A real scene three-dimensional model dynamic updating system

Abstract

The application discloses a kind of real scene three-dimensional model dynamic updating system, specifically related to three-dimensional model dynamic updating technical field, and the depth change detection network based on graph attention mechanism obtains candidate change area;Through incremental local reconstruction and preliminary insertion simulation, in combination with two types of high-precision evaluation indexes, namely, redundant reconstruction abnormality coefficient and structure damage pollution coefficient, the reconstruction effect and insertion quality are quantified in real time, and the potential structure continuity damage risk is accurately captured;Secondly, the model updating degradation objective function and degradation index evaluation mechanism are constructed, so that the system can quantitatively monitor the risk of each update behavior, and timely identify the closed-loop trend of "misjudgment-redundant update-structure pollution-again misjudgment";Finally, based on the fluctuation control strategy of updating degradation index, the automatic freezing and pending review marking of high-risk candidate area are realized, and the continuous promotion of safe update path is realized, to ensure the seamless connection of model geometry and texture.

Description

Technical Field

[0001] This invention relates to the field of dynamic updating technology for 3D models, and more specifically, to a dynamic updating system for real-world 3D models. Background Technology

[0002] With the widespread application of real-scene 3D modeling technology in fields such as digital twin cities, digital preservation of cultural relics, and intelligent operation and maintenance, higher demands are placed on the dynamic updating capabilities of 3D models. To improve the timeliness and consistency with reality, existing technologies mostly adopt incremental update methods, that is, periodically collecting urban or scene data through multi-source sensors and performing local reconstruction and insertion updates for areas that have changed. However, in actual operation, the system often faces the risk of overlapping issues of misjudgment of changed areas and destruction of structural logic, forming a long-neglected problem of a coupling chain between structural contamination and false evolution.

[0003] Specifically, current change detection modules often misjudge short-term, non-structural changes as structural alterations when dealing with factors such as changes in lighting, dynamic occlusions (e.g., pedestrians, vehicles, tree shadows), and inconsistent data collection. This triggers unnecessary update operations. Furthermore, during redundant reconstruction and the insertion of local models into the original global 3D model, failure to maintain structural consistency can easily disrupt the geometric continuity of the original model, causing issues such as discontinuities, floating, and overlapping. More seriously, this structural contamination is often re-identified as "new changes" by subsequent change detection modules, leading to repeated updates and forming a closed loop of update degradation: "misjudgment—redundant update—structural contamination—re-misjudgment." Therefore, a dynamic update method for real-world 3D models with robust change perception, structural consistency assurance capabilities, and contamination suppression mechanisms is urgently needed to effectively break this deterioration coupling chain and improve the overall stability and quality of model updates. Summary of the Invention

[0004] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a dynamic update system for real-scene 3D models to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A dynamic updating system for a real-scene 3D model includes a region perception module, a region reconstruction insertion module, a reconstruction damage module, an update deterioration assessment module, and an update decision control module.

[0007] The region awareness module is used to acquire multi-source real-scene datasets through multiple channels, preprocess the multi-source real-scene datasets, and use a deep change detection network based on graph attention mechanism to obtain candidate change regions.

[0008] The region reconstruction insertion module is used to perform incremental local 3D reconstruction and local reconstruction model insertion simulation on candidate changed regions.

[0009] The reconstruction damage module is used to obtain redundant reconstruction information of candidate change regions and structural consistency information of local reconstruction model insertion. The redundant reconstruction information includes a redundant reconstruction anomaly coefficient, and the structural consistency information includes a structural damage contamination coefficient.

[0010] The updated degradation assessment module is used to construct the objective function of model update degradation based on the redundancy reconstruction anomaly coefficient and the structural damage pollution coefficient, obtain the model update degradation index, and assess the degradation risk of the model update process.

[0011] The update decision control module is used to control and optimize update decisions based on the degradation risk during the model update process.

[0012] In a preferred embodiment, the logic for obtaining the redundancy reconstruction anomaly coefficients is as follows:

[0013] For each candidate change region, disparity matching is performed on the M viewpoint image pairs corresponding to the candidate change region to extract the disparity tensor of the candidate change region. ;

[0014] The parallax fluctuation value is obtained by extracting the magnitude of parallax variation under different viewpoints. ;

[0015] Mapping parallax fluctuation values ​​to spatial disturbance indices ;

[0016] At T historical moments, record the probability response values ​​of regions detected as candidate change regions to construct a probability response value sequence;

[0017] Calculate the polarization degree of the detected candidate change region ;

[0018] Mapping polarization degree to time fluctuation response index ;

[0019] Extract point cloud sets from candidate changed regions during reconstruction;

[0020] For each point cloud neighborhood Calculate the local covariance matrix ;

[0021] For the local covariance matrix Perform eigenvalue calculation and sort the eigenvalues ​​in descending order. Calculate the local degradation ratio : ,in Let them represent the local covariance matrices respectively. The eigenvalues; mapping the local degradation ratio to the regional geometric degradation index. : ;

[0022] The redundancy reconstruction anomaly coefficient is calculated based on the spatial disturbance index, the temporal fluctuation response index, and the regional geometric degradation index. : ,in These represent the preset proportional coefficients for the spatial disturbance index, the temporal fluctuation response index, and the regional geometric degradation index, respectively. All are greater than 0.

[0023] In a preferred embodiment, the logic for obtaining the structural damage pollution coefficient is as follows:

[0024] Reconstruct the model with local redundancy Compared with the original global 3D model The insertion interface is marked as ; Obtain the set of boundary points , The number of boundary points, each boundary point Carrying normal vector curvature Spatial position vector ;

[0025] For each boundary point Calculate the normal perturbation : Thus, the discrete perturbation vector field is obtained;

[0026] For each boundary point Calculate curvature perturbation : Thus, the discrete curvature perturbation field is obtained;

[0027] Construct a spatial disturbance propagation graph based on the set of boundary points. ,in , Given a set of edges, obtain the edge weights based on their spatial location vectors. ;

[0028] Using the normal perturbation and curvature perturbation as input signals, perform a perturbation propagation convolution operation: ;

[0029] Obtain the global maximum perturbation concentration : ;

[0030] Calculate the spatial discrete value of the disturbance concentration : ,in This represents the average concentration of disturbances. ;

[0031] Calculate the pollution coefficient of structural damage : .

[0032] In a preferred embodiment, an objective function for model update degradation is constructed based on the redundancy reconstruction anomaly coefficient and the structural damage contamination coefficient, and the model update degradation index is obtained. The objective function for model update degradation is as follows: In the formula Update the degradation index for the model. To redundancy, reconstruct anomaly coefficients. The pollution coefficient is the structural damage factor. These represent the preset proportional coefficients for the redundancy reconstruction anomaly coefficient and the structural damage contamination coefficient, respectively. All are greater than 0.

[0033] In a preferred embodiment, the update decision is controlled and optimized based on the degradation risk of the model update process, as follows:

[0034] Based on the objective function of model update degradation, obtain the model update degradation index after each round of model update and construct an update degradation risk set: ,in Let be the model update degradation index after the d-th round of model updates. Update the total number of rounds for the model;

[0035] The volatility of the degradation index is updated based on the updated degradation risk set calculation model. : ,in Update the average value of the degradation index for the model. .

[0036] In a preferred embodiment, the fluctuation value of the model update degradation index is compared with a preset fluctuation threshold to control and optimize the update decision, as follows:

[0037] If the fluctuation value of the model update degradation index is greater than the fluctuation threshold, the candidate change region currently participating in the local 3D reconstruction and the local reconstruction model insertion simulation will be marked as the region to be reviewed. At the same time, the update operation will be frozen and abnormal metadata will be recorded for manual review.

[0038] If the volatility of the model's updated degradation index is less than or equal to the volatility threshold, the current update strategy will continue.

[0039] The technical effects and advantages of this invention are as follows:

[0040] 1. This invention utilizes a graph attention-based depth change detection network to effectively distinguish between real structural changes and false interferences such as short-term lighting and dynamic occlusion, thereby acquiring candidate change regions. Secondly, through incremental local reconstruction and preliminary insertion simulation, combined with two high-precision evaluation indicators—redundant reconstruction anomaly coefficient and structural damage contamination coefficient—the reconstruction effect and insertion quality are quantified in real time, accurately capturing potential risks of structural continuity damage. Thirdly, a model update degradation objective function and degradation index evaluation mechanism are constructed, enabling the system to quantitatively monitor the risks of each round of updates and promptly identify the closed-loop trend of "misjudgment—redundant update—structural contamination—re-misjudgment." Finally, based on a fluctuation control strategy for the update degradation index, automatic freezing and review marking of high-risk candidate regions are achieved, along with continuous advancement of safe update paths, ensuring seamless integration of model geometry and texture. This avoids structural defects such as discontinuities, floating, and interpenetration common in traditional incremental updates. Through dynamic risk feedback and decision optimization, the efficiency and long-term reliability of model updates in application scenarios such as digital twin cities, cultural relic protection, and intelligent operation and maintenance are significantly improved. Attached Figure Description

[0041] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings;

[0042] Figure 1 This is a flowchart of the system according to an embodiment of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] Example: The present invention provides, as follows Figure 1 The system shown is a dynamic update system for a real-scene 3D model, including a region perception module, a region reconstruction and insertion module, a reconstruction and destruction module, an update and deterioration assessment module, and an update decision control module.

[0045] The region awareness module is used to acquire multi-source real-scene datasets through multiple channels, preprocess the multi-source real-scene datasets, and use a deep change detection network based on graph attention mechanism to obtain candidate change regions.

[0046] The region reconstruction insertion module is used to perform incremental local 3D reconstruction and local reconstruction model insertion simulation on candidate changed regions.

[0047] The reconstruction damage module is used to obtain redundant reconstruction information of candidate change regions and structural consistency information of local reconstruction model insertion. The redundant reconstruction information includes a redundant reconstruction anomaly coefficient, and the structural consistency information includes a structural damage contamination coefficient.

[0048] The updated degradation assessment module is used to construct the objective function of model update degradation based on the redundancy reconstruction anomaly coefficient and the structural damage pollution coefficient, obtain the model update degradation index, and assess the degradation risk of the model update process.

[0049] The update decision control module is used to control and optimize update decisions based on the degradation risk of the model update process;

[0050] In the area perception module, multi-source real-scene datasets covering the target scene are acquired through multi-channel perception devices (such as UAV imagery, vehicle-mounted LiDAR, ground cameras, fixed video surveillance, etc.). All acquisition devices synchronize timestamps through GNSS receiving modules or NTP protocols to ensure that data acquisition is within a unified reference time window. All raw acquisition frame data are encapsulated with metadata such as timestamp, device ID, pose matrix, coordinate reference system number, etc.

[0051] Preprocessing of multi-source real-scene datasets includes:

[0052] Coordinate transformation and unified projection:

[0053] Transform all raw coordinates collected by sensing devices (such as local UTM, vehicle-mounted LIDAR coordinate system, pixel image coordinates, etc.) to a unified global coordinate system (WGS84 / ENU, etc.).

[0054] A pixel-geographic mapping matrix is ​​used to map each pixel to a geographic location in the image.

[0055] The point cloud was registered as a rigid body using a transformation matrix.

[0056] Data quality enhancement and noise suppression:

[0057] The image is enhanced with CLAHE (limited contrast histogram equalization) and a dehazing algorithm to improve texture clarity.

[0058] Sparse downsampling is performed on the point cloud using a voxel grid filter to remove floating and isolated points;

[0059] Perform inter-frame optical flow consistency verification on consecutive frames in the image sequence to remove erroneous frames caused by occlusion or blurring.

[0060] The preprocessed multi-source real-scene dataset is used as network input, and a deep change detection network based on graph attention mechanism is used to obtain candidate change regions.

[0061] A deep change detection network based on graph attention mechanism refers to a network architecture that models the input data as a graph structure and learns the spatiotemporal differences between nodes through graph convolutional propagation and attention weight adjustment. Its network structure includes:

[0062] The graph construction module is used to model the acquisition scene as a heterogeneous graph, where nodes in the heterogeneous graph represent image blocks or point cloud sub-blocks, and edges represent spatial adjacency or semantic connection.

[0063] A graph attention encoder is used to further calculate the information propagation weights between nodes using a self-attention mechanism to transform an initial heterogeneous graph into an attention graph structure.

[0064] The dual-branch residual comparison module is used to compare the attention map structure constructed from past historical multi-source real-scene data and current multi-source real-scene data to generate a significant change map.

[0065] Saliency decoder: Used to restore the saliency map of changes to a spatial mask, obtaining a set of candidate change regions. ;

[0066] In the region reconstruction insertion module, local datasets are formed by cropping the image frame regions, point cloud voxels, and depth map windows covered by each candidate change region from the multi-source dataset. These datasets include image subframes, LiDAR point cloud slices, pose information, and timestamp metadata.

[0067] For each candidate region of change in the local dataset, the following incremental local 3D reconstruction simulation process is performed:

[0068] Step A1: Extract sparse keypoints (such as SIFT / SuperPoint) and feature descriptors from the image frame, and use epipolar geometry and RANSAC filtering to perform cross-view feature matching to obtain the initial matching map.

[0069] Step A2: Apply the SfM (Structure-from-Motion) method to recover the sparse 3D point set and camera pose from the matching map;

[0070] Step A3: Use multi-view stereo algorithms (such as PatchMatchNet, ACMM, etc.) to perform dense disparity estimation and point cloud fusion to generate a high-density local reconstruction mesh model.

[0071] Obtain the common reference points between the local reconstructed model and the original global 3D model, use the Iterative Closest Point (ICP) algorithm to perform fine registration of the local region, solve the rigid body transformation matrix to update the position of the local reconstructed model, extract the boundary of the local reconstructed model and the boundary of the original global 3D model to construct a boundary overlap region map in space, and calculate the boundary overlap degree. ,in For boundary overlap, For point distance similarity, The angle between the normal vectors, These represent the preset scaling factors for point distance similarity and the angle between normal vectors, respectively. (All values ​​are greater than 0) to determine if there is a potential conflict:

[0072] If the boundary overlap is greater than the preset boundary overlap threshold, then boundary stitching interpolation processing is performed to stitch the boundary between the local reconstructed model and the original global 3D model, thus completing the insertion simulation of the local reconstructed model.

[0073] If the boundary overlap is less than or equal to the preset boundary overlap threshold, then perform an insertion rejection simulation.

[0074] It should be noted that the above formulas are all dimensionless calculations. Commonly used methods for removing dimensions include Min-Max normalization and Z-Score standardization, which will not be elaborated here. The settings should be tailored to the specific circumstances. For example, an expert-empowered approach could be adopted, where experts in relevant fields are invited to determine the pre-defined proportions for each indicator through professional opinion surveys and comprehensive evaluations. It can be 0.5 or 0.5;

[0075] The reconstruction damage module is used to obtain redundant reconstruction information of candidate change regions and structural consistency information of local reconstruction model insertion. The redundant reconstruction information includes a redundant reconstruction anomaly coefficient, and the structural consistency information includes a structural damage contamination coefficient.

[0076] In this invention, the redundancy reconstruction anomaly coefficient is a core indicator used to quantify the risk of a locally changed region being misjudged as "needing an update" during dynamic 3D modeling, triggering unnecessary reconstruction. This coefficient comprehensively considers the response uncertainty during the change detection phase, the structural stability of the local region, the temporal continuity characteristics, and the effectiveness of the actual reconstruction. This effectively determines whether the "change" in a region is a real-world evolution or a false update caused by environmental disturbances, sensor errors, or model drift. A large coefficient usually indicates that although the region has been detected as a changed region multiple times, the reconstruction results consistently fail to form an effective supplementary structure, or multiple insertions may lead to structural contamination phenomena such as model boundary breaks, density drift, and topological conflicts, representing a typical redundant reconstruction pattern. Conversely, a smaller redundancy reconstruction anomaly coefficient typically indicates that the region's change response has clear spatiotemporal consistency, the reconstruction results are well-connected to the model structure, and the overall modeling quality is improved, representing a necessary and effective structural supplementation. Introducing the redundancy reconstruction anomaly coefficient into the degradation risk assessment of the 3D model update process firstly endows the model update decision with "source discrimination capability," enabling quantitative judgment on the authenticity and value of candidate change areas before performing local reconstruction. This significantly reduces the system's sensitivity to false changes caused by dynamic occlusion, environmental shadows, and periodic disturbances (such as water ripples and vegetation swaying). Secondly, through the degradation risk objective function jointly constructed with the structural damage contamination coefficient, the entire process of "contamination caused by redundant modeling" can be modeled and measured. This clarifies the potential quality fluctuation impact of each model update, thereby supporting a more robust model evolution mechanism. This effectively improves the structural consistency, geometric continuity, and visual stability of the real-world 3D model during long-term evolution, achieving high-quality, high-reliability multi-temporal 3D reconstruction.

[0077] The logic for obtaining the redundancy reconstruction anomaly coefficients is as follows:

[0078] For each candidate change region, there are M viewpoint image pairs ( Perform disparity matching to extract disparity tensors from candidate regions of change. ,in Image pixel coordinates, This represents the parallax from viewpoint i to viewpoint j;

[0079] It is used to measure the disparity of a pixel under different viewing angles;

[0080] The parallax fluctuation value is obtained by extracting the magnitude of parallax variation under different viewpoints. : ,in For average parallax, ;

[0081] Mapping parallax fluctuation values ​​to spatial disturbance indices : ;

[0082] At T historical moments, record the probability response values ​​of regions detected as candidate change regions, and construct a sequence of probability response values: , , Let be the probability response value at time t;

[0083] Calculate the polarization degree of the detected candidate change region : ,in This represents the average probability response value. , This is to prevent division by zero by a very small constant (generally taken as...). );

[0084] Mapping polarization degree to time fluctuation response index : ;

[0085] Extracting point cloud sets from candidate changed regions during reconstruction Each It is a three-dimensional point including three-dimensional spatial coordinates. This represents the number of point clouds;

[0086] For each point cloud neighborhood Calculate the local covariance matrix : ,in Let be the three-dimensional coordinate vector of the k-th point cloud in the neighborhood. Represents the transpose of a vector. ,in The number of point clouds in the neighborhood;

[0087] For the local covariance matrix Perform eigenvalue calculation and sort the eigenvalues ​​in descending order. Calculate the local degradation ratio : ,in Let them represent the local covariance matrices respectively. The eigenvalues; mapping the local degradation ratio to the regional geometric degradation index. : ;

[0088] The redundancy reconstruction anomaly coefficient is calculated based on the spatial disturbance index, the temporal fluctuation response index, and the regional geometric degradation index. : ,in These represent the preset proportional coefficients for the spatial disturbance index, the temporal fluctuation response index, and the regional geometric degradation index, respectively. All are greater than 0;

[0089] It should be noted that the above formulas are all dimensionless calculations. Commonly used methods for removing dimensions include Min-Max normalization and Z-Score standardization, which will not be elaborated here. The settings should be tailored to the specific circumstances. For example, an expert-empowered approach could be adopted, where experts in relevant fields are invited to determine the pre-defined proportions for each indicator through professional opinion surveys and comprehensive evaluations. It can be 0.4, 0.2, or 0.3;

[0090] In this invention, the structural damage contamination coefficient is an important evaluation index used to measure the degree of damage to the structural integrity and geometric continuity of the original global 3D model during the insertion of locally redundant reconstructed models. Its core function is to quantitatively model the local abnormal changes in the model caused by the insertion behavior from dimensions such as geometric topology, spatial continuity, and semantic consistency, thereby assisting the dynamic 3D model system in avoiding structural degradation and erroneous evolution risks during the update process. During the dynamic update of the 3D model, if candidate change regions are redundantly reconstructed and directly inserted into the global model, local damage phenomena such as structural breakage, surface interpenetration, hierarchical floating, and semantic conflicts are highly likely to occur. Once such damage occurs, it will propagate a series of subsequent effects in the model space, leading to problems such as increased geometric insolvability, blurred semantic boundaries, and reduced change detection accuracy. A large structural damage contamination coefficient indicates that during the model insertion process, the newly reconstructed local data and the original model have significant inconsistencies in geometry, boundary continuity, or spatial connectivity, often manifesting as typical distortion features such as model boundary breakage, mesh non-manifolding, and high-frequency distortion of point clouds. Inserting such regions can easily create model "cracks" or "contamination zones," disrupting the overall coherence of the 3D structure and causing a chain reaction of degradation in subsequent tasks such as change detection, navigation, and recognition. It can even form a closed-loop structural contamination propagation chain. Conversely, a smaller structural damage contamination coefficient indicates that the local reconstructed model has high structural compatibility with the original 3D model during insertion. Its spatial geometry connects smoothly with the original model's boundaries, its normal direction is continuous, and its semantic attributes are highly consistent. After insertion, the overall quality of the model will not decrease; it may even improve local accuracy and integrity. By using the structural damage contamination coefficient as a quantitative indicator of structural risk in the dynamic 3D update framework, potential damage impacts can be dynamically perceived. Based on this, an update degradation risk objective function can be further constructed, which, together with the redundant reconstruction anomaly coefficient, constitutes a two-factor decision model. This model outputs a global model update degradation index, and update decisions are controlled based on this index, ensuring that dynamic updates are only performed under conditions of low redundancy risk and mild structural contamination. It effectively avoids the closed loop of "misjudgment - redundant reconstruction - structural damage - detection and re-misjudgment" and significantly improves the long-term stability, scalability and semantic consistency of the real-scene 3D model, which is of great significance for building a high-precision and evolvable city-level digital twin.

[0091] The logic for obtaining the structural damage pollution coefficient is as follows:

[0092] Reconstruct the model with local redundancy Compared with the original global 3D model The insertion interface is marked as ,in and These represent the boundary representations of the locally redundant reconstruction model and the original global 3D model, respectively; thus, the set of boundary points is obtained. , The number of boundary points, each boundary point Carrying normal vector curvature Spatial position vector ;

[0093] For each boundary point Calculate the normal perturbation : ,in For the boundary point The normal vector of the locally redundant reconstruction model. For the boundary point The discrete perturbation vector field is obtained by taking the normal vector of the original global 3D model. ;

[0094] For each boundary point Calculate curvature perturbation : ,in For the boundary point The curvature of the locally redundant reconstruction model For the boundary point The curvature of the original global 3D model is used to obtain the discrete curvature perturbation field. ;

[0095] Construct a spatial disturbance propagation graph based on the set of boundary points. ,in , Given a set of edges, obtain the edge weights based on their spatial location vectors. : ,in Boundary point The spatial position vector at that location, Boundary point The spatial position vector at that location, For Euclidean distance These are parameters for controlling the diffusion scale.

[0096] It should be noted that the diffusion scale control parameter is used to adjust the edge weight function in the spatial perturbation propagation diagram. The propagation attenuation rate determines the range of the disturbance propagation in space. Common methods for obtaining it include local scale estimation based on K-nearest neighbor distance and local point set standard deviation estimation.

[0097] Using the normal perturbation and curvature perturbation as input signals, perform a perturbation propagation convolution operation: ,in Represents boundary points The concentration of disturbances. Boundary point K-nearest neighbors in a spatial disturbance propagation diagram;

[0098] Obtain the global maximum perturbation concentration : ;

[0099] Calculate the spatial discrete value of the disturbance concentration : ,in This represents the average concentration of disturbances. ;

[0100] Calculate the pollution coefficient of structural damage : ;

[0101] It should be noted that the above formulas are all dimensionless calculations. Commonly used methods for removing dimensions include Min-Max normalization and Z-Score standardization, which will not be elaborated here.

[0102] The updated degradation assessment module is used to construct the objective function of model update degradation based on the redundancy reconstruction anomaly coefficient and the structural damage pollution coefficient, obtain the model update degradation index, and assess the degradation risk of the model update process.

[0103] Based on the redundancy reconstruction anomaly coefficient and the structural damage contamination coefficient, an objective function for model update degradation is constructed, and the model update degradation index is obtained. The objective function for model update degradation is as follows: In the formula Update the degradation index for the model. To redundancy, reconstruct anomaly coefficients. The pollution coefficient is the structural damage factor. These represent the preset proportional coefficients for the redundancy reconstruction anomaly coefficient and the structural damage contamination coefficient, respectively. All are greater than 0;

[0104] It should be noted that the above formulas are all dimensionless calculations. Commonly used methods for removing dimensions include Min-Max normalization and Z-Score standardization, which will not be elaborated here. The settings should be tailored to the specific circumstances. For example, an expert-empowered approach could be adopted, where experts in relevant fields are invited to determine the pre-defined proportions for each indicator through professional opinion surveys and comprehensive evaluations. It can be 0.5 or 0.5;

[0105] As can be seen from the above calculation expressions, the larger the redundancy reconstruction anomaly coefficient and the larger the structural damage contamination coefficient, the larger the model update degradation index, indicating that the model faces a higher risk of quality degradation during this round of updates. Conversely, the smaller the redundancy reconstruction anomaly coefficient and the smaller the structural damage contamination coefficient, the smaller the model update degradation index, indicating that the local update behavior has almost no negative impact on the structural integrity and semantic accuracy of the original 3D model, and the model update process has high quality stability and reliability.

[0106] The update decision control module is used to control and optimize update decisions based on the degradation risk during the model update process, as follows:

[0107] Based on the objective function of model update degradation, obtain the model update degradation index after each round of model update and construct an update degradation risk set: ,in Let be the model update degradation index after the d-th round of model updates. Update the total number of rounds for the model;

[0108] The volatility of the degradation index is updated based on the updated degradation risk set calculation model. : ,in Update the average value of the degradation index for the model. ;

[0109] The fluctuation value of the model update degradation index is compared with the preset fluctuation threshold, and the update decision is controlled and optimized as follows:

[0110] If the fluctuation value of the model update degradation index is greater than the fluctuation threshold, it indicates that there are significant fluctuations in the degradation index between different rounds of the current model update process. This is manifested as the system's unstable identification of changed areas and large fluctuations in the quality of update results. There is a high possibility of structural pollution caused by misjudgment of candidate areas or excessive redundant reconstruction. The system is in an uncertain evolutionary state. Continuing to update automatically will amplify the pollution risk. The candidate changed areas currently participating in the local 3D reconstruction and the local reconstruction model insertion simulation are marked as areas to be reviewed. At the same time, the update operation is frozen and abnormal metadata is recorded for manual review or intelligent correction.

[0111] If the fluctuation value of the model update degradation index is less than or equal to the fluctuation threshold, it means that the model degradation risk remains in a low and stable range during the current multiple updates, the redundancy reconstruction control and structural fusion strategies have a certain robustness, the update operation is relatively reliable, and the current update strategy should continue to be executed.

[0112] This invention utilizes a graph attention-based depth change detection network to effectively distinguish between real structural changes and false interferences such as short-term lighting and dynamic occlusion, thereby acquiring candidate change regions. Secondly, through incremental local reconstruction and preliminary insertion simulation, combined with two high-precision evaluation indicators—redundant reconstruction anomaly coefficient and structural damage contamination coefficient—the reconstruction effect and insertion quality are quantified in real time, accurately capturing potential risks of structural continuity damage. Thirdly, a model update degradation objective function and degradation index evaluation mechanism are constructed, enabling the system to quantitatively monitor the risks of each round of updates and promptly identify the closed-loop trend of "misjudgment—redundant update—structural contamination—re-misjudgment." Finally, based on a fluctuation control strategy for the update degradation index, high-risk candidate regions are automatically frozen and marked for review, and safe update paths are continuously advanced, ensuring seamless integration of model geometry and texture. This avoids structural defects such as discontinuities, floating, and interpenetration common in traditional incremental updates. Through dynamic risk feedback and decision optimization, the efficiency and long-term reliability of model updates in application scenarios such as digital twin cities, cultural relic protection, and intelligent operation and maintenance are significantly improved.

[0113] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0114] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0115] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0116] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0117] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A dynamic updating system for a real-scene 3D model, characterized in that: It includes a regional perception module, a regional reconstruction insertion module, a reconstruction damage module, an update degradation assessment module, and an update decision control module; The region awareness module is used to acquire multi-source real-scene datasets through multiple channels, preprocess the multi-source real-scene datasets, and use a deep change detection network based on graph attention mechanism to obtain candidate change regions. The region reconstruction insertion module is used to perform incremental local 3D reconstruction and local reconstruction model insertion simulation on candidate changed regions. The reconstruction damage module is used to obtain redundant reconstruction information of candidate change regions and structural consistency information of local reconstruction model insertion. The redundant reconstruction information includes a redundant reconstruction anomaly coefficient, and the structural consistency information includes a structural damage contamination coefficient. The updated degradation assessment module is used to construct the objective function of model update degradation based on the redundancy reconstruction anomaly coefficient and the structural damage pollution coefficient, obtain the model update degradation index, and assess the degradation risk of the model update process. The update decision control module is used to control and optimize update decisions based on the degradation risk of the model update process; Based on the redundancy reconstruction anomaly coefficient and the structural damage contamination coefficient, an objective function for model update degradation is constructed, and the model update degradation index is obtained. The objective function for model update degradation is as follows: In the formula Update the degradation index for the model. To redundancy, reconstruct anomaly coefficients. The pollution coefficient is the structural damage factor. These represent the preset proportional coefficients for the redundancy reconstruction anomaly coefficient and the structural damage contamination coefficient, respectively. All are greater than 0; The update decision is controlled and optimized based on the degradation risk of the model update process, as follows: Based on the objective function of model update degradation, obtain the model update degradation index after each round of model update and construct an update degradation risk set: ,in Let be the model update degradation index after the d-th round of model updates. Update the total number of rounds for the model; The volatility of the degradation index is updated based on the updated degradation risk set calculation model. : ,in Update the average value of the degradation index for the model. ; The fluctuation value of the model update degradation index is compared with the preset fluctuation threshold, and the update decision is controlled and optimized as follows: If the fluctuation value of the model update degradation index is greater than the fluctuation threshold, the candidate change region currently participating in the local 3D reconstruction and the local reconstruction model insertion simulation will be marked as the region to be reviewed. At the same time, the update operation will be frozen and abnormal metadata will be recorded for manual review. If the volatility of the model's updated degradation index is less than or equal to the volatility threshold, the current update strategy will continue.

2. The real-scene 3D model dynamic update system according to claim 1, characterized in that: The logic for obtaining the redundancy reconstruction anomaly coefficients is as follows: For each candidate change region, disparity matching is performed on the M viewpoint image pairs corresponding to the candidate change region to extract the disparity tensor of the candidate change region. ; The parallax fluctuation value is obtained by extracting the magnitude of parallax variation under different viewpoints. ; Mapping parallax fluctuation values ​​to spatial disturbance indices ; At T historical moments, record the probability response values ​​of regions detected as candidate change regions to construct a probability response value sequence; Calculate the polarization degree of the detected candidate change region ; Mapping polarization degree to time fluctuation response index ; Extract point cloud sets from candidate changed regions during reconstruction; For each point cloud neighborhood Calculate the local covariance matrix ; For the local covariance matrix Perform eigenvalue calculation and sort the eigenvalues ​​in descending order. Calculate the local degradation ratio : ,in Let them represent the local covariance matrices respectively. eigenvalues; Mapping the local degradation ratio to the regional geometric degradation index : ; The redundancy reconstruction anomaly coefficient is calculated based on the spatial disturbance index, the temporal fluctuation response index, and the regional geometric degradation index. : ,in These represent the preset proportional coefficients for the spatial disturbance index, the temporal fluctuation response index, and the regional geometric degradation index, respectively. All are greater than 0.

3. The dynamic updating system for a real-scene 3D model according to claim 1, characterized in that: The logic for obtaining the structural damage pollution coefficient is as follows: Reconstruct the model with local redundancy Compared with the original global 3D model The insertion interface is marked as ; Obtain the set of boundary points , The number of boundary points, each boundary point Carrying normal vector curvature Spatial position vector ; For each boundary point Calculate the normal perturbation : Thus, the discrete perturbation vector field is obtained; For each boundary point Calculate curvature perturbation : Thus, the discrete curvature perturbation field is obtained; Construct a spatial disturbance propagation graph based on the set of boundary points. ,in , Given a set of edges, obtain the edge weights based on their spatial location vectors. ; Using the normal perturbation and curvature perturbation as input signals, perform a perturbation propagation convolution operation: ; Obtain the global maximum perturbation concentration : ; Calculate the spatial discrete value of the disturbance concentration : ,in This represents the average concentration of disturbances. ; Calculate the pollution coefficient of structural damage : .

Images