Multi-equipment collaborative deformation monitoring and data processing methods in engineering survey

Abstract

This invention belongs to the field of deformation monitoring technology, specifically relating to a multi-device collaborative deformation monitoring and data processing method in engineering surveying. The method includes: acquiring observation data from target monitoring elements collected by multiple observation devices and generating observation constraint factors; constructing a closed-loop constraint graph based on the observation constraint factors and the feature point distance constraints of the target monitoring elements, calculating the pose residuals of the closed-loop path, and obtaining closed-loop residual features; decomposing the residual components of actual deformation, equipment drift, reference drift, and target monitoring element anomalies based on the closed-loop residual features, generating residual attribution results; determining deformation risk results based on the residual attribution results and deformation coordination constraints; and, when multiple candidate attribution results exist, determining the target re-measurement action and updating the deformation risk results using the re-measurement data. This invention can distinguish multiple sources of anomalies, improving the accuracy and traceability of deformation risk assessment.

Description

Technical Field

[0001] This invention relates to the field of deformation monitoring technology, specifically to a multi-device collaborative deformation monitoring and data processing method in engineering surveying. Background Technology

[0002] Reliable identification of minute deformations of objects such as the ground surface, slopes, foundation pit support, tunnel cross-sections, and beam structures during the engineering survey phase is crucial for assessing the stability of geological bodies, the stress state of structures, and the risks of subsequent construction. With the increasing use of equipment such as visual measurement, UAV photogrammetry, total stations, GPS receivers, and terrestrial laser scanning in the field, monitoring data sources are becoming more abundant, but the relationships between these data points are also becoming more complex.

[0003] Existing methods typically use a single benchmark point or a few stable points as coordinate constraints, then weighted and fused or differentially judged the displacement results obtained from different devices. While this approach can generate monitoring results in normal scenarios, it easily mixes errors from different sources into a single displacement value when there are simultaneous issues such as on-site obstruction, equipment attitude drift, slight movement of the candidate benchmark area, target loosening, or actual deformation. Especially under extreme conditions, if the benchmark point drifts slowly, or a certain observation device experiences systematic deviation, traditional weighted fusion will spread the error to the overall result; if the monitored target itself experiences local loosening, it may also be misjudged as deformation of the engineering object. This leads to problems such as unstable risk level assessment, lack of specificity in re-measurement actions, and difficulty in tracing the source of anomalies, affecting the reliability of the survey conclusions and the efficiency of subsequent handling. Summary of the Invention

[0004] This invention provides a multi-device collaborative deformation monitoring and data processing method in engineering surveying, which is used to at least solve the problem of how to accurately attribute and output reliable deformation risk results when multi-device observation errors, benchmark drift, monitoring element anomalies and actual deformation coexist.

[0005] This invention provides a method for multi-equipment collaborative deformation monitoring and data processing in engineering surveying, the method comprising: The target monitoring element is obtained by acquiring observation data from multiple observation devices, and observation constraint factors are generated based on the observation data. The target monitoring element is an asymmetric three-dimensional coded monitoring element that includes three non-collinear feature points. A closed-loop constraint graph is constructed based on the observation constraint factor and the distance constraint between three non-collinear feature points. The pose residual of the closed-loop path is calculated to obtain the closed-loop residual characteristics. Based on the characteristics of the closed-loop residuals, the residual sources are decomposed to obtain the residual components corresponding to the anomalies of the actual deformation, equipment drift, reference drift and target monitoring elements, respectively, and the residual attribution results are generated. Deformation risk results are determined based on residual attribution results and deformation coordination constraints. When the residual attribution result includes multiple candidate attribution results, the target retest action is determined based on the discriminativeness of the candidate retest actions, the retest data corresponding to the target retest action is obtained, and the closed-loop residual characteristics, residual attribution results and deformation risk results are updated based on the retest data, and the deformation risk results are output.

[0006] In one possible implementation, the target monitoring primitive also includes an identification coding area and a direction identifier. The identification coding area is used to determine the primitive identifier of the target monitoring primitive, and the direction identifier and three non-collinear feature points are used to determine the attitude orientation of the target monitoring primitive.

[0007] In one possible implementation, generating observation constraint factors based on observation data includes: identifying the identification coding area and extracting the observation positions of three non-collinear feature points; and generating pose observation constraint factors corresponding to the target monitoring primitive based on the primitive identifier, the observation position, and the calibration parameters of the observation equipment that collects observation data containing the observation position.

[0008] In one possible implementation, a closed-loop constraint graph is constructed based on the observation constraint factor and the distance constraint between three non-collinear feature points. This includes: using the target monitoring primitive, multiple observation devices, and candidate reference areas for determining reference drift as graph nodes, and using the observation constraint factor, the distance constraint between the three non-collinear feature points, and the relative position constraint between the candidate reference areas as graph edges to construct the closed-loop constraint graph.

[0009] In one possible implementation, calculating the pose residual of the closed-loop path includes: determining the target closed-loop path in the closed-loop constraint graph; determining multiple relative pose observations based on the observation constraint factors in the target closed-loop path; combining the multiple relative pose observations according to the connection order of the target closed-loop path to obtain the closed-loop pose transformation; and determining the pose residual of the closed-loop path based on the difference between the closed-loop pose transformation and the unit pose transformation.

[0010] In one possible implementation, the closed-loop residual features include device association features, spatial neighborhood features, feature point distance features, and closed-loop type features. The device association features are used to indicate the observation devices that participate in forming the pose residuals of the closed-loop path among multiple observation devices. The spatial neighborhood features are used to indicate the spatial distribution range of the pose residuals of the closed-loop path. The feature point distance features are used to indicate the distance variation between three non-collinear feature points. The closed-loop type features are used to indicate the source of constraints corresponding to the closed-loop path.

[0011] In one possible implementation, residual source decomposition is performed based on closed-loop residual characteristics, including: determining the residual component corresponding to the actual deformation based on spatial neighborhood characteristics; determining the residual component corresponding to equipment drift based on equipment association characteristics; determining the residual component corresponding to the reference drift based on closed-loop type characteristics; and determining the residual component corresponding to the anomaly of the target monitoring element based on feature point distance characteristics.

[0012] In one possible implementation, the deformation coordination constraints include at least one of the following: geological block displacement continuity constraints, support structure horizontal displacement curve constraints, tunnel cross-section convergence constraints, and beam deflection curve constraints; the deformation risk results are determined based on the residual attribution results and the deformation coordination constraints, including: determining the corresponding deformation coordination constraints according to the type of engineering survey object, and determining the deformation risk results based on the matching results between the residual attribution results and the corresponding deformation coordination constraints.

[0013] In one possible implementation, the target retest action is determined based on the discriminative power of the candidate retest actions, including: for each candidate retest action, determining the predicted observations corresponding to multiple candidate attribution results when executing the candidate retest action; determining the discriminative power of the candidate retest action based on the differences between the predicted observations corresponding to different candidate attribution results, the observation noise of the candidate retest action, and the execution cost; and determining the candidate retest action with the highest discriminative power as the target retest action.

[0014] In one possible implementation, updating the closed-loop residual features, residual attribution results, and deformation risk results based on the retest data includes: generating retest observation constraint factors based on the retest data; adding the retest observation constraint factors to the closed-loop constraint graph; calculating the pose residual of the closed-loop path associated with the target retest action and updating the closed-loop residual features; and updating the residual attribution results and deformation risk results based on the updated closed-loop residual features.

[0015] Compared with the prior art, the advantages and beneficial effects of the present invention are as follows: By employing an asymmetric three-dimensional coded monitoring primitive including three non-collinear feature points, the monitoring object is transformed from a single measuring point into a rigid observation object with identifiable identity, direction, and internal distance constraints, which can distinguish between the anomalies of the monitoring primitive itself and the actual deformation of the monitored object.

[0016] By constructing a closed-loop constraint graph based on observation constraint factors and feature point distance constraints, a unified expression of multi-device observation relationships, primitive internal geometric relationships, and candidate reference area relative relationships is achieved, enabling the source of observation anomalies to be traced along the closed-loop path.

[0017] By calculating the pose residual of the closed-loop path and generating closed-loop residual features, the simple numerical residual is transformed into a structured judgment basis that includes device association, spatial neighborhood, feature point distance and closed-loop type.

[0018] By decomposing the residual sources based on the characteristics of the closed-loop residuals, the system can separately identify the actual deformation, equipment drift, reference drift, and target monitoring element anomalies, thus avoiding the direct misinterpretation of system errors as engineering deformation.

[0019] By combining deformation coordination constraints to determine deformation risk results, a matching judgment between residual attribution results and deformation patterns of engineering objects was achieved.

[0020] By selecting the target retest action based on the discrimination of multiple candidate attribution results, the retest resources are concentrated on the observation action that can best distinguish the source of the anomaly. The retest data is used to update the closed-loop residual characteristics, residual attribution results and deformation risk results, so that the risk output has stronger traceability and stability. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the process of the present invention; Figure 2 This is a schematic diagram of the monitoring deployment in an embodiment of the present invention; Figure 3 This is a diagram showing the change in closed-loop pose residuals in an embodiment of the present invention. Figure 4 This is a diagram showing the decomposition results of residual sources in an embodiment of the present invention; Figure 5 This is a comparison diagram of the displacement profile before and after the remeasurement in an embodiment of the present invention. Detailed Implementation

[0022] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the invention.

[0023] Multi-device collaborative deformation monitoring is not simply a matter of summarizing displacement data from multiple observation devices. Instead, it requires establishing verifiable constraints between measurement results from different observation scales, coordinate systems, and error characteristics. Visual observation provides information on local target attitude and feature point changes; UAV observation supplements large-scale spatial relationships; ground-based laser scanning reflects surface geometry; and total stations or GPS receivers provide relative positions or reference relationships. Only when the observation results from various devices undergo identity matching, pose association, closed-loop verification, and anomaly attribution within the same constraint system can the true deformation, device errors, reference drift, and anomalies of the monitored target be distinguished from the multi-source observation data. Based on this understanding, this invention focuses on multi-device collaborative deformation monitoring and data processing in engineering surveying. It proposes a processing method that uses asymmetric three-dimensional coded monitoring primitives as the observation object, closed-loop constraint diagrams as the data organization basis, and residual source decomposition and re-measurement updates as the judgment mechanism.

[0024] like Figure 1 As shown, a method for multi-equipment collaborative deformation monitoring and data processing in engineering surveying includes: The target monitoring element is obtained by acquiring observation data from multiple observation devices, and observation constraint factors are generated based on the observation data. The target monitoring element is an asymmetric three-dimensional coded monitoring element that includes three non-collinear feature points. Several target monitoring elements are deployed within the engineering survey area, and multiple observation devices capable of covering these elements are configured. These observation devices may include at least two of the following types: fixed visual acquisition devices, UAV image acquisition devices, ground laser scanning devices, total stations, and global navigation satellite system receiving devices. The target monitoring elements are installed at slopes, support structures, tunnel cross-sections, beams, retaining walls, or surface settlement monitoring locations to form stable, observable objects. The target monitoring elements employ an asymmetric three-dimensional coded structure, including at least three non-collinear feature points, which maintain a fixed spatial relationship on the rigid substrate of the target monitoring element. Multiple observation devices collect observation data from the target monitoring elements within the same or adjacent monitoring cycles. The observation data includes image data, point cloud data, distance observation data, angle observation data, or relative coordinate data. After receiving the observation data, the data processing terminal records the observation device identification, acquisition time, acquisition angle, and observation data type, and standardizes the data format. The data processing unit generates observation constraint factors based on the observation results of identifiable target monitoring primitives and feature points in the observation data. The observation constraint factors are used to describe the pose observation relationship between the target monitoring primitives and the observation equipment, and serve as the data input for the subsequent construction of the closed-loop constraint diagram.

[0025] The target monitoring unit also includes an identification coding area and a direction identifier. The identification coding area is used to determine the unit identifier of the target monitoring unit, and the direction identifier and three non-collinear feature points are used to determine the attitude orientation of the target monitoring unit.

[0026] In one embodiment, the target monitoring element is further equipped with an identification coding area and a direction identifier based on three non-collinear feature points. The identification coding area determines the element identifier of the target monitoring element, and the direction identifier, combined with the three non-collinear feature points, determines the orientation of the target monitoring element. The purpose of introducing the identification coding area is to enable the target monitoring elements at different monitoring positions to be automatically distinguished, avoiding object matching errors when multiple target monitoring elements appear in the same field of view. The identification coding area can be a QR code, a matrix coding pattern, or an engineering visual coding pattern with a unique number. The coding content includes at least the element number, and may further include the installation area number, installation surface number, and initial installation time. The identification coding area is set on the visible surface of the target monitoring element and maintains a fixed relative position with the three non-collinear feature points. After the data processing terminal completes decoding in the identification coding area, it binds the decoded element identifier with the feature point extraction results in the observation data.

[0027] The orientation marker is used to determine the local orientation of the target monitoring element, avoiding instability in attitude and orientation judgment due to changes in shooting angle or partial occlusion when relying solely on three non-collinear feature points. The orientation marker can be a directional bar mark, arrow mark, asymmetrical color block, or geometric pattern with directional boundaries. The orientation marker and the three non-collinear feature points together constitute the basis for determining the local coordinates of the target monitoring element. In specific implementation, the geometric center formed by the three non-collinear feature points is used as the local reference point of the target monitoring element, and the fixed line connecting two feature points is used as the local reference direction. The orientation marker then confirms the positive direction of this local reference direction. The three non-collinear feature points are not arranged in an equilateral triangle or other rotationally symmetric manner, but rather in a scalene triangle or a spatial arrangement with a clear main direction, enabling the data processing end to distinguish the rotational attitude of the target monitoring element.

[0028] During on-site installation, the rigid substrate of the target monitoring element is fixedly connected to the surface of the monitored component or geological body. The fixing method can be bolts, anchor blocks, adhesive bases, or welded supports, depending on the material of the monitored object. After installation, the initial distance between the three non-collinear feature points, the initial orientation of the direction indicator, and the element identifier in the identification coding area are recorded, forming the initial configuration record of the target monitoring element. This initial configuration record is used to subsequently determine whether the target monitoring element has become loose, damaged, or misidentified. If the identification coding area cannot be decoded, but the three non-collinear feature points can still be stably extracted, the data processing end marks this observation data as unconfirmed observation data and does not directly generate an observation constraint factor with a defined element identifier. If the decoding result of the identification coding area is inconsistent with the installation area record, or if the orientation given by the direction indicator conflicts with the local direction formed by the three non-collinear feature points, the data processing end marks this observation data as abnormal observation data and waits for cross-confirmation of the observation results from other observation equipment within the same monitoring cycle. With the above settings, the target monitoring primitive can not only provide the basis for position observation, but also provide the observation basis that can distinguish identity and attitude, providing a stable input for the generation of subsequent pose observation constraint factors.

[0029] The observation constraint factors are generated based on the observation data, including: identifying the identification coding area and extracting the observation positions of three non-collinear feature points; and generating the pose observation constraint factors corresponding to the target monitoring primitives based on the primitive identifier, the observation position, and the calibration parameters of the observation equipment that collects the observation data containing the observation position.

[0030] In one embodiment, when generating the observation constraint factor, the data processing end decodes the identification coding region in the observation data and extracts the observation positions of three non-collinear feature points. Then, based on the primitive identifier, the observation positions, and the calibration parameters of the observation equipment that collected the observation data containing the observation positions, the pose observation constraint factor corresponding to the target monitoring primitive is generated. This embodiment further defines the generation process of the observation constraint factor compared to the foregoing, making the observation constraint factor not just a simple record of the observation data, but structured data containing the identity of the target monitoring primitive, the observation positions of the feature points, and the calibration relationship of the observation equipment. The decoding result of the identification coding region is used to determine the primitive identifier, the observation positions of the three non-collinear feature points are used to determine the geometric observation relationship of the target monitoring primitive in the coordinate system of the observation equipment, and the calibration parameters of the observation equipment are used to convert the original observation positions into data that can participate in the pose constraint calculation.

[0031] For image-based observation data, the data processing unit locates and identifies the coded region in the image and reads the primitive identifiers. After decoding the coded region, the data processing unit searches for three non-collinear feature points in the neighborhood of the coded region and determines the image observation position of the feature points using edge extraction, gray-level centroid, corner detection, or circular marker detection methods. The image observation position of the feature points can be recorded in pixel coordinates, along with the feature point extraction quality. The feature point extraction quality can be determined based on the contrast of the feature point region, edge integrity, positioning residual, and occlusion ratio. When all three non-collinear feature points are extracted, and the relative image relationships between the three non-collinear feature points match the initial configuration record, the data processing unit calls the intrinsic, extrinsic, and distortion correction parameters of the observation equipment that acquired the image to generate a pose observation constraint factor. The pose observation constraint factor includes at least the primitive identifier, observation equipment identifier, acquisition time, observation position of the three non-collinear feature points, observation equipment calibration parameter identifier, and observation quality identifier.

[0032] For point cloud observation data, the data processing unit extracts the spatial observation positions of three non-collinear feature points based on the geometric structure or reflection intensity differences on the surface of the target monitoring element, and determines the element identifier based on the identification coding area on the target monitoring element or a pre-established spatial index. If the identification coding area cannot be directly read from the point cloud data, the data processing unit can perform candidate matching based on the spatial distance between feature points in the point cloud, the installation area of ​​the target monitoring element, and the scanning station position of the observation equipment. After successful matching, a pose observation constraint factor with matching confidence is generated. For observation data generated by total station or global navigation satellite system receiving equipment, the data processing unit converts the distance, angle, or relative coordinates into the observation position of the local reference point of the target monitoring element, and writes the station parameters, antenna height parameters, or prism constant of the observation equipment as part of the calibration parameters into the observation constraint factor.

[0033] Before generation, observation constraint factors need to undergo consistency verification. Verification includes checking if the primitive identifier exists in the initial configuration record, whether all three non-collinear feature points have valid observation positions, whether the ratio of observation distances between feature points is within the allowable deviation range, and whether the calibration parameters of the observation equipment match the acquisition time. The allowable deviation range is determined based on the manufacturing and installation errors of the target monitoring primitive and the nominal accuracy of the observation equipment; it can be set as a proportional deviation of the initial distance or a fixed length deviation. If only two of the three non-collinear feature points are extracted, the data processing end does not generate complete pose observation constraint factors but instead generates supplementary observation records. If the observation equipment calibration parameters are missing or the calibration time exceeds the set validity period, the data processing end marks the corresponding observation data as invalid observation data. The generated pose observation constraint factors are added to the observation constraint factor set and indexed by primitive identifier, observation equipment identifier, and acquisition time for subsequent use in constructing the closed-loop constraint graph.

[0034] A closed-loop constraint graph is constructed based on the observation constraint factor and the distance constraint between three non-collinear feature points. The pose residual of the closed-loop path is calculated to obtain the closed-loop residual characteristics. After obtaining the observation constraint factors, the data processing end integrates the target monitoring primitive, multiple observation devices, and candidate reference areas into a unified graph structure for management. The target monitoring primitive serves as the observed object node, the observation device as the observation source node, and the candidate reference area as the reference region node used to determine reference drift. The data processing end establishes graph edges between the observation device and the target monitoring primitive based on the observation constraint factors, establishes distance constraint graph edges within the target monitoring primitive based on the initial distance and current observation distance between three non-collinear feature points, and establishes reference relative position graph edges based on the relative positional relationships between candidate reference areas. After the closed-loop constraint graph is constructed, the data processing end searches the graph structure for closed-loop paths that can return to the starting node, sequentially combines the pose observation relationships within the closed-loop path, and compares the combination result with the unit pose transformation to obtain the pose residual of the closed-loop path. The pose residual is further organized into closed-loop residual features for subsequent residual source decomposition.

[0035] A closed-loop constraint graph is constructed based on the observation constraint factor and the distance constraint between three non-collinear feature points. This includes: using the target monitoring primitive, multiple observation devices, and candidate reference areas used to determine reference drift as graph nodes, and using the observation constraint factor, the distance constraint between the three non-collinear feature points, and the relative position constraint between candidate reference areas as graph edges to construct the closed-loop constraint graph.

[0036] In one embodiment, the construction of the closed-loop constraint graph is further limited to a structured generation process of nodes and graph edges. This limitation is introduced because the observation data from multiple devices come from different sources. If only a single observation result is used as the basis for subsequent calculations, device drift, target monitoring element anomalies, and baseline drift can easily appear mixed in the same residual, making them difficult to separate in subsequent steps. The data processing end sets target monitoring elements, multiple observation devices, and candidate baseline regions used to determine baseline drift as graph nodes, giving each type of object an independent identity and a traceable position in the graph structure. Target monitoring element nodes at least record the element identifier, the initial distance between three non-collinear feature points, the installation area, the current visibility status, and the observation quality identifier; observation device nodes at least record the device identifier, device type, calibration parameter index, acquisition time range, and station status; candidate baseline region nodes at least record the candidate baseline region number, the baseline region spatial range, the set of target monitoring elements participating in the observation, and the relative position record between the candidate baseline regions and other candidate baseline regions.

[0037] The graph edges are generated according to the source of the constraints. Observation constraint factors form observation graph edges, which connect observation equipment nodes and target monitoring primitive nodes, expressing the pose observation relationship formed by the observation equipment on the target monitoring primitive. Distance constraints between three non-collinear feature points form primitive internal graph edges, which express whether the rigid structure of the target monitoring primitive itself remains stable. Relative position constraints between candidate reference areas form reference relative position graph edges, which record the relative distance, relative direction, or relative elevation changes between different candidate reference areas in adjacent monitoring cycles. Candidate reference areas are not fixed absolute references, but rather reference areas participating in reference drift judgment. The data processing end determines whether there is overall drift based on the internal geometric stability of the candidate reference area and the relative position changes between candidate reference areas.

[0038] The data processing end synchronously writes graph edge attributes when generating graph edges. Graph edge attributes include constraint source, acquisition time, observation equipment type, observation quality, constraint weight, and associated nodes. Observation quality can be determined based on feature point localization residuals in the image, point cloud matching residuals, total station observation accuracy, or GNS positioning quality. Constraint weights are not used to directly replace residual source decomposition, but rather to indicate the reliability of the graph edge in closed-loop residual calculation. If the observation data corresponding to a certain observation graph edge has occlusion, expired calibration, inconsistent acquisition time, or incomplete feature point extraction, the data processing end marks the observation graph edge as a low-confidence graph edge, does not directly delete the graph edge, but reduces its participation priority during closed-loop path search. Through the above method, each node and each graph edge in the closed-loop constraint graph has a clear source and purpose, and the pose residual of subsequent closed-loop paths can be traced back to specific target monitoring primitives, specific observation equipment, and specific candidate reference areas.

[0039] Calculating the pose residual of the closed-loop path includes: determining the target closed-loop path in the closed-loop constraint graph; determining multiple relative pose observations based on the observation constraint factors in the target closed-loop path; combining the multiple relative pose observations according to the connection order of the target closed-loop path to obtain the closed-loop pose transformation; and determining the pose residual of the closed-loop path based on the difference between the closed-loop pose transformation and the unit pose transformation.

[0040] In one embodiment, the pose residual of the closed-loop path is obtained by combining multiple relative pose observations in the target closed-loop path in sequence. This constraint point further clarifies the calculation boundary of the pose residual, ensuring a definite relationship between the observation direction, connection sequence, and unit pose transformation within the same closed-loop path. When searching for the target closed-loop path in the closed-loop constraint graph, the data processing terminal selects a path that starts from the target monitoring primitive node, passes through at least one observation device node or candidate reference area node, and returns to the starting target monitoring primitive node. The target closed-loop path can be a cross-device closed loop formed between the target monitoring primitive, the observation device, and another target monitoring primitive; it can also be a reference closed loop formed between multiple target monitoring primitives and candidate reference areas; or it can be a spatial geometric closed loop formed between the observation device, the target monitoring primitive, and the point cloud patch.

[0041] For each observation edge in the target closed-loop path, the data processing unit determines the relative pose observation based on the observation constraint factor. The relative pose observation is used to express the translation and attitude relationships from the current node coordinate system to the next node coordinate system. The relative pose observation corresponding to image-type observation data can be calculated from the observation positions of feature points of the target monitoring primitives and the calibration parameters of the observation equipment; the relative pose observation corresponding to point cloud-type observation data can be obtained from the local point cloud registration results; and the relative pose observation corresponding to total station or global navigation satellite system observation data can be obtained by converting relative coordinates, distance, and azimuth. If the direction of a certain edge in the target closed-loop path is opposite to the direction of a stored relative pose observation, the data processing unit performs an inverse transformation on that relative pose observation, ensuring that all relative pose observations are arranged according to the connection order of the target closed-loop path.

[0042] Closed-loop pose transformation can be expressed by the following formula:

[0043] Where is the pose residual vector of the closed-loop path. For the target closed-loop path, the first The relative pose observations corresponding to the edges of the bar graph. Number the edges of the graph sequentially. This represents the number of graph edges contained in the target closed-loop path. This formula maps pose transformation to translational and rotational residuals. The input to the formula is multiple relative pose observations along the target's closed-loop path, and the output is a pose residual vector representing the closed-loop deviation. When multiple relative pose observations are identical, the closed-loop pose transformation obtained by sequentially combining them approximates a unit pose transformation, and the pose residual vector approaches zero. When observation equipment drift, candidate reference area drift, target monitoring element anomalies, or actual deformation occurs, the closed-loop pose transformation deviates from a unit pose transformation, and the corresponding components in the pose residual vector change.

[0044] After obtaining the pose residual vector, the data processing unit decomposes it into translational and rotational residual components, and records the target closed-loop path, participating graph edges, participating observation devices, and associated target monitoring primitives corresponding to the pose residual vector. If the number of low-confidence graph edges in the target closed-loop path exceeds a set number, the data processing unit marks the pose residual as a reference residual and does not use it as the basis for decomposing the main residual source. The set number is determined based on the total number of graph edges in the target closed-loop path; for example, the number of low-confidence graph edges can be required to not exceed half of the total number of graph edges. If the target closed-loop path lacks invertible transformation relationships, or if the relative pose observations cannot form a complete connection, the data processing unit does not calculate the pose residual of that path and marks the path as an incomplete path. Thus, the pose residual of the closed-loop path can reflect both the degree of inconsistency in closed-loop observations and retain the observation sources that generate the residuals.

[0045] The closed-loop residual features include device association features, spatial neighborhood features, feature point distance features, and closed-loop type features. Device association features are used to indicate the observation devices that participate in forming the pose residuals of the closed-loop path among multiple observation devices. Spatial neighborhood features are used to indicate the spatial distribution range of the pose residuals of the closed-loop path. Feature point distance features are used to indicate the distance variation between three non-collinear feature points. Closed-loop type features are used to indicate the source of constraints corresponding to the closed-loop path.

[0046] In one embodiment, the closed-loop residual features consist of device association features, spatial neighborhood features, feature point distance features, and closed-loop type features. This constraint point is used to transform the pose residual from a single numerical value into a structured feature capable of participating in residual source decomposition, avoiding subsequent judgment of true deformation solely based on residual magnitude. The device association features record the observation devices involved in forming the pose residual of the closed-loop path among multiple observation devices. Specifically, the data processing end extracts the observation device identifier, device type, acquisition time, and device calibration status from the observation map edges in the target closed-loop path and writes this information into the device association features. If multiple pose residuals are concentrated in the closed-loop path involving the same observation device, while closed-loop paths involving other observation devices do not show the same type of residual, then the device association features provide a basis for subsequent judgment of device drift.

[0047] Spatial neighborhood features are used to indicate the spatial distribution range of pose residuals in closed-loop paths. The data processing unit determines whether the pose residuals are concentrated in a single target monitoring unit, a local area, or a continuous area based on the installation area of ​​the target monitoring units associated with the target closed-loop path, their adjacency relationships, and the surface location of the engineering object. If the pose residuals are concentrated only in a single target monitoring unit, and the closed-loop paths corresponding to surrounding target monitoring units remain stable, the spatial neighborhood features are recorded as a point distribution. If the pose residuals appear continuously among multiple adjacent target monitoring units, the spatial neighborhood features are recorded as a local distribution. If the pose residuals continuously extend along the slope, support structure, tunnel cross-section, or beam direction, the spatial neighborhood features are recorded as a continuous distribution. The range of spatial neighborhood features is determined based on the spatial distance between target monitoring units, the structural zoning of the engineering object, and the grid division of the monitoring area. Adjacency distances can be set according to the size of the engineering survey object and the density of target monitoring unit deployment; fixed values ​​unrelated to the engineering scale should not be used.

[0048] The feature point distance feature is used to indicate the distance changes between three non-collinear feature points. The data processing unit compares the observed distances between the three non-collinear feature points in the current monitoring cycle with the distances recorded in the initial configuration to form the feature point distance feature. If the distances between the three non-collinear feature points all remain within the allowable deviation range, the rigid structure of the target monitoring element is considered to remain stable; if any distance exceeds the allowable deviation range, the data processing unit records the feature point pair with the distance anomaly and writes this information into the feature point distance feature. The allowable deviation range is determined based on the manufacturing error of the target monitoring element, the installation method, and the accuracy of the observation equipment, and can be set during system initialization based on the field calibration results. This feature is used to distinguish between abnormalities in the target monitoring element itself and the actual deformation of the monitored object, preventing loosening or damage of the target monitoring element from being mistaken for displacement of the engineering object.

[0049] The closed-loop type feature indicates the constraint source corresponding to the closed-loop path. The data processing end classifies the closed-loop path into cross-device closed loops, baseline closed loops, spatial geometric closed loops, or hybrid closed loops based on the graph edge type in the target closed-loop path. Cross-device closed loops are mainly formed by observations of the same target monitoring element or adjacent target monitoring elements by different observation devices; baseline closed loops are mainly formed by relative position constraints between candidate baseline areas; spatial geometric closed loops are mainly formed by internal distance constraints of target monitoring elements and spatial relationships between adjacent target monitoring elements; hybrid closed loops include multiple constraint sources mentioned above. The closed-loop type feature, along with device association features, spatial neighborhood features, and feature point distance features, are written into the closed-loop residual feature record. Each closed-loop residual feature record includes at least the closed-loop path number, pose residual vector, participating nodes, participating graph edges, device association features, spatial neighborhood features, feature point distance features, and closed-loop type feature. This record enters the residual source decomposition stage to distinguish between true deformation, device drift, baseline drift, and target monitoring element anomalies.

[0050] Based on the characteristics of the closed-loop residuals, the residual sources are decomposed to obtain the residual components corresponding to the anomalies of the actual deformation, equipment drift, reference drift and target monitoring elements, respectively, and the residual attribution results are generated. After obtaining the closed-loop residual features, the data processing unit establishes a source decomposition record for each feature. This record includes the closed-loop path number, pose residual vector, spatial neighborhood features, equipment association features, closed-loop type features, and feature point distance features. The data processing unit decomposes the pose residuals into four sources: actual deformation generated by the engineering object, equipment drift generated by the observation equipment, reference drift generated by the candidate reference area, and anomalies generated by the target monitoring element itself. During the decomposition process, spatial neighborhood features are used to identify continuously distributed deformation residuals, equipment association features are used to identify drift residuals concentrated on the same observation equipment, closed-loop type features are used to identify drift residuals related to the candidate reference area, and feature point distance features are used to identify rigid structural anomalies of the target monitoring element. After the four residual components are determined, the data processing unit generates residual attribution results, which then proceed to the deformation risk assessment stage.

[0051] The residual sources are decomposed based on the closed-loop residual characteristics, including: determining the residual components corresponding to the actual deformation based on spatial neighborhood characteristics; determining the residual components corresponding to equipment drift based on equipment association characteristics; determining the residual components corresponding to the reference drift based on closed-loop type characteristics; and determining the residual components corresponding to the anomalies of the target monitoring elements based on feature point distance characteristics.

[0052] In one embodiment, residual source decomposition is further defined as a process of determining four types of residual components based on four types of closed-loop residual features. This limitation is introduced because the pose residual of a single closed-loop path only indicates that the observation relationship is not closed; it cannot directly indicate whether the residual originates from actual deformation, equipment drift, reference drift, or target monitoring element anomaly. The data processing unit aggregates the closed-loop residual features within the same monitoring period into a residual feature set and establishes four types of residual candidate channels for each closed-loop residual feature. Each type of residual candidate channel records the judgment criteria for the corresponding source, the residual magnitude, participating nodes, and the confidence status, avoiding the mixing of residuals from different sources in the same value.

[0053] The residual decomposition relation can be expressed as:

[0054] in, The composite residual corresponding to the closed-loop residual characteristics; These are the residual components corresponding to the actual deformation; This refers to the residual components corresponding to equipment drift. The residual components corresponding to the baseline drift; The residual components corresponding to the anomalies of the target monitoring primitives; This represents random observation error. The formula describes the data relationships in the residual source decomposition. The inputs are the closed-loop residual features and the pose residual vector; the outputs are four types of residual components and random observation error. Random observation error is not output as an independent attribution result; it is only used to accommodate observation noise, feature point localization fluctuations, and minor deviations in point cloud matching.

[0055] The data processing unit determines the residual components corresponding to the actual deformation based on spatial neighborhood characteristics. If the pose residual of the closed-loop path appears continuously among multiple adjacent target monitoring elements, and this continuous distribution is consistent with the spatial orientation of the engineering object, the data processing unit classifies this part of the residual into the candidate channel for actual deformation. For slopes or geological bodies, a continuous distribution can be manifested as similar displacement directions between adjacent blocks; for support structures, a continuous distribution can be manifested as segmental changes along the vertical or horizontal direction; for tunnel cross-sections, a continuous distribution can be manifested as a convergence trend of the same cross-section or adjacent cross-sections. If the pose residual is concentrated only in a single target monitoring element, and the surrounding target monitoring elements do not show the same trend, this residual is not preferentially classified into the candidate channel for actual deformation.

[0056] The data processing unit determines the residual components corresponding to equipment drift based on the equipment's associated characteristics. If multiple closed-loop paths all contain the same observation device, and the pose residuals of these closed-loop paths have similar directions or similar amplitudes of change, while closed-loop paths not containing the same observation device remain stable, the data processing unit will classify the corresponding residuals into the equipment drift candidate channel. The equipment drift candidate channel is also verified in conjunction with the calibration status of the observation device, the acquisition time, and the station records. If the equipment calibration parameters exceed their validity period, or if the same observation device introduces similar residuals simultaneously in multiple spatial regions, the reliability of the equipment drift candidate channel is improved.

[0057] The data processing unit determines the residual components corresponding to the baseline drift based on the characteristics of the closed-loop type. If the anomaly mainly occurs in the closed-loop path containing the relative position constraints of the candidate baseline region, and the relative relationships of the target monitoring primitives within the candidate baseline region remain stable, the data processing unit will classify the corresponding residuals into the baseline drift candidate channel. The candidate baseline region is not directly regarded as a fixed and unchanging area, but is judged based on relative position constraints and internal stable state. If the relative position of the candidate baseline region to the outside continuously changes, but the distance between feature points, local point cloud patches, and relationships between adjacent target monitoring primitives within the candidate baseline region do not change significantly, this change is recorded as the main basis for baseline drift.

[0058] The data processing unit determines the residual components corresponding to anomalies in the target monitoring element based on the distance characteristics of feature points. If the current observation distance between three non-collinear feature points exceeds the allowable deviation range, or if different observation devices all show anomalies in the internal distance judgment of the same target monitoring element, the data processing unit classifies the residual associated with that target monitoring element into the target monitoring element anomaly candidate channel. The allowable deviation range is set based on the manufacturing error of the target monitoring element, the installation and fixing method, and the positioning accuracy of the observation equipment. For single distance anomalies caused by temporary occlusion, the data processing unit marks the anomaly status as pending confirmation; for cases where distance anomalies exist in multiple consecutive monitoring cycles, the data processing unit removes the target monitoring element from the main basis for judging actual deformation, but retains its anomaly record for use in the retesting stage. After the four types of residual candidate channels are updated, the data processing unit compares the credibility status and residual proportion of each channel to generate residual attribution results.

[0059] Deformation risk results are determined based on residual attribution results and deformation coordination constraints. After obtaining the residual attribution results, the data processing terminal reads the type information of the engineering survey object and calls the corresponding deformation coordination constraints based on the type information. The type information of the engineering survey object can be determined by the project configuration file, monitoring area labels, or field modeling data, and includes at least one of the following: geological body, support structure, tunnel cross-section, and beam structure. The residual attribution results record the residual components and their spatial distribution of actual deformation, equipment drift, benchmark drift, and target monitoring element anomalies. The data processing terminal does not directly use the actual deformation residual components as risk conclusions, but rather matches the actual deformation residual components with the deformation coordination constraints corresponding to the engineering survey object. If the residual attribution results and the corresponding constraints meet the matching conditions in terms of spatial location, direction of change, continuity relationship, and temporal trend, the data processing terminal uses it as valid deformation evidence; if the residual attribution results and the corresponding constraints do not match, the data processing terminal transfers it to a pending review state. The deformation risk results include the risk area, risk type, risk level, and formation basis, which are used for subsequent retesting or early warning output.

[0060] Deformation coordination constraints include at least one of the following: geological block displacement continuity constraints, support structure horizontal displacement curve constraints, tunnel cross-section convergence constraints, and beam deflection curve constraints; the deformation risk results are determined based on the residual attribution results and deformation coordination constraints, including: determining the corresponding deformation coordination constraints according to the type of engineering survey object, and determining the deformation risk results based on the matching results between the residual attribution results and the corresponding deformation coordination constraints.

[0061] In one embodiment, deformation coordination constraints are selected according to the type of engineering survey object, and each constraint corresponds to specific input data, judgment boundaries, and matching methods. This constraint is introduced because the deformation patterns of different engineering objects vary significantly. Simply determining risk based on the magnitude of residual components can easily lead to misinterpreting equipment drift, localized observation anomalies, or loosening of target monitoring elements as engineering risks. During the project initialization phase, the data processing end establishes an engineering survey object type table. This table records the monitoring area number, engineering object type, target monitoring element placement location, adjacency relationship, allowable deformation direction, and the source of the risk judgment threshold. The threshold source can be engineering design documents, monitoring specifications, initial on-site observation statistics, or configuration values ​​from the project management end. If no threshold is provided in the project configuration file, the data processing end uses the observation fluctuation range during the initial stable monitoring period as a reference and sets conservative boundaries in conjunction with the engineering safety level.

[0062] For geological bodies or unstable rock masses, the data processing terminal invokes the geological body block displacement continuity constraint. This constraint treats the areas where adjacent target monitoring elements are located as several block units, determining whether the true deformation residual components within the same block unit maintain continuity in displacement direction and amplitude. If there are no crack boundaries, patch separation, or significant terrain abrupt changes between adjacent target monitoring elements, the displacement direction within the same block unit should not change abruptly; if there are displacement differences on both sides of a crack boundary, discontinuity of the residual components on both sides of the boundary is allowed. The data processing terminal determines the block boundary based on spatial neighborhood relationships, point cloud patch boundaries, and crack lines marked on-site. If the true deformation residual components continuously extend along the slope direction, and the proportion of residual components with reference drift and target monitoring element anomalies is lower than a set ratio, the data processing terminal identifies the corresponding area as a geological body deformation risk area; the set ratio is determined based on the engineering safety level, the residual ratio during the initial stable monitoring period, and the nominal accuracy of the observation equipment; the higher the engineering safety level, the lower the set ratio.

[0063] For foundation pit support structures, retaining walls, or similar vertical structures, the data processing terminal invokes the horizontal displacement curve constraint of the support structure. This constraint arranges target monitoring elements on the same vertical profile or the same structural axis in order of elevation or mileage, forming a sequence of horizontal displacement changes. The data processing terminal checks whether the actual deformation residual components can form a continuous horizontal displacement curve. If the displacement of a single target monitoring element suddenly increases, but adjacent target monitoring elements above and below or left and right do not form the same trend, the data processing terminal does not directly output the deformation risk of the support structure, but instead checks the residual components corresponding to the target monitoring element anomalies and equipment drift. If multiple adjacent target monitoring elements form a continuous bending trend along the vertical direction, and the equipment drift residual components cannot explain this trend, the data processing terminal uses this trend as evidence of support structure deformation.

[0064] For tunnels, utility tunnels, or underground caverns, the data processing terminal invokes tunnel cross-section convergence constraints. These constraints use the monitoring positions of the crown, waist, sidewalls, and invert within a single cross-section as the judgment objects, checking whether the actual deformation residual components of the target monitoring elements within the cross-section conform to convergence, settlement, or uplift relationships. If both sidewalls within the same cross-section shift towards the cross-section center, and the crown settlement component appears simultaneously, the data processing terminal uses this as a basis for assessing cross-section convergence risk. If only a single target monitoring element exhibits abnormal displacement, and the feature point distance characteristics indicate that the target monitoring element itself has an anomaly, the data processing terminal marks this result as a non-structural anomaly.

[0065] For bridges, trestle bridges, or beam-type components, the data processing terminal invokes the beam deflection curve constraint. This constraint, based on the arrangement order of target monitoring elements along the beam's length, determines whether the actual deformation residual components can form a continuous deflection change. The residual components in the mid-span region, support region, and pier top region participate in curve matching respectively. If a continuous downward deflection trend appears in the mid-span region, and no baseline drift that can explain all changes appears at the support or pier top monitoring positions, the data processing terminal outputs the beam deflection risk result. If the degree of matching between the residual attribution result and the corresponding deformation coordination constraint reaches the risk judgment condition, the deformation risk result records the risk level; the risk judgment condition is determined based on cumulative displacement, rate of change, number of consecutive monitoring periods, and constraint matching degree. The larger the cumulative displacement, the higher the rate of change, the more consecutive periods, and the higher the constraint matching degree, the higher the risk level. If the matching degree does not reach the risk judgment condition, but the residual attribution result persists, the data processing terminal outputs a result pending review and submits the corresponding area to the retesting process.

[0066] When the residual attribution result includes multiple candidate attribution results, the target retest action is determined based on the discriminativeness of the candidate retest actions, the retest data corresponding to the target retest action is obtained, and the closed-loop residual characteristics, residual attribution results and deformation risk results are updated based on the retest data, and the deformation risk results are output.

[0067] When multiple candidate attribution results exist in the residual attribution results, the data processing end does not directly output a unique risk conclusion, but instead enters the retest selection process. Multiple candidate attribution results can include at least two of the following: actual deformation, equipment drift, baseline drift, and target monitoring element anomalies. The data processing end generates candidate retest actions based on the candidate attribution results. These actions can include changing the observation equipment's perspective, increasing target area scanning, supplementing the relative position of the candidate baseline area, or re-acquiring local observation data of the target monitoring element. The data processing end estimates the predicted observation values ​​that each candidate retest action may obtain under different candidate attribution results and calculates the discriminant score by combining observation noise and execution cost. The candidate retest action with the highest discriminant score is determined as the target retest action. After the target retest action is executed, retest data is obtained. The retest data generates retest observation constraint factors and writes them into the closed-loop constraint graph. The data processing end recalculates the pose residuals of the associated closed-loop paths, updates the closed-loop residual characteristics, residual attribution results, and deformation risk results, and outputs the updated deformation risk results.

[0068] The target retest action is determined based on the discriminative power of the candidate retest actions, including: for each candidate retest action, determining the predicted observations corresponding to multiple candidate attribution results when executing the candidate retest action; determining the discriminative power of the candidate retest action based on the differences between the predicted observations corresponding to different candidate attribution results, the observation noise of the candidate retest action, and the execution cost; and determining the candidate retest action with the highest discriminative power as the target retest action.

[0069] In one embodiment, the discriminative power of candidate retest actions is determined based on the predicted observations corresponding to multiple candidate attribution results. This constraint is used to prevent retest actions from being executed solely based on human experience or a fixed sequence. After generating multiple candidate attribution results, the data processing end establishes a predicted observation state for each candidate attribution result. The predicted observation state is jointly determined by the current closed-loop residual characteristics, the residual source decomposition results, the position of the participating target monitoring primitives, the type of participating observation equipment, and the observation method of the candidate retest actions. If the candidate attribution result is a true deformation, the predicted observations should show spatially continuous changes between adjacent target monitoring primitives; if the candidate attribution result is equipment drift, the predicted observations should mainly show consistent offsets with the same observation equipment; if the candidate attribution result is benchmark drift, the predicted observations should mainly appear in the closed-loop path containing the relative position constraints of the candidate benchmark area; if the candidate attribution result is a target monitoring primitive anomaly, the predicted observations should be concentrated on the three non-collinear feature points of the target monitoring primitive and their internal distance changes.

[0070] Candidate retest actions can be drawn from a pre-configured set of actions. For fixed visual acquisition equipment, candidate retest actions could include increasing the sampling frequency, switching to a backup fixed visual acquisition equipment, or adjusting local exposure parameters. For UAV image acquisition equipment, candidate retest actions could include adding lateral flight paths, decreasing flight altitude, or expanding the overlap area. For ground-based laser scanning equipment, candidate retest actions could include rescanning target areas or adding scanning stations. For total stations or GNS receivers, candidate retest actions could include retesting the relative positions between candidate reference areas. Execution costs are determined based on equipment scheduling time, acquisition duration, on-site safety constraints, and data processing volume. Observation noise is determined based on the equipment's nominal accuracy, the current acquisition environment, historical observation stability, and calibration status.

[0071] The discrimination of candidate retest actions can be determined by the following formula:

[0072] in, The discrimination of candidate retest actions; Number the candidate retest action; Number a candidate attribution result; Number another candidate attribution result; To perform candidate retesting action At that time, in the candidate attribution results The predicted observations obtained below; To perform candidate retesting action At that time, in the candidate attribution results The predicted observations obtained below; the observation noise for candidate retest actions; The execution cost of the candidate retest action; The execution cost weight is used to convert execution costs into a penalty comparable to observation noise. The execution cost weight is configured based on the project's emphasis on retest timeliness, equipment occupancy, and on-site safety constraints, or it can be determined based on the conversion relationship between the unit execution cost and observation noise of historical retest actions. The input to this expression is the predicted observation value, observation noise, and execution cost under different candidate attribution results; the output is the discriminative power of the candidate retest action. If a candidate retest action produces significantly different predicted observation values ​​under different candidate attribution results, while having low observation noise and execution cost, this candidate retest action has high discriminative power. The data processing end determines the candidate retest action with the highest discriminative power as the target retest action; if multiple candidate retest actions have the same discriminative power, the candidate retest action with lower execution cost and higher satisfaction of on-site safety constraints is prioritized.

[0073] The closed-loop residual characteristics, residual attribution results, and deformation risk results are updated based on the retest data, including: generating retest observation constraint factors based on the retest data; adding the retest observation constraint factors to the closed-loop constraint graph; calculating the pose residuals of the closed-loop path associated with the target retest action and updating the closed-loop residual characteristics; and updating the residual attribution results and deformation risk results based on the updated closed-loop residual characteristics.

[0074] In one embodiment, the update process for the remeasurement data is limited to writing the remeasurement observation constraint factors into the closed-loop constraint graph and recalculating the closed-loop path associated with the target remeasurement action. This constraint point is used to ensure that the remeasurement results are not recorded independently, but can return to the original closed-loop residual calculation chain to participate in attribution updates. After the target remeasurement action is completed, the data processing terminal receives the remeasurement data and generates remeasurement observation constraint factors according to the source of the remeasurement data. If the remeasurement data comes from an image acquisition device, the data processing terminal identifies the primitive identifier of the target monitoring primitive, extracts the observation positions of three non-collinear feature points, and calls the calibration parameters of the corresponding image acquisition device to generate remeasurement observation constraint factors. If the remeasurement data comes from a ground laser scanning device, the data processing terminal extracts the spatial observation positions of the target patch and the target monitoring primitive, and generates point cloud remeasurement observation constraint factors. If the remeasurement data comes from a total station or a global navigation satellite system receiving device, the data processing terminal writes the supplemented distance, angle, relative elevation, or relative coordinates into the remeasurement observation constraint factors.

[0075] When the data processing end adds the re-measurement observation constraint factors to the closed-loop constraint graph, it retains the original observation constraint factors and does not directly overwrite the original observation records. The re-measurement observation constraint factors are added as new graph edges to connect the corresponding observation equipment nodes, target monitoring primitive nodes, or candidate reference area nodes, and are written with the re-measurement time, target re-measurement action number, re-measurement equipment identifier, and re-measurement quality identifier. The re-measurement quality identifier is determined based on the completeness of feature point extraction, point cloud registration error, total station angle and distance measurement accuracy, global navigation satellite system positioning status, and on-site occlusion. If the re-measurement data does not match the target re-measurement action, such as the re-measurement data not covering the target monitoring primitive, not covering the candidate reference area, or the acquisition equipment identifier being inconsistent, the data processing end marks the re-measurement data as invalid re-measurement data and does not update the closed-loop constraint graph.

[0076] After the re-measurement observation constraint factors are added to the closed-loop constraint graph, the data processing end filters the closed-loop paths associated with the target re-measurement actions. Each associated closed-loop path must contain at least the newly added re-measurement observation constraint factors and have a judgment relationship with at least one of the multiple candidate attribution results. The data processing end calculates the pose residuals of the associated closed-loop paths and writes the calculation results into a new closed-loop residual feature record. If the re-measurement equipment association features show that the residuals are still concentrated in the same observation equipment, the residual component corresponding to equipment drift is increased; if the re-measurement spatial neighborhood features show that there are still continuous changes between adjacent target monitoring elements, the residual component corresponding to the actual deformation is increased; if the re-measurement closed-loop type features show that the closed loops in the candidate reference area are continuously abnormal, the residual component corresponding to the reference drift is increased; if the re-measurement feature point distance features show that the distance between three non-collinear feature points is abnormal, the residual component corresponding to the target monitoring element anomaly is increased.

[0077] The updated closed-loop residual characteristics enter the residual source decomposition process, and the data processing end regenerates the residual attribution results. If the updated residual attribution results retain only one primary candidate attribution result, the data processing end updates the deformation risk results based on this primary candidate attribution result and deformation coordination constraints. If the updated residual attribution results still include multiple candidate attribution results, the data processing end determines whether to continue retesting. The triggering conditions for continuing retesting are determined based on the number of remaining candidate attribution results, the availability of candidate retesting actions, on-site safety conditions, and the upper limit of the number of retests. The upper limit of the number of retests can be set by the project configuration file or determined according to the engineering risk level. If a unique attribution cannot be formed even after reaching the upper limit of the number of retests, the data processing end outputs the deformation risk results pending manual review and records the candidate attribution results that could not be eliminated. Through the above process, the retest data is incorporated into the closed-loop constraint diagram and residual attribution results, and the deformation risk results can be updated synchronously with new observation evidence.

[0078] In one specific embodiment, taking a highway slope engineering survey scenario as an example, the monitoring area extends approximately 55m along the slope, with an elevation projection range of approximately 36m. Twelve target monitoring elements, designated M1 to M12, are deployed on-site. Each target monitoring element employs an asymmetric three-dimensional coded monitoring element, with three non-collinear feature points within each element. The designed distances between these three non-collinear feature points are 120.00mm, 100.00mm, and 140.00mm, respectively. Four observation devices are configured on-site, including two fixed visual observation devices (C1 and C2), one total station (T1), and one UAV (UAV) (U1). Three candidate reference areas (B1, B2, and B3) are also established. The monitoring cycle is set to six times, at 08:00, 10:00, 12:00, 14:00, 16:00, and 18:00.

[0079] like Figure 2As shown, this map was drawn based on the coordinates of the on-site layout. It illustrates the slope extent, the locations of 12 target monitoring elements, the locations of 4 observation devices, and the positional relationships of 3 candidate benchmark areas. M4 to M7 are located within the potential slip zone in the middle of the slope, B2 is positioned at the edge of the slope crest platform, and C2 is located on the right side of the slope toe. This layout allows fixed visual observation equipment to cover the middle and toe areas of the slope, drones to supplement the overall imagery of the slope crest and slope surface, and total stations to perform supplementary measurements on the relative positions between the candidate benchmark areas.

[0080] At 08:00, the system reads the observation data collected by each observation device, identifies the identification coding area and direction marker of each target monitoring element, and extracts the observation positions of three non-collinear feature points. Taking M6 as an example, the actual distances between the three non-collinear feature points measured at the initial time are 120.04mm, 100.08mm, and 139.96mm, respectively, with deviations from the design values ​​of 0.04mm, 0.08mm, and 0.04mm, respectively, all less than 0.10mm, indicating that the target monitoring element itself is stable. The system combines the element identifier, observation position, and observation device calibration parameters to generate a pose observation constraint factor, and uses the target monitoring element, observation device, and candidate reference area as graph nodes, and the pose observation constraint factor, feature point distance constraint, and relative position constraint between candidate reference areas as graph edges to construct a closed-loop constraint graph.

[0081] At each monitoring time, the system extracts the target closed-loop path from the closed-loop constraint graph and calculates the pose residual. Taking closed-loop path L3 as an example, its path passes through C1, M5, C2, M6, B2, and T1. At 16:00, a significant difference appears between the closed-loop pose transformation and the unit pose transformation of L3, resulting in a pose residual of 13.2 mm; the corresponding pose residual for L2 is 6.0 mm, and for L1 it is 1.5 mm. Figure 3 As shown, this figure was plotted based on closed-loop pose residual data at six monitoring times. The horizontal axis represents the monitoring time, and the vertical axis represents the pose residual. The figure shows that L1 remained consistently within the range of 1.1mm to 1.5mm, L2 continuously increased from 1.3mm to 6.2mm, and L3 increased from 1.5mm to 13.5mm. This result indicates a significant abnormal increase in the central part of the monitoring area after 12:00, but the source of the anomaly cannot be directly determined based solely on the closed-loop pose residual.

[0082] The system further extracts the closed-loop residual features and decomposes the sources of the residuals. At 16:00, the residuals of the closed-loop paths corresponding to M4 to M7 are decomposed, and the actual deformation residual components, equipment drift residual components, reference drift residual components, and target monitoring element anomaly residual components of M4 are 4.2mm, 1.5mm, 0.7mm, and 0.2mm, respectively; those of M5 are 6.3mm, 2.1mm, 0.8mm, and 0.3mm; those of M6 are 8.1mm, 2.4mm, 1.1mm, and 0.3mm; and those of M7 are 6.8mm, 2.0mm, 1.0mm, and 0.4mm.

[0083] like Figure 4 As shown, this figure was drawn based on the residual source decomposition results from M4 to M7 at 16:00. The horizontal axis represents the target monitoring element, and the vertical axis represents the residual components. The figure shows the superposition relationship of four types of residual components: actual deformation, equipment drift, reference drift, and target monitoring element anomaly. It can be seen that the actual deformation residual component is significantly higher at positions M5 to M7, the equipment drift residual component is mainly concentrated in the closed-loop path involving C2, the reference drift residual component is generally small, and the target monitoring element anomaly residual component is consistently below 0.4 mm. This result indicates that the dominant source of anomalies is not damage to the target monitoring element, but rather actual deformation, superimposed with a small amount of equipment drift and reference drift.

[0084] The system invokes deformation coordination constraints based on the type of engineering survey object. In this embodiment, the engineering object is a slope geological body, and the geological body block displacement continuity constraint is adopted. The system performs a continuity check on the actual deformation residual components from M4 to M7, and finds that a continuous displacement zone is formed along the slope direction in this area, and the displacement direction is basically consistent. If the original closed-loop pose residual is directly used for threshold judgment, the 13.2mm corresponding to M6 may be misjudged as a high-risk single-point mutation; after decomposing the residual source, it can be confirmed that about 8.1mm comes from actual deformation, about 2.4mm comes from equipment drift, about 1.1mm comes from reference drift, and the rest are target monitoring element anomalies and random errors. Based on this, the system forms two candidate attribution results: one is that the slope is mainly due to local slippage and superimposed with slight drift of C2, and the other is that the slope is mainly due to local drift of B2 and accompanied by actual deformation of the slope.

[0085] To distinguish between the two candidate attribution results, the system generated three candidate retest actions. Retest action A1 involves adding a lateral low-altitude flight path for UAV U1, focusing on supplementing measurements from M4 to M7 and B2; retest action A2 involves supplementing measurements of the relative positions between B2 and B3 using total station T1; retest action A3 involves C2 performing a self-check and repeatedly acquiring data from M5 and M6. The system predicts the difference in observation values ​​for each candidate retest action under different candidate attribution results, and calculates the discrimination index by combining observation noise and execution cost. The discrimination indexes are 2.31 for A1, 1.48 for A2, and 0.96 for A3. Therefore, A1 is selected as the target retest action.

[0086] After the retest is completed, the system adds the retest observation constraint factor to the closed-loop constraint graph and recalculates the pose residuals of the associated closed-loop paths. After the update, the actual deformation displacements of M4 to M7 are 2.4 mm, 5.8 mm, 8.4 mm, and 7.3 mm, respectively. Figure 5 The data was drawn based on the actual deformation and displacement profiles before and after the re-measurement. The horizontal axis represents the target monitoring element number, and the vertical axis represents the actual deformation and displacement. Figure 5 As shown, the displacement curves before the retest showed an upward trend near M5 and M6. After the retest, this trend became clearer, with the deformation peak concentrated near M6. A continuous slip zone profile was formed from M5 to M7.

[0087] Based on the residual attribution results, the continuous constraint of geological block displacement, and the updated re-measurement results, the system ultimately classifies areas M5 to M7 as Level II deformation risk zones, with M6 being a key risk point. It is recommended to initiate intensified monitoring and support verification on-site. Compared to conventional methods that do not employ residual source decomposition and active re-measurement selection, this embodiment can separate equipment drift and baseline drift from the comprehensive anomaly, avoiding the direct attribution of observation system errors as engineering deformation. Simultaneously, through discrimination-driven re-measurement action selection, new observations can be focused on differentiating candidate attribution results, providing a clearer data source and attribution chain for risk assessment. As demonstrated in this embodiment, the present invention can identify true deformation in engineering survey scenarios with multiple source errors and generate traceable deformation risk results.

[0088] The above are merely embodiments of the present invention and are not intended to limit the invention. Those skilled in the art can make various modifications, equivalent substitutions, and improvements within the spirit and principles of the present invention.

Claims

1. A method for multi-equipment collaborative deformation monitoring and data processing in engineering surveying, characterized in that, The method includes: The observation data of the target monitoring element collected by multiple observation devices are acquired, and the observation constraint factor is generated based on the observation data. The target monitoring element is an asymmetric three-dimensional coded monitoring element including three non-collinear feature points. Based on the observation constraint factor and the distance constraint between the three non-collinear feature points, a closed-loop constraint graph is constructed, the pose residual of the closed-loop path is calculated, and the closed-loop residual characteristics are obtained. Based on the closed-loop residual characteristics, residual source decomposition is performed to obtain residual components corresponding to actual deformation, equipment drift, reference drift and the anomaly of the target monitoring unit, respectively, and residual attribution results are generated. Deformation risk results are determined based on the residual attribution results and deformation coordination constraints. When the residual attribution result includes multiple candidate attribution results, the target retest action is determined based on the discriminativeness of the candidate retest actions, the retest data corresponding to the target retest action is obtained, and the closed-loop residual features, the residual attribution result and the deformation risk result are updated based on the retest data, and the deformation risk result is output.

2. The method according to claim 1, characterized in that, The target monitoring element further includes an identification coding area and a direction identifier. The identification coding area is used to determine the element identifier of the target monitoring element, and the direction identifier and the three non-collinear feature points are used to determine the attitude orientation of the target monitoring element.

3. The method according to claim 2, characterized in that, The step of generating observation constraint factors based on the observation data includes: Identify the identification coding region and extract the observation positions of the three non-collinear feature points; Based on the primitive identifier, the observation location, and the calibration parameters of the observation equipment that collects observation data containing the observation location, a pose observation constraint factor corresponding to the target monitoring primitive is generated.

4. The method according to claim 1, characterized in that, The step of constructing a closed-loop constraint graph based on the observation constraint factor and the distance constraint between the three non-collinear feature points includes: The target monitoring primitive, the multiple observation devices, and the candidate reference area used to determine the reference drift are used as graph nodes, and the observation constraint factor, the distance constraint between the three non-collinear feature points, and the relative position constraint between the candidate reference areas are used as graph edges to construct the closed-loop constraint graph.

5. The method according to claim 4, characterized in that, The calculation of the pose residual of the closed-loop path includes: Determine the target closed-loop path in the closed-loop constraint graph; Multiple relative pose observations are determined based on the observation constraint factors in the target closed-loop path; The multiple relative pose observations are combined according to the connection order of the target closed-loop path to obtain the closed-loop pose transformation. The pose residual of the closed-loop path is determined based on the difference between the closed-loop pose transformation and the unit pose transformation.

6. The method according to claim 1, characterized in that, The closed-loop residual features include device association features, spatial neighborhood features, feature point distance features, and closed-loop type features; The device association feature is used to indicate the observation device that participates in forming the pose residual of the closed-loop path among the plurality of observation devices. The spatial neighborhood feature is used to indicate the spatial distribution range of the pose residual of the closed-loop path. The feature point distance feature is used to indicate the distance change between the three non-collinear feature points. The closed-loop type feature is used to indicate the constraint source corresponding to the closed-loop path.

7. The method according to claim 6, characterized in that, The residual source decomposition based on the closed-loop residual characteristics includes: The residual components corresponding to the actual deformation are determined based on the spatial neighborhood characteristics. Determine the residual component corresponding to the device drift based on the device association characteristics; The residual component corresponding to the baseline drift is determined based on the closed-loop type characteristics; the residual component corresponding to the anomaly of the target monitoring element is determined based on the feature point distance characteristics.

8. The method according to claim 1, characterized in that, The deformation coordination constraints include at least one of the following: geological block displacement continuity constraints, support structure horizontal displacement curve constraints, tunnel section convergence constraints, and beam deflection curve constraints. The determination of deformation risk results based on the residual attribution results and deformation compatibility constraints includes: The deformation coordination constraints are determined according to the type of the engineering survey object, and the deformation risk result is determined according to the matching result between the residual attribution result and the corresponding deformation coordination constraint.

9. The method according to claim 1, characterized in that, The step of determining the target retest action based on the discriminative power of the candidate retest actions includes: For each candidate retest action, determine the predicted observation value corresponding to the multiple candidate attribution results when the candidate retest action is executed; The discriminative power of the candidate retest action is determined based on the differences between the predicted observations corresponding to different candidate attribution results, the observation noise of the candidate retest action, and the execution cost. The candidate retest action with the highest discriminative power is determined as the target retest action.

10. The method according to claim 9, characterized in that, The step of updating the closed-loop residual characteristics, the residual attribution results, and the deformation risk results based on the retest data includes: Generate retest observation constraint factors based on the retest data; Add the re-measurement observation constraint factor to the closed-loop constraint graph; Calculate the pose residual of the closed-loop path associated with the target retest action, and update the closed-loop residual features; The residual attribution results and the deformation risk results are updated based on the updated closed-loop residual characteristics.

Images