Ancient building bracket system node three-dimensional deformation dynamic monitoring method

Abstract

The invention discloses a three-dimensional deformation dynamic monitoring method for ancient building bracket system nodes. The method comprises the following steps: acquiring high-precision point cloud data of the surfaces of the bracket system nodes by adopting a non-contact three-dimensional laser scanning technology; according to the method, the key feature points of the bracket system nodes are extracted from the digital twin model by using the deep learning algorithm, and the four-dimensional motion trail model is constructed, so that the dynamic characteristics of the bracket system nodes can be deeply understood; through an environment interference intelligent filtering algorithm, deformation data in the four-dimensional motion trail model is corrected, the algorithm can effectively identify and eliminate non-structural displacement caused by environmental factors, retain real structural displacement data and improve the accuracy and reliability of the deformation data, and according to the corrected deformation data, the accuracy and reliability of the four-dimensional motion trail model are improved. A multi-level early warning mechanism is triggered through dynamic threshold early warning, different early warning measures can be taken according to the severity of deformation, the early warning response is more reasonable and effective, and potential safety hazards of the ancient building bracket system nodes can be found in time.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ancient building monitoring, in particular to a three-dimensional deformation dynamic monitoring method for ancient building corbel bracket joints. BACKGROUND

[0002] Ancient buildings, as valuable carriers of historical and cultural heritage, bear rich historical information and cultural value. However, with the passage of time and changes in the environment, ancient buildings inevitably undergo various forms of deformation, such as wall cracking, beam and column tilting, etc. These deformations not only affect the aesthetics of ancient buildings, but also seriously threaten their structural safety, and may even lead to the collapse of ancient buildings.

[0003] Traditional methods of monitoring the deformation of ancient buildings mainly rely on periodic manual inspection and simple sensor monitoring. Manual inspection is inefficient and difficult to achieve real-time monitoring, making it difficult to detect potential deformation problems in a timely manner. Although simple sensor monitoring can achieve some degree of real-time monitoring, it can only obtain single deformation data and lacks comprehensive consideration of environmental factors. For example, temperature changes will cause thermal expansion and contraction of building materials, humidity changes will affect the moisture content of wood, and wind speed will exert dynamic loads on the building structure, all of which will affect the deformation of ancient buildings. Traditional methods fail to fully consider these factors, resulting in inaccurate assessment of ancient building deformation and low reliability of early warning. SUMMARY

[0004] To solve the above technical problems, a three-dimensional deformation dynamic monitoring method for ancient building corbel bracket joints is provided, which solves the problems raised in the background.

[0005] To achieve the above purposes, the technical solution adopted by the present application is as follows: In the first aspect of the present application, a three-dimensional deformation dynamic monitoring method for ancient building corbel bracket joints is provided, comprising: Using non-contact three-dimensional laser scanning technology to obtain high-precision point cloud data of the corbel bracket joint surface; Based on the point cloud data, and collecting tilt photogrammetry data in the same monitoring area, performing multi-source data registration and fusion to generate a digital twin model; Using a deep learning algorithm to extract key feature points of the corbel bracket joint from the digital twin model and constructing a four-dimensional motion trajectory model; Using an environmental interference intelligent filtering algorithm to correct the deformation data in the four-dimensional motion trajectory model and distinguish between structural displacement and non-structural interference; According to the corrected deformation data, triggering a multi-level early warning mechanism through a dynamic threshold early warning.

[0006] Preferably, the non-contact three-dimensional laser scanning technology is used to obtain high-precision point cloud data of the corbel node surface, specifically including: A phase laser scanner is selected as the scanning device, and the corbel node geometric center is taken as the reference to obtain point cloud data of the corbel node surface through multi-view scanning; A statistical outlier removal algorithm is used, the threshold of the number of neighborhood points is set to 20, and the standard deviation multiple is set to 1.5 to remove noise points in the obtained point cloud data; Then, a bilateral filtering algorithm is applied, the spatial domain standard deviation is set to 0.5, and the color domain standard deviation is set to 0.3 to smooth the point cloud while retaining the edge features of the point cloud data; The processed point cloud data is unified to the WGS84 coordinate system and stored as a LAS format file according to the monitoring time sequence number to construct a high-precision point cloud data set containing spatial coordinates (X, Y, Z), reflection intensity, and time stamp.

[0007] Preferably, based on the point cloud data, oblique photogrammetry data in the same monitoring area are collected, multi-source data registration and fusion are performed, and a digital twin model is generated, specifically including: A multi-view oblique photogrammetry device is used to take aerial photographs of the same monitoring area to obtain oblique photogrammetry original image data containing the overall appearance of the ancient building, a three-dimensional reconstruction software is used to perform aerial triangulation encryption processing on the original image data to generate an oblique photogrammetry three-dimensional model with a geographic coordinate system WGS84, and building contour lines and surface texture information are extracted from the model; In the high-precision point cloud data set, a feature extraction algorithm based on geometric constraints is used to extract feature point sets of key components such as corbel arches, arch centers, and tenon heads, and record their three-dimensional coordinates (X, Y, Z) and normal vectors (Nx, Ny, Nz); In the oblique photogrammetry three-dimensional model, semantic segmentation network is used to identify building components at corresponding positions and extract their feature point coordinates, and based on the double constraint conditions of Euclidean distance and normal vector angle between feature points, point cloud feature points and oblique photogrammetry feature points are initially matched to construct an initial corresponding point pair set; An iterative closest point (ICP) algorithm is used to perform initial registration on the initial corresponding point pair set, and an initial transformation matrix is generated by dynamically adjusting the iteration threshold; Based on the initial transformation matrix, the point cloud data is converted to the oblique photogrammetry coordinate system, and a bundle adjustment algorithm is used to optimize the global coordinate system to register the point cloud data and the oblique photogrammetry data by minimizing the re-projection error; The registered point cloud data and oblique photogrammetry data are fused with the texture information of the oblique photogrammetry three-dimensional model, wherein the point cloud data provides millimeter-level geometric accuracy, and the oblique photogrammetry texture provides real surface color and material information; A fused 3D mesh model is generated using the Poisson reconstruction algorithm, and the oblique photographic texture is mapped onto the mesh surface to ultimately generate a digital twin model.

[0008] Preferably, the step of extracting key feature points of the bracket set nodes from the digital twin model using a deep learning algorithm and constructing a four-dimensional motion trajectory model specifically includes: A deep learning network is constructed. The input layer is configured to receive 3D point cloud data from the digital twin model. The middle layer of the network adopts a three-level nested sampling. The first level samples the farthest point with a radius of 0.2m to obtain the global context. The second level extracts local geometric features through dynamic radius search with a radius range of 0.05m-0.1m. The third level samples the key areas of the mortise and tenon joint with a fixed radius of 0.02m. The output layer adopts a dual-branch structure, where the geometry branch generates point-level three-dimensional displacement vectors (ΔX, ΔY, ΔZ) through MLP regression, and the texture branch generates a displacement field heatmap that matches the resolution of the input point cloud through deconvolution operation. Based on the displacement field data of continuous time series in the displacement field heat map, the Kalman filter algorithm is used to smooth the discrete displacement points, and the DBSCAN clustering algorithm is used to identify the displacement clusters of the same bracket node at different times, and a four-dimensional motion trajectory model containing spatial coordinates (X, Y, Z), timestamp T and displacement D is constructed.

[0009] Preferably, the step of correcting the deformation data in the four-dimensional motion trajectory model using an intelligent environmental interference filtering algorithm to distinguish between structural displacement and non-structural interference specifically includes: The dominant frequency of the vibration signal was identified by spectrum analysis, which distinguished between the walking frequency of tourists and the natural frequency of the building. The walking frequency of tourists ranged from 1 to 3 Hz, while the natural frequency of the building ranged from 0.1 to 0.5 Hz. The Kalman filter algorithm is used to smooth the structural displacement data corresponding to the identified building natural frequency to suppress high-frequency noise interference. The state transition matrix of the Kalman filter algorithm is constructed based on the building natural frequency characteristics. A vibration energy threshold model is established, and a threshold for tourist vibration energy is set. When the real-time monitored tourist vibration energy exceeds the threshold, a data acquisition pause mechanism is triggered and an abnormal period is marked. The affected data within the abnormal period is corrected using cubic spline interpolation.

[0010] Preferably, the step of triggering a multi-level early warning mechanism based on the corrected deformation data through dynamic threshold warning specifically includes: A Long Short-Term Memory (LSTM) network model is constructed, whose input layer receives corrected deformation data and synchronously collected environmental parameters, including temperature, humidity, and wind speed. The double-layer LSTM unit structure is arranged in the hidden layer, and a long-term dependence relationship modeling of the deformation trend is realized through a gating mechanism, which includes a collaborative calculation of a forgetting gate, an input gate and an output gate; A full connection network is configured in the output layer to generate probability values corresponding to the yellow, orange and red three-level early warning; According to the highest-level early warning probability, corresponding operations are performed, the yellow early warning is to push real-time deformation monitoring data and early warning information to a management terminal, the orange early warning is to start an audible and light alarm device of a physical place and notify a responsible person, and the red early warning is to automatically close a building visiting channel and start a preset emergency reinforcement program.

[0011] Preferably, the input layer in the LSTM neural network model is: ; Among them, is a comprehensive deformation value of the ancient building structure in the three-dimensional space at time t, is an air temperature of an environment where the ancient building is located at time t, is a relative humidity of the environment where the ancient building is located at time t, is a wind speed measured on the surface of the ancient building at time t.

[0012] Preferably, the full connection network is configured in the output layer, and a yellow, orange and red three-level early warning probability vector is generated through a Softmax function, and the Softmax function is: ; Among them, The three-level early warning probability vector generated by the output layer is used to determine the possibility of the occurrence of the yellow, orange and red early warning levels at the current time t; The function is an activation function, which converts the input vector into a probability distribution; The weight matrix of the output layer, is a parameter of a linear transformation, which is used to map the state of the hidden layer to an output space, and the dimension of the weight matrix depends on the dimension of the state of the hidden layer and the number of output early warning categories; The state vector of the hidden layer at time t; The bias vector of the output layer, the bias vector acts on the state of the hidden layer together with the weight matrix .

[0013] In the second aspect of the present application, a three-dimensional deformation dynamic monitoring system for a corbel bracket joint of an ancient building is also provided, comprising: The acquisition module is used for obtaining high-precision point cloud data of the node surface of the arch of the ancient building by using a non-contact three-dimensional laser scanning technology; The fusion module is used for fusing and registering multi-source data based on the point cloud data and collecting the oblique photogrammetry data in the same monitoring area to generate a digital twin model; The construction module is used for extracting key feature points of the node of the arch from the digital twin model by using a deep learning algorithm and constructing a four-dimensional motion trajectory model; The correction module is used for correcting deformation data in the four-dimensional motion trajectory model by using an environmental interference intelligent filtering algorithm to distinguish structural displacement from non-structural interference; The early warning module is used for triggering a multi-level early warning mechanism by a dynamic threshold early warning according to the corrected deformation data.

[0014] Compared with the prior art, the application provides a three-dimensional deformation dynamic monitoring method for the node of the arch of an ancient building, and has the following beneficial effects: The application extracts key feature points of the node of the arch from the digital twin model by using a deep learning algorithm and constructs a four-dimensional motion trajectory model, which helps to deeply understand the dynamic characteristics of the node of the arch; the deformation data in the four-dimensional motion trajectory model is corrected by using an environmental interference intelligent filtering algorithm, the algorithm can effectively identify and eliminate non-structural displacement caused by environmental factors, retain real structural displacement data, improve the accuracy and reliability of the deformation data, trigger a multi-level early warning mechanism by a dynamic threshold early warning according to the corrected deformation data, improve the sensitivity and accuracy of the early warning, and the multi-level early warning mechanism can take different early warning measures according to the severity of the deformation, so that the early warning response is more reasonable and effective, and the safety hazards of the node of the arch of the ancient building can be found in time, thereby providing a scientific basis for the protection and maintenance of the ancient building. BRIEF DESCRIPTION OF DRAWINGS

[0015] Figure 1 It is a method flowchart of S101-S105 in the application; Figure 2 It is a method flowchart of S201-S204 in the application; Figure 3 It is a method flowchart of S301-S307 in the application; Figure 4 It is a method flowchart of S401-S403 in the application; Figure 5 It is a method flowchart of S501-S503 in the application; Figure 6 It is a method flowchart of S601-S604 in the application. Detailed Implementation

[0016] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0017] Example 1 Please refer to Figure 1 As shown, in a first aspect of the present invention, a method for dynamic monitoring of three-dimensional deformation of bracket sets in ancient buildings is provided, comprising: S101. High-precision point cloud data of the surface of the bracket arch node is obtained by using non-contact three-dimensional laser scanning technology; S102. Based on point cloud data, and by collecting oblique photogrammetry data within the same monitoring area, multi-source data registration and fusion are performed to generate a digital twin model. S103. Use deep learning algorithms to extract key feature points of the bracket nodes from the digital twin model and construct a four-dimensional motion trajectory model. S104. Through an intelligent environmental interference filtering algorithm, the deformation data in the four-dimensional motion trajectory model is corrected to distinguish between structural displacement and non-structural interference. S105. Based on the corrected deformation data, a multi-level early warning mechanism is triggered through dynamic threshold early warning.

[0018] The skilled in the art can understand that the present application adopts non-contact three-dimensional laser scanning technology to obtain high-precision point cloud data of the corbel node surface, avoids the damage to the ancient building caused by the traditional contact measurement, and can comprehensively and carefully capture the geometric information of the corbel node surface, thereby providing accurate basic data for subsequent analysis; the oblique photogrammetry data can reflect the overall situation of the monitoring area from a macro perspective, and is combined with the microscopic details of the point cloud data, so that the digital twin model is more real and complete in restoring the corbel node of the ancient building and the surrounding environment, thereby providing more abundant information for subsequent feature extraction and motion analysis; the deep learning algorithm is used to extract the key feature points of the corbel node from the digital twin model, and a four-dimensional motion trajectory model is constructed. The deep learning algorithm can automatically learn and identify the complex features of the corbel node, accurately extract the key feature points, and the four-dimensional motion trajectory model can comprehensively describe the motion of the corbel node in the three-dimensional space and the change over time, which helps to deeply understand the dynamic characteristics of the corbel node; the deformation data in the four-dimensional motion trajectory model is corrected through the environmental interference intelligent filtering algorithm, and the structural displacement and non-structural interference are distinguished. The algorithm can effectively identify and eliminate the non-structural displacement caused by environmental factors, retain the real structural displacement data, improve the accuracy and reliability of the deformation data, and provide more accurate data support for subsequent early warning; according to the corrected deformation data, a multi-level early warning mechanism is triggered through a dynamic threshold. The dynamic threshold can be adaptively adjusted according to the actual deformation of the corbel node of the ancient building and the environmental changes, thereby improving the sensitivity and accuracy of the early warning. The multi-level early warning mechanism can take different early warning measures according to the severity of the deformation, so that the early warning response is more reasonable and effective, and the safety hazards of the corbel node of the ancient building can be found in time, thereby providing a scientific basis for the protection and maintenance of the ancient building.

[0019] Please refer to Figure 2 The high-precision point cloud data of the corbel node surface is obtained by using non-contact three-dimensional laser scanning technology, specifically including: S201, selecting a phase laser scanner as a scanning device, taking the geometric center of the corbel node as a reference, and scanning through multiple perspectives to obtain point cloud data of the corbel node surface; S202, using a statistical outlier removal algorithm, setting the threshold of the number of neighborhood points to 20 and the multiple of the standard deviation to 1.5, and removing the noise points in the obtained point cloud data; S203, applying a bilateral filtering algorithm again, setting the spatial domain standard deviation to 0.5 and the color domain standard deviation to 0.3, smoothing the point cloud while retaining the edge features of the point cloud data; S204, storing the processed point cloud data in LAS format files under the WGS84 coordinate system according to the monitoring time sequence number, and constructing a high-precision point cloud data set containing spatial coordinates (X, Y, Z), reflection intensity and time stamp.

[0020] Please refer to Figure 3 Based on point cloud data, and collect the same monitoring area in the oblique photogrammetry data, multi-source data registration fusion, generate digital twin model, specific including: S301, using multi-view oblique photography equipment for the same monitoring area flight shooting, get the original image data of oblique photogrammetry containing the whole picture of ancient buildings, through the three-dimensional reconstruction software to the original image data for space three encryption processing, generate oblique photogrammetry three-dimensional model with geographic coordinate system WGS84, and extract the building contour line and surface texture information from the model; S302, in the high-precision point cloud data set, using feature extraction algorithm based on geometric constraint, extracting the feature point set of key components such as arch mouth, arch center and tenon, recording its three-dimensional coordinates (X, Y, Z) and normal vector (Nx, Ny, Nz); S303, in the oblique photogrammetry three-dimensional model, the building components at the corresponding position are identified through the semantic segmentation network, and the feature point coordinates are extracted, based on the double constraint conditions of Euclidean distance and normal vector angle between feature points, the point cloud feature points and oblique photography feature points are matched, and the initial corresponding point pair set is constructed; S304, using iterative closest point (ICP) algorithm to initial registration of initial corresponding point pair set, through dynamic adjustment of iteration threshold, generate initial transformation matrix; S305, based on the initial transformation matrix, convert the point cloud data to the oblique photography coordinate system, and use the bundle adjustment algorithm to optimize the global coordinate system, and through the minimization of the reprojection error, the point cloud data and the oblique photography data are registered; S306, the registered point cloud data and oblique photography data are fused with the texture information of the oblique photogrammetry three-dimensional model, wherein the point cloud data provides millimeter level geometric accuracy, and the oblique photography texture provides real surface color and material information; S307, through the poisson reconstruction algorithm to generate the fused three-dimensional grid model, and map the oblique photography texture to the grid surface, finally generate the digital twin model.

[0021] Please refer to Figure 4 As shown in the figure, the key feature points of arch node are extracted from the digital twin model by using deep learning algorithm, and a four-dimensional motion trajectory model is constructed, including: S401, construct a deep learning network, the input layer is configured to receive three-dimensional point cloud data in the digital twin model, the middle layer of the network adopts three-level nested sampling, wherein the first level performs farthest point sampling with a radius of 0.2 m to obtain global context, the second level extracts local geometric features through dynamic radius search, and the radius range is 0.05 m-0.1 m, and the third level adopts a fixed radius of 0.02 m for sampling in the preset key area of the mortise and tenon joint; S402, the output layer adopts a double-branch structure, wherein the geometric branch generates a point-level three-dimensional displacement vector (ΔX, ΔY, ΔZ) through MLP regression, and the texture branch generates a displacement field heat map matching the resolution of the input point cloud through deconvolution operation; S403, based on the displacement field data of the continuous time sequence in the displacement field heat map, the Kalman filtering algorithm is used to smooth the discrete displacement points, the DBSCAN clustering algorithm is used to identify the displacement clusters of the same corbel node at different times, and a four-dimensional motion trajectory model containing spatial coordinates (X, Y, Z), time stamp T and displacement D is constructed.

[0022] Please refer to Figure 5 , the deformation data in the four-dimensional motion trajectory model is corrected through an environmental interference intelligent filtering algorithm, and structural displacement and non-structural interference are distinguished, which specifically includes: S501, identify the main frequency of the vibration signal through spectrum analysis, and distinguish the tourist step frequency from the building natural frequency, wherein the tourist step frequency ranges from 1-3 Hz, and the building natural frequency ranges from 0.1-0.5 Hz; S502, the Kalman filtering algorithm is used to smooth the structural displacement data corresponding to the identified building natural frequency, and high-frequency noise interference is suppressed, and the state transition matrix of the Kalman filtering algorithm is constructed based on the building natural frequency characteristics; S503, establish a vibration energy threshold model, set a tourist vibration energy threshold, when the real-time monitored tourist vibration energy exceeds the threshold, trigger the data collection pause mechanism and mark the abnormal period, and use the cubic spline interpolation method to correct the affected data in the abnormal period.

[0023] Please refer to Figure 6 , according to the corrected deformation data, a multi-level early warning mechanism is triggered through a dynamic threshold early warning, which specifically includes: S601, construct a long short-term memory network LSTM model, the input layer of which receives the corrected deformation data and synchronously collected environmental parameters, including temperature, humidity and wind speed; S602, set a double-LSTM unit structure in the hidden layer, model the long-term dependence relationship of the deformation trend through the gating mechanism, and the gating mechanism includes the cooperative calculation of the forgetting gate, the input gate and the output gate; S603, configuring a full connection network in an output layer to generate probability values corresponding to yellow, orange and red early warning; S604, performing corresponding operations according to the probability of the highest level early warning, the yellow early warning is to push real-time deformation monitoring data and early warning information to a management terminal, the orange early warning is to start an audible and visual alarm device of a physical place and notify a responsible person, and the red early warning is to automatically close a building visiting channel and start a preset emergency reinforcement program.

[0024] The input layer in the LSTM neural network model is: ; wherein, is a comprehensive deformation value of the ancient building structure in a three-dimensional space at time t, is an air temperature of an environment where the ancient building is located at time t, is a relative humidity of the environment where the ancient building is located at time t, is a wind speed measured on the surface of the ancient building at time t.

[0025] A full connection network is configured in the output layer to generate a yellow, orange and red early warning probability vector through a Softmax function, and the Softmax function is: ; wherein, The three-level early warning probability vector generated by the output layer is used to determine the possibility of occurrence of the yellow, orange and red early warning levels at the current time t; The function is an activation function which converts the input vector into a probability distribution; The weight matrix of the output layer, is a parameter of a linear transformation, which is used to map the state of the hidden layer to an output space, and the dimension of the weight matrix depends on the dimension of the state of the hidden layer and the number of output early warning categories; The state vector of the hidden layer at time t; The bias vector of the output layer, the bias vector acts on the state of the hidden layer together with the weight matrix .

[0026] In the second aspect of the present application, a three-dimensional deformation dynamic monitoring system for a gable arch node of an ancient building is also provided, comprising: A collection module, which is used to obtain high-precision point cloud data of the gable arch node surface by using a non-contact three-dimensional laser scanning technology; The fusion module is used to perform multi-source data registration and fusion based on point cloud data and oblique photogrammetry data collected in the same monitoring area to generate a digital twin model. The building module is used to extract key feature points of the bracket nodes from the digital twin model using deep learning algorithms and to build a four-dimensional motion trajectory model. The correction module is used to correct the deformation data in the four-dimensional motion trajectory model through an intelligent environmental interference filtering algorithm, distinguishing between structural displacement and non-structural interference. The early warning module is used to trigger a multi-level early warning mechanism based on the corrected deformation data and dynamic threshold warnings.

[0027] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for dynamic monitoring of three-dimensional deformation of bracket nodes in ancient buildings, characterized in that, include: High-precision point cloud data of the surface of the bracket arch node was obtained using non-contact 3D laser scanning technology; Based on point cloud data and by collecting oblique photogrammetry data within the same monitoring area, multi-source data registration and fusion are performed to generate a digital twin model. Using deep learning algorithms, key feature points of the bracket nodes are extracted from the digital twin model, and a four-dimensional motion trajectory model is constructed. The deformation data in the four-dimensional motion trajectory model is corrected by an intelligent environmental interference filtering algorithm to distinguish between structural displacement and non-structural interference. Based on the corrected deformation data, a multi-level early warning mechanism is triggered through dynamic threshold warnings.

2. The method for three-dimensional deformation dynamic monitoring of ancient building bracket nodes according to claim 1, characterized in that, The method of acquiring high-precision point cloud data of the dougong (bracket set) node surface using non-contact 3D laser scanning technology specifically includes: A phase-type laser scanner was selected as the scanning device. The geometric center of the bracket node was used as the reference, and the point cloud data of the bracket node surface was obtained by scanning from multiple perspectives. A statistical outlier removal algorithm was used, with a neighborhood point threshold of 20 and a standard deviation factor of 1.5, to remove noise points from the acquired point cloud data. Then, a bilateral filtering algorithm is applied, with the spatial domain standard deviation set to 0.5 and the color domain standard deviation set to 0.3, to smooth the point cloud while preserving the edge features of the point cloud data; The processed point cloud data were unified into the WGS84 coordinate system and stored as LAS format files according to the monitoring time series number, thus constructing a high-precision point cloud dataset containing spatial coordinates (X, Y, Z), reflection intensity, and timestamps.

3. The method for dynamic monitoring of three-dimensional deformation of bracket nodes in ancient buildings according to claim 2, characterized in that, The process of generating a digital twin model based on point cloud data and by collecting oblique photogrammetry data within the same monitoring area, performing multi-source data registration and fusion, specifically includes: Aerial photography of the same monitoring area was carried out using multi-view oblique photography equipment to obtain oblique photogrammetry raw image data containing the whole view of ancient buildings. The raw image data was then subjected to aerial triangulation processing by 3D reconstruction software to generate an oblique photogrammetry 3D model with geographic coordinate system WGS84, and the building outline and surface texture information were extracted from the model. In the high-precision point cloud dataset, a feature extraction algorithm based on geometric constraints is used to extract feature point sets of key components such as the bracket opening, arch core, and tenon, and record their three-dimensional coordinates (X, Y, Z) and normal vectors (Nx, Ny, Nz). In the oblique photogrammetry 3D model, the corresponding building components are identified by the semantic segmentation network and their feature point coordinates are extracted. Based on the dual constraints of the Euclidean distance and the angle between the normal vectors between the feature points, the point cloud feature points and the oblique photogrammetry feature points are initially matched to construct an initial set of corresponding point pairs. The iterative nearest point (ICP) algorithm is used to perform initial registration on the initial set of corresponding point pairs. The initial transformation matrix is ​​generated by dynamically adjusting the iteration threshold. Based on the initial transformation matrix, the point cloud data is transformed into the oblique photogrammetry coordinate system, and the global coordinate system is optimized using the bundle adjustment algorithm. By minimizing the reprojection error, the point cloud data and the oblique photogrammetry data are registered. The registered point cloud data and oblique photogrammetry data are fused with the texture information of the oblique photogrammetry 3D model. The point cloud data provides millimeter-level geometric accuracy, while the oblique photogrammetry texture provides real surface color and material information. A fused 3D mesh model is generated using the Poisson reconstruction algorithm, and the oblique photographic texture is mapped onto the mesh surface to ultimately generate a digital twin model.

4. The method for three-dimensional deformation dynamic monitoring of ancient building bracket nodes according to claim 3, characterized in that, The process of extracting key feature points of the bracket set nodes from the digital twin model using deep learning algorithms and constructing a four-dimensional motion trajectory model specifically includes: A deep learning network is constructed. The input layer is configured to receive 3D point cloud data from the digital twin model. The middle layer of the network adopts a three-level nested sampling. The first level samples the farthest point with a radius of 0.2m to obtain the global context. The second level extracts local geometric features through dynamic radius search with a radius range of 0.05m-0.1m. The third level samples the key areas of the mortise and tenon joint with a fixed radius of 0.02m. The output layer adopts a dual-branch structure, where the geometry branch generates point-level three-dimensional displacement vectors (ΔX, ΔY, ΔZ) through MLP regression, and the texture branch generates a displacement field heatmap that matches the resolution of the input point cloud through deconvolution operation. Based on the displacement field data of continuous time series in the displacement field heat map, the Kalman filter algorithm is used to smooth the discrete displacement points, and the DBSCAN clustering algorithm is used to identify the displacement clusters of the same bracket node at different times, and a four-dimensional motion trajectory model containing spatial coordinates (X, Y, Z), timestamp T and displacement D is constructed.

5. The method for dynamic monitoring of three-dimensional deformation of bracket nodes in ancient buildings according to claim 4, characterized in that, The method employs an intelligent environmental interference filtering algorithm to correct deformation data in the four-dimensional motion trajectory model, distinguishing between structural displacements and non-structural disturbances. Specifically, this includes: The dominant frequency of the vibration signal was identified by spectrum analysis, which distinguished between the walking frequency of tourists and the natural frequency of the building. The walking frequency of tourists ranged from 1 to 3 Hz, while the natural frequency of the building ranged from 0.1 to 0.5 Hz. The Kalman filter algorithm is used to smooth the structural displacement data corresponding to the identified building natural frequency to suppress high-frequency noise interference. The state transition matrix of the Kalman filter algorithm is constructed based on the building natural frequency characteristics. A vibration energy threshold model is established, and a threshold for tourist vibration energy is set. When the real-time monitored tourist vibration energy exceeds the threshold, a data acquisition pause mechanism is triggered and an abnormal period is marked. The affected data within the abnormal period is corrected using cubic spline interpolation.

6. The method for dynamic monitoring of three-dimensional deformation of bracket nodes in ancient buildings according to claim 5, characterized in that, The step of triggering a multi-level early warning mechanism based on the corrected deformation data and through dynamic threshold warning specifically includes: A Long Short-Term Memory (LSTM) network model is constructed, whose input layer receives corrected deformation data and synchronously collected environmental parameters, including temperature, humidity, and wind speed. A two-layer LSTM unit structure is set in the hidden layer, and the long-term dependency relationship of deformation trend is modeled through a gating mechanism, which includes the collaborative computation of forget gate, input gate and output gate. Configure a fully connected network at the output layer to generate probability values ​​for the corresponding yellow, orange, and red three-level warnings; The corresponding operation is executed according to the highest level of warning probability. The yellow warning is to push real-time deformation monitoring data and warning information to the management terminal. The orange warning is to activate the sound and light alarm device of the physical site and notify the responsible personnel. The red warning is to automatically close the building visitor passage and start the preset emergency reinforcement program.

7. The method for dynamic monitoring of three-dimensional deformation of bracket nodes in ancient buildings according to claim 6, characterized in that, The input layer in the LSTM neural network model is: ； in, Let be the comprehensive deformation value of the ancient building structure in three-dimensional space at time t. The air temperature of the environment in which the ancient building is located at time t. The relative humidity of the environment in which the ancient building is located at time t. Let t be the wind speed measured on the surface of the ancient building.

8. The method for three-dimensional deformation dynamic monitoring of ancient building bracket nodes according to claim 7, characterized in that, The output layer is configured with a fully connected network, and a three-level warning probability vector (yellow, orange, and red) is generated using the Softmax function. The Softmax function is: ； in, The output layer generates a three-level warning probability vector, which is used to determine the probability of a yellow, orange, or red warning occurring at the current time t. The function is an activation function that transforms an input vector into a probability distribution; The weight matrix of the output layer, It is a parameter of a linear transformation used to change the state of the hidden layer. Mapped to the output space, and the weight matrix The dimension depends on the hidden layer state. The dimensions and the number of output warning categories; The state vector of the hidden layer at any given time; The bias vector of the output layer, the bias vector With weight matrix Together they affect the hidden layer state .

9. A three-dimensional deformation dynamic monitoring system for bracket sets in ancient buildings, used to implement the three-dimensional deformation dynamic monitoring method for bracket sets in ancient buildings as described in any one of claims 1-8, characterized in that, include: The acquisition module is used to acquire high-precision point cloud data of the surface of the bracket arch node using non-contact three-dimensional laser scanning technology; The fusion module is used to perform multi-source data registration and fusion based on point cloud data and oblique photogrammetry data collected in the same monitoring area to generate a digital twin model. A construction module is used to extract key feature points of the bracket nodes from the digital twin model using deep learning algorithms and to construct a four-dimensional motion trajectory model. The correction module is used to correct the deformation data in the four-dimensional motion trajectory model through an intelligent environmental interference filtering algorithm, and to distinguish between structural displacement and non-structural interference. The early warning module is used to trigger a multi-level early warning mechanism based on the corrected deformation data through dynamic threshold warning.

10. An electronic device, comprising at least one processor; and a memory communicatively connected to said at least one processor; characterized in that, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-8.

Images