes: 4. Displacement measuring points around the foundation pit; 5. Displacement measuring points for the foundation pit retaining structure; and 6. Connecting routes for the measuring points. The specific layout of the surface displacement / structural displacement measuring points around the foundation pit is designed according to actual conditions and needs. The basic model considers: points located on the outer surface of the foundation pit, along underground pipelines, and around the foundations of adjacent buildings, incorporated into the sensing and monitoring unit as visual identification target observation points; and points spaced along the foundation pit capping beam and support sidewalls, incorporated into the sensing and monitoring unit, focusing on monitoring the deformation of the support structure. Both also simultaneously consider artificial settlement and horizontal displacement methods, using measured data to input into the prediction model, as a reserve for true value verification. The visual perception acquisition unit includes a visual terminal array base station 7, an industrial camera 8, a machine vision micro-motion device 9, a tripod, fixing and adjustment device 10, and a visual terminal array base station observation range 11. It is deployed based on the foundation pit capping beam and the surrounding stable walls, and is installed using the aforementioned multi-mode layout according to specific conditions, supporting the entire set of image acquisition equipment. The image acquisition equipment is the core imaging device of the visual perception acquisition unit, responsible for continuously capturing real-time images of the foundation pit and surrounding monitoring area from multiple angles. Multiple projects work together to complete multi-mode deployment and synchronous acquisition of images across the entire area in confined spaces. The cloud-edge collaborative transmission unit includes a real-time sensing displacement information transmission unit 12 and an information integration base station 13, which completes the aggregation, synchronization and uplink transmission of multi-source data on site to the cloud. The intelligent analysis and reasoning unit includes image information collection and preprocessing (14), image information extraction and analysis (15), and an intelligent reasoning module for foundation pit deformation (16). It achieves image calculation, deformation field construction, intelligent prediction of physical constraints, and coupled safety assessment. A truth value verification channel is reserved in the data acquisition module, and the results are used to provide feedback to the visual perception acquisition unit and the intelligent analysis and reasoning unit. Data output and feedback (17) is the result application output unit, assisting in the visualization of monitoring results, risk warning, and engineering decision feedback.

[0024] Figure 4 This is a schematic diagram illustrating the principle of multi-view three-dimensional displacement calculation in this invention. It shows the basic relationship between displacement data in the world coordinate system, camera coordinate system, and photo coordinate system, which helps in understanding the relevant principles and basic formulas.

[0025] Figure 5This is a flowchart illustrating the intelligent inference process for spatiotemporal data of foundation pit deformation in this invention. The diagram shows the process of acquiring foundation pit image information, summarizing foundation pit displacement data sets, and using an intelligent inference model to predict the overall situation and temporal trend of the foundation pit deformation field. It demonstrates the basic relationships between foundation pit deformation patterns, development trend predictions, and related data, as well as typical visualization results. The list data is a schematic diagram of the foundation pit deformation sensing data set. The upper sub-figure is a schematic diagram of the overall analysis of the deformation spatial field, considering the multi-measurement point data pattern of the foundation pit space based on the sensing data set, and comprehensively considering surrounding existing buildings as protected objects, presented in the form of settlement deformation contour lines. The lower sub-figure is the deformation time development trend prediction; the hollow data points within the dashed boxes indicated by the arrows are typical results of measurement point time-series predictions considering a longer time span.

[0026] like Figure 1 As shown, this invention provides a method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array, including: Step S1: Obtain basic data for the soft soil foundation pit project in the context of urban renewal. The basic data includes the length, width, and excavation depth of the foundation pit, soil stratification, soil physical and mechanical properties, and construction procedures. At the same time, it includes the requirements of the surrounding environment for monitoring the deformation of the foundation pit in terms of scope, accuracy, and frequency. This clarifies the basic elements such as the specific layout of the foundation pit measuring point network, the accuracy of the measurement process, and the measurement frequency, so as to improve the layout plan of the visual terminal array, including specific design and equipment selection. Figure 2 The illustration shows typical construction elements in the context of urban renewal, which is the focus of this invention. In step S1, for the complex working conditions corresponding to the concepts of building symbiosis, resilient building, and green and low-carbon in urban renewal, the monitoring scope of the foundation pit and surrounding buildings, underground pipe corridors, transportation facilities, and community environment is clarified, taking into account the comprehensive management and control needs of building safety, traffic environment, and underground space development. The visual perception acquisition unit and intelligent analysis and reasoning unit in the subsequent steps are also dedicated to adapting to the complex construction environment in this context, integrating the innovative concepts of progressive construction, adaptive reuse of buildings, and green and low-carbon development in the context of urban renewal, and providing high-precision and high-reliability safety perception and intelligent decision support for the protection of various existing buildings and infrastructure, underground space development, and community environment improvement.

[0027] Step S2 involves deploying a multi-mode visual terminal array under the visual perception system. This includes: conducting field-of-view analysis in the stable area surrounding the foundation pit; addressing the challenges of limited space, restricted visibility, and difficulty in large-scale scaffolding erection in urban renewal projects; and conducting multi-mode point location surveys in the stable area surrounding the foundation pit or at stable structures, including capping beams and walers, using scaffolding, wall-mounted, and hanging methods to address the limited construction space; ensuring no blind spots in key areas through multi-camera cross-coverage; achieving anti-occlusion perception in non-visual areas; and ensuring visual perception effectiveness by having key areas cross-observed by at least two cameras. A measurement control network for foundation pit settlement monitoring is established. In the absolutely stable area, using high-precision instruments such as total stations and high-level leveling points, a set of three-dimensional coordinates of control points are deployed and measured as the spatial reference for the entire visual measurement system. Based on the equipment list, site location map, network topology diagram, and accuracy and frequency requirements, the hardware deployment is determined, and intelligent vision terminal nodes are installed, including components such as integrated industrial cameras and edge computing units. Targets, such as high-contrast circular coded markers, are placed on the monitored targets to ensure visual recognition effects. Simultaneously, the correct calibration of the corresponding sensing system is performed during the deployment of the vision terminal array to ensure the accuracy of the results. This involves using the deployed control points and a bundle adjustment algorithm to perform high-precision system calibration of the entire vision array, accurately obtaining the intrinsic and extrinsic parameter matrices of each camera. Information from the intelligent vision terminal nodes is integrated to form hardware synchronization or high-precision software synchronization, acquiring continuous image sequences from the foundation pit site.

[0028] Twelve intelligent vision terminal nodes were deployed on the stable capping beams and surrounding ground around the foundation pit. Each node used a 20-megapixel industrial CMOS camera with a built-in edge computing unit to establish cloud communication. Images with a resolution of 1920×1080 pixels were acquired for four control points P1, P2, P3, and P4 and two cameras C1 and C2. The camera sensor size was approximately 6.4mm×4.8mm. For world coordinates measured by a total station, the control points (X, Y, Z) (m) were (0.0, 0.0, 0.0), (10.0, 0.0, 0.0), (10.0, 8.0, 0.0), and (0.0, 8.0, 0.0), respectively. Two cameras captured images containing these control points from different angles, and sub-pixel precision image point coordinates were extracted using image processing algorithms. For C1, the coordinates are (312.45, 524.78), (1256.89, 498.32), (1320.12, 823.56), and (295.67, 845.91); for C2, the coordinates are (1589.34, 432.18), (1420.56, 756.89), (987.45, 892.34), and (1123.78, 412.67). After simple calibration, the initial values ​​are set as follows: f x f yGiven a resolution of 1500 pixels, u0 and v0 are 960 and 540 respectively, and considering initial values ​​of certain coefficients of variation k1=k2=p1=p2=0. For C1, the extrinsic parameter matrix is ​​as follows; C2 is similar. ; Using C1 to observe P1, the world coordinates are transformed to camera coordinates and then projected onto the normalized image plane. After applying a distortion model, the coordinates are transformed back to pixel coordinates. The reprojection error is calculated and the process is iterated multiple times. After iterative optimization, the intrinsic and extrinsic parameters of C1 are adjusted as follows. The total reprojection error after optimization is 0.86 pixels, and the average reprojection error per point is approximately 0.15 pixels. ; .

[0029] In step S3, during data acquisition and edge processing, all cameras synchronously acquire images at a set frequency. The edge computing unit runs a lightweight model based on an improved YOLOv5s, identifies targets in the images in real time, outputs the sub-pixel-level image coordinates (u,v) of the target center corresponding to the measurement point, and uploads them to the cloud, establishing a cloud-edge collaborative data transmission link. After receiving the data at the cloud data processing center, the images uploaded to the cloud platform via data stream from the visual terminal array are preprocessed, including denoising, contrast enhancement, and haze degradation model correction, to improve image quality. Data captured by different cameras at the same time are spatiotemporally aligned. Subsequently, for each target, using its image coordinates in different cameras, combined with the calibrated camera parameters, its current three-dimensional coordinates are calculated through forward intersection and bundle adjustment. Using the coordinates of the first phase before excavation as a reference, the cumulative displacement of each point is calculated. The high-precision three-dimensional coordinates of each target / feature point in the world coordinate system are calculated. The real-time calculated three-dimensional coordinates are compared with the initial reference coordinates to obtain the three-dimensional displacement vector (ΔX, ΔY, ΔZ) of each monitoring point, forming a displacement time series database.

[0030] Combination Figure 4 It represents the relative relationship between the overall coordinates, camera coordinates, and image coordinates, so as to facilitate the calculation and understanding of the displacement information of the foundation pit measuring points using the bundle adjustment method.

[0031] For a specific target, the image coordinates (pixels) observed by cameras C1 and C2 are p1=[312.45,524.78]ᵀ and p2=[1589.34,432.18]ᵀ, respectively. A system of linear equations is established by the intersection of these equations, expanded into equation form, and simplified for calculation. ; Combining observations from C1 and C2, the relevant equations are obtained. After SVD decomposition and normalization, we have P = [0.101 / (-0.040), 0.048 / (-0.040), 0.994 / (-0.040), 1]ᵀ = [-2.525, -1.200, -24.850, 1]ᵀ. Therefore, P w =[-2.525,-1.200,-24.850]ᵀ (m). Reprojecting C1, compared with the observed value [312.45,524.78]ᵀ, the error is calculated as e1=[0.04,0.02]ᵀ pixels, ||e1||=0.045 pixels. The same applies to C2. For a specific point ΔP=[0.002,0.002,-0.002]ᵀ meters, the horizontal displacement, settlement, and total displacement are obtained.

[0032] Step S4 focuses on the overall deformation field of the foundation pit and the coupled deformation of key local areas with the surrounding environment. Specifically, this includes: generating a global displacement field covering the entire monitoring area using spatial interpolation algorithms from the three-dimensional displacement data of discrete monitoring points, i.e., displacement contour maps and strain field distribution maps, to intuitively express the overall spatial deformation situation of the foundation pit; based on the global deformation field, conducting detailed displacement vector analysis on the retaining structure, support system, and key corners of the foundation pit; and, considering the dense distribution of buildings and underground pipelines around the urban renewal foundation pit, extracting the settlement along the building line, differential settlement of adjacent points, building tilt rate, and relative deformation and additional strain along the pipeline axis, and using the true values ​​of traditional measuring points as anchor points to verify the rationality of the global deformation field; and by unifying and analyzing the deformation of the foundation pit itself and the deformation of the surrounding environment, forming the coupled deformation characteristics of the foundation pit and the protected object, solving the problem that traditional monitoring only focuses on the foundation pit itself and cannot comprehensively evaluate the linkage deformation of the soft soil foundation pit's surrounding environment, thus improving the completeness and accuracy of safety assessment in complex and sensitive scenarios. Among them, the true value of the traditional measuring point is the specific form of the anchor point, which is selected from the corresponding measurement results of the total station, inclinometer, and settlement magnetic ring. Combined with... Figure 5 It illustrates the data prediction process of obtaining the overall situation and time trend of the foundation pit deformation field after acquiring the foundation pit image information, summarizing the foundation pit displacement data set, and obtaining the foundation pit deformation field based on the intelligent reasoning framework.

[0033] Step S5: Based on the time-series dataset, establish a prediction model incorporating the physical constraints of soft soil, utilizing LSTM combined with a Transformer model. For the LSTM, hidden layers are set, using 2-3 layers stacked to achieve shallow layers capturing high-frequency fluctuations and deep layers extracting low-frequency trends. For small to medium-sized foundation pits, the number of units in a single hidden layer is set to 64 to 128; for large and complex foundation pits, this is increased to 128 to 256, balancing feature extraction capability and computational efficiency. A forgetting gate threshold is set, and the tendency to enhance long-term memory is enhanced and short-term noise is weakened by initializing the bias, such as -0.1 to -0.5, or by data-driven learning. For input gate activation, the Tanh activation function is used, matching the standardized displacement data. Data normalization is involved, and higher weights are assigned to input features of sudden working conditions, assisted by an attention mechanism, with weight coefficients of 1.2 to 1.5. The Transformer model overcomes the temporal distance limitation through a self-attention mechanism, globally capturing cross-time-period correlations of information, making it suitable for long-term trend prediction of foundation pits. For the number of multi-head attention sensors, a strict adherence to the formula `number of sensors = max(1, feature dimension / / 64)` is considered to effectively capture sub-associations and avoid feature fragmentation caused by an excessive number of sensors. The number of encoder / decoder layers is selected based on both data volume and sequence complexity, with two layers each chosen for specific basic scenarios. For the fusion of location encoding and construction features, basic temporal encoding is considered, employing sine and cosine encoding. Regarding the feedforward network dimension, the feedforward hidden layer dimension is considered to be equal to the feature dimension × 4, ensuring a balance between nonlinear fitting capability and computational efficiency. A comprehensive intelligent inference model is utilized to train the model on perceived data, inputting the latest displacement data to predict the deformation development trend over the next few hours to days, achieving relatively real-time prediction of the foundation pit deformation trend and meeting certain accuracy requirements. A loss function is also considered to achieve a dual orientation of engineering constraints and data fitting. For basic loss, Huber loss is used for small samples to resist outliers; MAE is used for large datasets to align with engineering error assessment practices. For physical constraint loss, constraints such as deformation continuity and rate limits are added to prevent prediction results from violating physical laws. Considering dynamic weighting, the loss weight of the settlement component is set to 1.5 times that of the horizontal component, matching the priority given to settlement in the matching project. Specializations related to soft soil should be considered based on the background of the foundation pit project and the actual soil properties; the creep correlation coefficient is selected between 15d and 60d.

[0034] Step S6: When the prediction error analysis results are appropriate, it indicates to some extent that the foundation pit deformation prediction results obtained by the intelligent inference model of this method are good, and it reflects the effect of using displacement information extracted based on visual perception array in the context of urban renewal. Based on the constructed global deformation field, combined with the intelligent inference deformation prediction results embedded with soft soil physical constraints, and the consistency judgment conclusion between visual perception data and traditional calibration points, a coupled safety judgment system for foundation pits and protected objects is established; foundation pit retaining deformation, building safety, and pipeline safety are incorporated into a unified system for joint judgment; combined with the strict control requirements of the surrounding environment of soft soil foundation pits in the context of urban renewal, graded early warning information is output according to risk level, and targeted construction control, support adjustment, and surrounding protection auxiliary decision-making suggestions are given, realizing full-cycle, highly reliable, and systematic safety perception and intelligent decision-making for deep foundation pits in soft soil in urban renewal. Therefore, it can assist engineers in predicting whether the foundation pit deformation meets expectations and is within a reasonable and controllable range by combining relevant results with actual engineering conditions, and also provides a good comprehensive perspective for deformation-based risk control during foundation pit construction.

[0035] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. 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 method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array, characterized in that: Including but not limited to the following steps: S1. Based on the basic situation of the soft soil deep foundation pit project under urban renewal conditions, clarify the monitoring scope of the foundation pit, form a preliminary monitoring plan for foundation pit deformation, determine the layout network of visual perception system measuring points, and determine the accuracy and frequency elements of the measurement. S2. Using close-range photography technology, a visual perception system is established for the measurement points. A multi-mode visual terminal array is selected and deployed. Through multi-camera cross-coverage and perspective compensation, anti-occlusion perception in non-line-of-sight areas is achieved, and image information acquisition of the foundation pit and surrounding sensitive targets is completed. In the stable area or stable structure outside the foundation pit, including the capping beam and waler, multi-mode point location is determined, including bracket type, wall-mounted type, and hanging type, to cope with the limited construction space. S3. Establish a data flow transmission of "acquisition-processing-analysis"; S4. Based on image information, combined with the complex environmental interference of urban renewal sites in small spaces, soft soil foundation pits, and multi-array deployment, preprocessing methods are used to enhance the processed images. By using multi-view spatiotemporal alignment and multi-machine cross-observation constraints, high-precision three-dimensional displacement vectors of foundation pit retaining, ground surface and visible sensitive objects are calculated to form a reliable displacement time series data sequence. S5. Combining the array layout of measuring points with the environmental characteristics of the urban renewal foundation pit, based on the three-dimensional displacement time series data calculated from multiple perspectives, a continuous deformation field covering the foundation pit, surrounding surface and sensitive targets is constructed. Displacement contour lines and strain field distribution are generated through spatial interpolation and deformation continuity constraints, and calibration points are connected to verify the spatial rationality of the deformation field. The key parts of the retaining structure, support system and foundation pit corner are analyzed. At the same time, the differential settlement, tilt rate and additional deformation characteristics of surrounding buildings and underground pipelines are extracted to realize the coupled analysis of foundation pit deformation and surrounding environment response. S6. Based on the construction process of the foundation pit excavation and the characteristics of soft soil, a multi-source time series dataset is constructed based on displacement time series and global deformation field; an LSTM+Transformer intelligent inference framework is established that incorporates the physical constraints of soft soil, and the concepts of deformation continuity, rate limit or creep law are embedded into the model learning and calculation in the form of formulas to make a unified judgment on the deformation time series trend of the foundation pit and surrounding buildings and pipelines. To address the rheological, creep, consolidation hysteresis, and deformation rate sensitivity of soft soil, physical constraints on soft soil deformation are embedded in the LSTM+Transformer prediction model. These constraints include rheological time-series decay laws, consolidation deformation trends, displacement continuity constraints, and deformation rate limit constraints. The core formulas involved include: ； Among them: S, S e S v S p For total deformation, elastic deformation, viscoelastic deformation, and plastic flow, τ is the coefficient considering the influence of creep time; S7. Based on deformation field analysis, time series prediction results and data integration, conclusions are made, coupled deformation safety assessment of foundation pit and protected object is carried out, the overall safety status of soft soil foundation pit in urban renewal sensitive environment is comprehensively evaluated, deformation development trend is predicted and graded early warning information and auxiliary decision-making suggestions are output, forming a complete closed loop including perception, calculation, prediction and decision-making.

2. The method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array according to claim 1, characterized in that: In step S1, the basic information of the soft soil deep foundation pit project under urban renewal conditions includes the length, width and excavation depth of the foundation pit, soil stratification, soil physical and mechanical properties and construction procedures; at the same time, it includes the requirements of the surrounding environmental characteristics of the soft soil deep foundation pit project on the monitoring range, accuracy and frequency of foundation pit deformation, so as to clarify the basic elements of the specific layout of the foundation pit measuring point network, the accuracy of the measurement process and the measurement frequency, so as to improve the layout scheme of the visual perception system.

3. The method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array according to claim 1, characterized in that: In step S2, the multi-mode visual terminal array under the visual perception system is arranged to ensure that there are no blind spots in the key area through the cross coverage of multiple cameras, and to achieve anti-occlusion perception in non-line-of-sight areas. The key area is cross-observed by at least two cameras to ensure the visual perception effect. Establish a measurement control network for monitoring foundation pit settlement. In an absolutely stable area, use a total station high-precision instrument, combined with high-level leveling points, to set up and measure the three-dimensional coordinates of a set of control points as the spatial reference for the entire visual measurement system. Based on the equipment list, site map, network topology diagram, and accuracy and frequency requirements, determine the hardware deployment and install intelligent vision terminal nodes, including integrated industrial cameras and edge computing units; To ensure visual recognition effectiveness, targets are set up on the monitoring targets. At the same time, when deploying the visual terminal array, the corresponding sensing system is correctly calibrated. Using the deployed control points, the entire visual terminal array is calibrated with high precision through the bundle adjustment algorithm to accurately obtain the intrinsic and extrinsic parameter matrices of each camera. The information of intelligent vision terminal nodes is integrated to form hardware synchronization or high-precision software synchronization, so as to realize the acquisition of continuous image sequences at the foundation pit site. The core formulas related to the coordinate relationships in the bundle adjustment algorithm involved in the calibration process are as follows: ； ； Where matrix K is the intrinsic parameter matrix; f x f y The equivalent focal length of the camera in the x and y directions; u0 and v0 are the principal point coordinates, i.e., the intersection of the camera's optical axis and the image plane; s is the axis tilt factor, which is 0; λ is the scale factor, i.e., the Z coordinate in the camera coordinate system; (u v1) T (RT) is the homogeneous coordinate of the image pixel coordinates; (X) is the extrinsic parameter matrix of the camera; w Y w Z w 1) T These are homogeneous coordinates of the world coordinate system.

4. The method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array according to claim 1, characterized in that: The data flow transmission in step S3, "acquisition-processing-analysis," establishes a cloud-edge collaborative data transmission link. This includes: edge processing within each visual terminal, running a lightweight deep learning model, performing target recognition and tracking, and extracting sub-pixel-level image coordinates and feature descriptions; the edge nodes upload the result data to the cloud platform via a wireless network, forming a comprehensive cloud-edge collaboration. In subsequent steps, the cloud is responsible for massive data storage, large-scale computing, and intelligent analysis, while the edge is responsible for front-end perception and real-time processing, thus forming a highly efficient collaboration.

5. The method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array according to claim 1, characterized in that: In step S4, the images uploaded by the visual terminal array undergo adaptive denoising, contrast enhancement, low-light compensation, and anti-occlusion optimization to improve the stability of image features in complex environments. Data captured by different cameras at the same time are spatiotemporally aligned, and effective data integration is performed for the same physical point in different images. Based on the parameters calibrated in step S2, high-precision three-dimensional coordinates of each target / feature point in the world coordinate system are calculated using multi-view geometry and forward intersection principles. The high-precision three-dimensional coordinates calculated in real time are compared with the initial reference coordinates to obtain the three-dimensional displacement vector of each monitoring point, forming a displacement time series database to support the construction of the global deformation field and subsequent coupled inference.

6. The method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array according to claim 5, characterized in that: In step S4, the adaptive image denoising uses a bilateral filtering method, considering spatial weights, range weights, and weighted averages. The core formula is as follows: ； Where d is the spatial Euclidean distance between neighboring pixels, σ s The standard deviation of the spatial Gaussian kernel. Grayscale value σ represents the grayscale difference between neighboring pixels. r The standard deviation of the Gaussian kernel in the range, and the joint weighting coefficients. Spatial weight Sum range weight The product of the two values ​​outputs the grayscale value. The image contrast enhancement is obtained by weighted averaging of gray values ​​with joint weighting coefficients. The image contrast enhancement utilizes adaptive histogram equalization, which takes into account block division, histogram processing, gray-level mapping, and interpolation fusion steps, and supplements the CLAHE method to limit the contrast processing of the image.

7. The method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array according to claim 1, characterized in that: In step S5, the overall deformation field of the foundation pit is analyzed, focusing on the coupled deformation of key local areas with the surrounding environment. Specifically, this includes: generating a global displacement field, displacement contour map, and strain field distribution map covering the entire monitoring area from the three-dimensional displacement data of discrete monitoring points using a spatial interpolation algorithm, which visually expresses the overall spatial deformation trend of the foundation pit; based on the global deformation field, a detailed displacement vector analysis is performed on the retaining structure, support system, and key corners of the foundation pit; simultaneously, considering the dense distribution of buildings and underground pipelines around the urban renewal foundation pit, the relative deformation and additional strain along the building line, differential settlement between adjacent points, building tilt rate, and pipeline axis direction are extracted, and the true values ​​of the measuring points are used as anchor points to verify the rationality of the global deformation field; by unifying and analyzing the deformation of the foundation pit itself and the deformation of the surrounding environment, the coupled deformation characteristics of the foundation pit and the protected object are formed.

8. The method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array according to claim 1, characterized in that: In step S6, the intelligent inference framework based on the time-varying dataset of foundation pit deformation data includes: constructing a time-series dataset, mainly integrating the displacement time-series data generated in step S4, and combining it with other information to form a multi-source time-series dataset; using a prediction model LSTM incorporating urban soft soil foundation pit characteristics combined with Transformer to form a deep learning model, training on historical data, learning the deformation development pattern over time, and establishing a prediction model; inputting the latest displacement data into the prediction model to predict the deformation development trend in the next few hours to days, achieving real-time foundation pit deformation trend prediction and achieving high-precision prediction results; the core formula for the LSTM model gating mechanism is as follows: ； Where: f t i t Here, W represents the output of the forget gate and the input gate, b represents the bias term, and σ represents the sigmoid function. This indicates vector concatenation.

9. The method for real-time perception and intelligent reasoning of foundation pit deformation based on a visual terminal array according to claim 1, characterized in that: In step S7, the coupled deformation safety assessment of the foundation pit and the protected object includes: establishing a coupled safety assessment system for the foundation pit and the protected object based on the full-domain deformation field constructed in step S5, the intelligent inference deformation prediction results embedded in the soft soil physical constraints in step S6, and the consistency judgment conclusions of visual perception data and traditional calibration points; automatically calculating and extracting sensitive indicators such as differential settlement of surrounding buildings, tilt rate, and additional strain of underground pipelines, and incorporating foundation pit retaining deformation, building safety, and pipeline safety into a unified system for joint assessment; and, in conjunction with the strict control requirements of the surrounding environment of soft soil foundation pits in the context of urban renewal, outputting graded early warning information according to risk level, and providing targeted construction control, support adjustment, and auxiliary decision-making suggestions for surrounding protection, thereby achieving full-cycle, highly reliable, and systematic safety perception and intelligent decision-making for deep soft soil foundation pits in urban renewal.

Images