Historic building three-dimensional model digitization reconstruction method

Abstract

The application relates to the field of building energy-saving simulation technology and discloses a historical building three-dimensional model digital reconstruction method. The method comprises the following steps: extracting mesoscopic geometric configuration, material distribution topology and environmental parameters of a system, and constructing an initial simulation model; performing defect analysis on the model, identifying structure incoherent areas and physical field mutation nodes, and generating a model defect set; matching physical compensation strategies associated with each defect from a preset knowledge base, and calculating corresponding dynamic compensation factors for each defect to form a compensation factor set; and using the set to perform real-time dynamic correction on the initial model to obtain a high-fidelity mesoscopic simulation model. The method improves the accuracy of heat transfer simulation results and the reliability of the model through accurate positioning of defects, intelligent matching of compensation strategies and adaptive calculation of correction parameters.

Description

Technical Field

[0001] This invention relates to the field of building energy conservation simulation technology, specifically a method for digital reconstruction of three-dimensional models of historical buildings. Background Technology

[0002] In the field of historical building conservation and digital reconstruction, constructing high-fidelity 3D digital models is fundamental for accurate structural analysis, virtual restoration, and morphological evolution research. Conventional 3D reconstruction methods for historical buildings typically rely on point cloud scanning and texture mapping of the current state, generating geometric models through surface modeling techniques. These methods treat the collected surface data as a whole for 3D reconstruction, and the model construction process largely depends on the operator's experience and judgment of the historical building's form. They lack effective means to analyze and reconstruct the complex structural hierarchical relationships hidden within the building, formed by the superposition or modification of different historical periods.

[0003] Existing technical solutions have shortcomings. The model construction lacks historical logical support, making it difficult to accurately identify and locate areas of spatiotemporal information inconsistencies caused by renovations, material replacements, or structural damage throughout history. These inconsistencies include problems such as incorrect splicing of components from different periods, missing structural logic, or inconsistencies in era characteristics. Furthermore, the model optimization process is static and fixed, unable to adaptively adjust based on the specific types of logical inconsistencies and their spatial distribution. This results in insufficient fidelity in representing the authenticity of architectural history and the rationality of structural construction, leading to deviations between the model results and the actual historical construction logic.

[0004] Current reconstruction processes suffer from problems such as fuzzy historical trace analysis and rigid model logic verification when dealing with the complex spatiotemporal information and structural topology of historical buildings. This hinders the accurate revelation of the evolutionary mechanisms of a building's life cycle through digital means and also limits the reliability of in-depth historical research and conservation decisions based on 3D models. Summary of the Invention

[0005] The purpose of this invention is to provide a method for digital reconstruction of three-dimensional models of historical buildings, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a method for digital reconstruction of three-dimensional models of historical buildings, the method comprising:

[0007] Identify and analyze historical traces on the surface of the target historical building, and output the evolution path of the building structure;

[0008] Based on the building structure evolution path, the currently collected point cloud data and texture image data are subjected to layered parsing and asynchronous alignment to generate layered aligned building space data;

[0009] The hierarchical aligned architectural space data is used to drive a parametric generation model, which gradually synthesizes a preliminary three-dimensional voxel model of the target historical building through an iterative feedback mechanism.

[0010] A rule base based on historical construction principles is introduced to perform logical rationality verification on the preliminary three-dimensional voxel model, generating a three-dimensional model with logical verification marks.

[0011] Based on the logical verification mark, a new round of data collection requests is initiated for areas in the 3D model that have contradictions or missing data, triggering a targeted supplementary data collection process;

[0012] Integrate new data from the targeted supplementary data acquisition process with existing data, perform multiple rounds of iterative model optimization and refinement operations until the preset integrity and consistency thresholds are met, and output the optimized three-dimensional digital model;

[0013] Based on the aforementioned three-dimensional digital model, the evolution of architectural forms in different historical periods is simulated, and corresponding virtual restoration plans are generated.

[0014] Preferably, the identification and analysis of historical traces on the surface of the target historical building, and the output of the building structure evolution path, include:

[0015] Obtain the historical trace identification results, and based on the historical trace identification results, deduce the structural change hypothesis of the target historical building in the time dimension, and generate a set of structural change hypotheses;

[0016] By combining historical drawing scan data, the set of structural change assumptions is cross-validated and corrected, and a validated building structure evolution path is output.

[0017] The identification and analysis of historical traces on the surface of the target historical building includes:

[0018] The spectral reflectance distribution map of the building surface is obtained by using a high-resolution multispectral scanning device;

[0019] In the spectral reflectance characteristic distribution map, based on the differences in reflectance characteristics of different materials in different bands, areas suspected of being artificially modified, naturally weathered, or damaged are segmented and marked as initial trace areas;

[0020] For each of the initial trace regions, perform a micro-scale three-dimensional topography scan to extract the micro-undulation data and texture direction data of its surface geometry;

[0021] The micro-undulation data and texture direction data are matched and similarity calculated item by item with a pre-built historical construction method trace feature library.

[0022] Based on the matching and calculation results, one or more possible historical cause labels and corresponding confidence levels are assigned to each initial trace area. The historical cause labels include changes in masonry methods, material repairs, and structural crack propagation.

[0023] All tagged and confidence-rated trace regions are aggregated to form the historical trace identification results, which record the spatial location, geometric features, and inferred causes of the traces.

[0024] Preferably, the derivation of the structural changes of the target historical building over time includes:

[0025] The spatiotemporal distribution patterns and superposition order of various types of traces in the historical trace identification results are analyzed.

[0026] Based on the superposition order of the traces, the relative time sequence of the construction, renovation or damage of different parts of the building can be inferred;

[0027] By combining the historical causal tags, traces with the same or similar causes and spatiotemporal correlations are clustered to form several sets of trace events.

[0028] For each set of trace events, a descriptive event hypothesis model is constructed, which includes the event type, scope of impact, relative timing of occurrence, and possible structural state changes.

[0029] All event hypothesis models are sorted and logically connected according to the inferred relative time series to construct a hypothesis chain describing the entire process of the building's evolution from its original state through various events to its current state, which is the set of structural change hypotheses.

[0030] Preferably, the step of cross-validating and revising the set of structural change hypotheses by combining historical drawing scan data includes:

[0031] The acquired historical drawing scan data is vectorized and layered to extract building outlines, structural component lines, and annotation information.

[0032] The extracted historical drawing elements are roughly matched with the point cloud projection outline of the current building in terms of spatial location and scale to establish a coordinate system correspondence.

[0033] Under the coordinate system correspondence, the structural state change corresponding to each event hypothesis model in the set of structural change hypotheses is compared with the building form or component information of different periods marked in the historical drawings;

[0034] When there is a conflict between the event hypothesis model and historical drawing information, the credibility of the event hypothesis model will be downgraded or a correction process will be initiated based on the authority weight of the drawing information.

[0035] When the event hypothesis model is supported by historical drawing information, its credibility is increased, and the precise dimensions and shape information in the drawings are incorporated into the event hypothesis model.

[0036] After comparison and adjustment, a time-stamped evolution path of the building structure is output, which integrates trace reasoning and drawing evidence.

[0037] Preferably, the step of performing layered parsing and asynchronous alignment of the currently acquired point cloud data and texture image data includes:

[0038] Based on the structural characteristics of different periods in the building structure evolution path, the point cloud data is decomposed into point cloud subsets corresponding to different construction periods or structural levels according to spatial location and geometric features;

[0039] Each subset of the point cloud is independently reconstructed into a three-dimensional surface to generate a series of triangular mesh fragments representing surfaces of different periods or hierarchical structures.

[0040] Simultaneously, the texture image data is divided into texture image groups that roughly correspond to each triangular mesh segment based on its shooting angle and content;

[0041] Establish the projection mapping relationship between each triangular mesh fragment and its corresponding texture image group, and accurately map the texture to the corresponding mesh surface through feature point matching and perspective transformation;

[0042] In this process, triangular mesh fragments belonging to different periods or levels are allowed to partially overlap or have gaps in space. Their relative positional relationship is constrained by the logic of the architectural structure evolution path, and finally the layered aligned architectural space data is generated. The architectural space data contains multiple structural surfaces and their textures that are separated in time and space but logically related.

[0043] Preferably, the method of using hierarchically aligned architectural space data to drive parametric generation of the model includes:

[0044] Each triangular mesh segment and its spatiotemporal label in the hierarchically aligned architectural spatial data are fed into the parameterized generation model as an input sequence;

[0045] The core of the parametric generation model is a voxel generation network with a spatiotemporal attention mechanism. The voxel generation network parses the input sequence step by step and, according to the building structure evolution path, generates voxel structures that are added or changed in subsequent periods in the three-dimensional voxel space, starting from the basic form in the earliest period.

[0046] In each generation step, the network outputs an intermediate voxel model for the current step and compares the intermediate voxel model with the corresponding hierarchically aligned architectural space data.

[0047] The difference signals generated by the comparison are fed back to the network's attention mechanism to adjust the parameters of the next generation step, so that the generated voxel structure continuously approximates the real, layered architectural space data during the iteration, and finally converges to generate the preliminary three-dimensional voxel model.

[0048] Preferably, the step of introducing a rule base based on historical construction principles to perform logical rationality verification on the preliminary three-dimensional voxel model includes:

[0049] The rule base based on historical construction principles includes constraints related to the construction logic, material mechanics logic, and morphological aesthetic logic of the target historical building, which are related to its culture, region, and era.

[0050] The preliminary three-dimensional voxel model is converted into a symbolic structural relationship diagram suitable for rule-based reasoning. In the diagram, nodes represent building components, and edges represent connections, supports, or spatial relationships between components.

[0051] The symbolic structural relationship diagram is matched one by one with the constraints in the rule base;

[0052] When a relationship that violates the constraints is detected in the structural relationship diagram, such as a cantilevered component lacking a lower support or an impossible overlapping relationship between components from different periods, the aforementioned logical verification mark is added to the component or relationship that violates the rules. The mark types include logical conflict, suspected missing, and inconsistency in time period.

[0053] After traversing the entire structural relationship diagram, the three-dimensional model with logical verification marks is output. The three-dimensional model marks all areas where logical problems may exist in a visual or data form.

[0054] Preferably, initiating a new round of data collection requests for areas in the 3D model that contain contradictions or missing data includes:

[0055] The system automatically analyzes the type and spatial distribution of the logical check tags;

[0056] For areas marked as logically conflicting or suspected missing, the system generates targeted data acquisition task instructions. These instructions specify in detail the types of data that need to be collected, the key areas to focus on, the recommended acquisition equipment, and the acquisition perspective.

[0057] The data acquisition task instruction is encapsulated into the data acquisition request and sent to the connected data acquisition scheduling terminal or platform;

[0058] The targeted supplementary data acquisition process is triggered. Depending on the request content, the data acquisition process may include laser scanning of concealed parts of the building at specific angles, radar detection of suspected internal hollow areas or boundaries between different materials, or higher resolution photogrammetry of specific structural nodes.

[0059] Preferably, the system automatically analyzes the type and spatial distribution of the logical check mark, including:

[0060] The system first reads the attribute information of all logical verification tags, including the tag type, the building component identifier to which the tag is attached, and the coordinates of the tag in three-dimensional space;

[0061] Logical check tags are categorized according to their tag type. The categories include logical conflict, suspected missing tags, and outdated tags.

[0062] For each classification category, the system calculates the spatial density distribution of the label of that category on the surface of the 3D model and generates a density heatmap using a kernel density estimation algorithm;

[0063] The system identifies high-density regions in the density heatmap and delineates the boundaries between high-density and low-density regions.

[0064] By combining the marker type and spatial density distribution, the system generates an analysis report, which indicates the degree of aggregation and spatial distribution pattern of various markers, providing a priority basis for subsequent data collection requests.

[0065] Preferably, the multi-round iterative model optimization and refinement operation includes:

[0066] After receiving new data from the targeted supplementary data acquisition process, the new data is first merged with the previously corresponding layered aligned building space data to update the layered aligned building space data.

[0067] Using the updated hierarchically aligned architectural space data, repeat the steps of driving parametric model generation and performing logical rationality verification to generate a new version of the 3D digital model.

[0068] Check the number and severity of logic check tags left on the new version of the model;

[0069] If the logical verification flags fail to decrease to below the preset threshold, or if new contradictions are discovered, a new round of data collection requests will be generated based on the new version model and the legacy flags, and the next iteration will be initiated.

[0070] If the logical verification flags have met the preset integrity and consistency thresholds, the model is determined to have converged, and the current version of the 3D digital model is used as the final output.

[0071] During the iteration process, each model update and reassessment of historical traces may fine-tune the earlier architectural evolution path, forming a closed-loop optimization from data to reasoning.

[0072] Compared with the prior art, the beneficial effects of the present invention are:

[0073] By performing model defect analysis, incoherent structural regions and abrupt physical field changes in the initial simulation model are specifically identified, generating a structured set of model defects. This transforms the model's defects from vague overall deviations into a series of specific problems that can be clearly described and located, providing clear targets for subsequent precise interventions and overcoming the limitations of conventional methods that offer vague defect descriptions and unclear locations.

[0074] Based on the generated set of model defects, the system matches the physical compensation strategy associated with each specific defect from a pre-defined knowledge base. This process transforms trial-and-error correction, which relies on human experience, into automated and intelligent strategy invocation based on the mapping relationship between "defect features" and "compensation strategies." For defects of different natures, such as structural inconsistencies or abrupt changes in physical fields, the system can automatically invoke corresponding, validated specialized physical models or mathematical processing methods, achieving precise matching between correction strategies and defect types, and improving the scientific rigor and purposefulness of correction actions.

[0075] Based on the matching of specific physical compensation strategies, a dynamic compensation factor is further calculated for each defect, forming a set of compensation factors used for real-time dynamic correction of the model. The calculation of dynamic compensation factors means that the correction parameters are no longer fixed values, but can be generated in real time based on contextual information such as the local environment and severity of the defect. Applying this set of factors for correction achieves non-uniform, adaptive adjustment to defects of different properties and in different regions of the model. This allows the final mesoscopic simulation model to more accurately reflect the physical field distribution of the real system, improving the local fidelity and overall accuracy of the simulation results. Attached Figure Description

[0076] Figure 1 This is a structural diagram of the method for digital reconstruction of three-dimensional models of historical buildings as described in this invention;

[0077] Figure 2 A flowchart deriving the structural change assumption;

[0078] Figure 3 A flowchart for cross-validating structural change assumptions with historical drawings;

[0079] Figure 4 Analysis diagram for iterative optimization of 3D models of historical buildings;

[0080] Figure 5 A bar chart showing the statistical evolution of the structure of historical buildings. Detailed Implementation

[0081] 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.

[0082] Please see Figure 1 This invention provides a method for digital reconstruction of 3D models of historical buildings. The method includes: identifying and analyzing historical traces on the surface of the target historical building, outputting the architectural structure evolution path. Based on this architectural structure evolution path, layered analysis and asynchronous alignment are performed on the currently collected point cloud data and texture image data to generate layered aligned architectural space data. The layered aligned architectural space data drives a parametric generation model, which gradually synthesizes a preliminary 3D voxel model of the target historical building through an iterative feedback mechanism. A rule base based on historical construction principles is introduced to perform logical rationality verification on the preliminary 3D voxel model, generating a 3D model with logical verification marks. Based on the logical verification marks, a new round of data acquisition requests is initiated for areas with contradictions or missing data in the 3D model, triggering a targeted supplementary data acquisition process. The new data from the targeted supplementary data acquisition process is integrated with the original data, and multiple rounds of iterative model optimization and refinement operations are performed until the preset integrity and consistency thresholds are met, outputting the optimized 3D digital model. Based on the 3D digital model, it is possible to simulate the evolution of architectural forms in different historical periods and generate corresponding virtual restoration schemes.

[0083] Example 1: See Figure 2The study identifies and analyzes historical traces on the surface of target historical buildings, acquiring spectral reflectance distribution maps of the building surface using high-resolution multispectral scanning equipment. Based on the differences in reflectance characteristics of different materials in different wavelength bands, areas suspected of human alteration, natural weathering, or damage are segmented from these maps and marked as initial trace areas. Each initial trace area undergoes a micro-scale 3D topographic scan, extracting microscopic undulation data and texture direction data of its surface geometry. This microscopic undulation data and texture direction data are then matched and similarity calculated against a pre-built database of historical construction method traces. Based on the matching and calculation results, each initial trace area is assigned one or more historical cause labels and corresponding confidence levels. Historical cause labels include changes in masonry methods, material repairs, and structural crack propagation. All tagged and confidence-rated trace areas are summarized to form a historical trace identification result, which records the spatial location, geometric features, and inferred causes of the traces.

[0084] After obtaining the historical trace identification results, structural change hypotheses of the target historical building in the temporal dimension are derived based on these results. The spatiotemporal distribution patterns and superposition order of various traces in the historical trace identification results are analyzed. Based on the superposition order of the traces, the relative time series of construction, modification, or damage of different parts of the building are inferred. Combined with historical cause labels, traces with the same or similar causes and spatiotemporal correlations are clustered to form several trace event sets. A descriptive event hypothesis model is constructed for each trace event set, including the event type, scope of influence, relative period of occurrence, and the resulting structural state changes. All event hypothesis models are sorted and logically connected according to the inferred relative time series to construct a hypothesis chain describing the entire process of the building's evolution from its original state through various events to its current state—that is, the structural change hypothesis set. The structural change hypothesis set is cross-validated and corrected using historical drawing scan data, outputting a validated architectural structural evolution path.

[0085] In practical implementation, taking a brick and stone building with a history of multiple renovations as an example, the historical traces on the surface of the target historical building are identified and analyzed to output the evolution path of the building structure. A high-resolution multispectral scanning device is used to acquire the spectral reflectance distribution map of the building surface. The device collects data in the visible and near-infrared bands, generating images recording the reflectance intensity at different wavelengths. In the spectral reflectance distribution map, areas suspected of being artificially modified, naturally weathered, or damaged are segmented based on the differences in reflectance characteristics of different materials in different bands, and marked as initial trace areas. For example, areas with significantly higher reflectance in the near-infrared band are initially identified as areas that have been repaired with different mortars.

[0086] In practice, each initial trace area undergoes a micro-scale 3D topographic scan to extract microscopic undulation data and texture direction data of its surface geometry. The point cloud density acquired by the scanning equipment exceeds 200 points per square millimeter. The microscopic undulation data and texture direction data are then matched and similarity calculated item-by-item with a pre-built historical construction method trace feature library. This library stores typical texture patterns and geometric features left by different techniques, such as "header-stretcher masonry" and "random rubble masonry." Based on the matching and calculation results, each initial trace area is assigned one or more historical cause labels and corresponding confidence levels. For example, an area might be simultaneously labeled "change in masonry method" and "material repair," with confidence levels of 0.85 and 0.15, respectively. Historical cause labels include changes in masonry method, material repair, and structural crack propagation. All tagged and confidence-based trace areas are then aggregated to form historical trace identification results. These results are recorded in a database table format, showing the spatial coordinates, geometric feature parameters, and inferred causes of the traces.

[0087] In practice, after obtaining the historical trace identification results, hypotheses about the structural changes of the target historical building in the time dimension are derived based on these results. The spatiotemporal distribution patterns and superposition order of various traces in the historical trace identification results are analyzed. For example, it is found that "brick layer weathering trace A" is covered by "cement plaster layer B". Based on the superposition order of the traces, the relative time sequence of construction, renovation, or damage of different parts of the building is inferred, suggesting that the construction of cement plaster layer B was later than the formation of brick layer weathering trace A. By combining historical cause labels, traces with the same or similar causes and spatiotemporal correlations are clustered into several trace event sets. For example, multiple traces on the upper part of the wall with high-confidence "change in masonry method" labels and consistent brick sizes are grouped into "event set C". For each set of trace events, a descriptive event hypothesis model is constructed. This model includes the event type, scope of impact, relative timing, and resulting structural changes. For example, the model for "Event Set C" is recorded as: "Type: Wall heightening; Scope of impact: North section of the east wall; Relative timing: Phase 2; Structural change: Three layers of brickwork added to the top of the original wall." All event hypothesis models are sorted and logically connected according to the inferred relative time sequence, constructing a chain of hypotheses describing the entire process of a building's evolution from its original state through various events to its current state—this is the structural change hypothesis set.

[0088] In some embodiments, the derivation of the set of structural change hypotheses may incorporate quantitative evaluation. It can be understood that a quantitative method for evaluating the reliability of trace superposition order relationships is based on geometric interferometric depth analysis of the trace boundary region. This method defines an order confidence function:

[0089]

[0090] Where: symbol This represents the confidence score for inferring a temporal relationship between two specific trace regions, with a value range from 0 to 1. (Symbol) This represents the average geometric depth at which one trace region spatially covers or encroaches upon another trace region in three-dimensional topographic data. (Symbol) This represents the average scale of the surface undulation features between two adjacent trace regions. (Symbol) It is an empirical scaling factor used to adjust the sensitivity of the depth-to-scale ratio to the confidence level. (Signature) It is an introduced positive correction factor used to ensure that the denominator term... Always greater than zero, to avoid situations where both trace areas have abnormally smooth surfaces, leading to... A division-by-zero error occurs when the value approaches zero.

[0091] In some embodiments, the set of structural change hypotheses is cross-validated and revised using historical drawing scan data. Vectorization and layer separation are performed on the acquired historical drawing scan data to extract building outlines, structural component lines, and annotation information. For example, the element representing "newly added buttresses" is extracted from a 1950s renovation sketch. The extracted historical drawing elements are coarsely matched in spatial location and scale with the point cloud projection outline of the current building to establish a coordinate system correspondence. Under this coordinate system correspondence, the structural state change corresponding to each event hypothesis model in the set of structural change hypotheses is compared with the architectural form or component information of different periods annotated in the historical drawings. For example, the hypothetical "Phase 2 wall heightening" event is overlaid with the annotation range of "1948 addition" in the drawings. When there is a conflict between the event hypothesis model and the historical drawing information, the credibility of the event hypothesis model is downgraded or a revision process is initiated based on the authority weight of the drawing information. For example, if the drawings clearly show that a certain window opening is in the original design, but the hypothesis model extrapolates it as a later addition, the credibility of that hypothesis is reduced. When the event hypothesis model is supported by historical drawing information, its credibility is enhanced, and the precise dimensions and shape information from the drawings are incorporated into the event hypothesis model. For example, the precise height and facade form of the addition recorded in the drawings are updated in the corresponding event model. After comparison and adjustment, a time-stamped architectural structure evolution path that integrates trace reasoning and drawing evidence is output. The architectural structure evolution path is ultimately recorded in the form of a series of three-dimensional blocks or keyframes with timestamps.

[0092] Example 2: See Figure 3The process of cross-validating and revising the set of structural change hypotheses by combining historical drawing scan data includes: performing vectorization and layer separation processing on the acquired historical drawing scan data to extract building outlines, structural component lines, and annotation information; performing coarse spatial and scale matching between the extracted historical drawing elements and the current building's point cloud projection outline to establish a coordinate system correspondence; comparing the structural state change corresponding to each event hypothesis model in the set of structural change hypotheses with the building form or component information of different periods marked in the historical drawings; when there is a conflict between the event hypothesis model and the historical drawing information, the credibility of the event hypothesis model is downgraded or a revision process is initiated based on the authority weight of the drawing information; when the event hypothesis model is supported by the historical drawing information, its credibility is increased, and the precise dimensions and shape information from the drawings are incorporated into the event hypothesis model. After comparison and adjustment, a time-marked architectural structural evolution path that integrates trace reasoning and drawing evidence is output.

[0093] Based on this architectural evolution path, the currently acquired point cloud data and texture image data are subjected to layered analysis and asynchronous alignment. According to the structural characteristics of different periods in the architectural evolution path, the point cloud data is decomposed into subsets corresponding to different construction periods or structural levels based on spatial location and geometric features. Each point cloud subset is independently reconstructed into a 3D surface, generating a series of triangular mesh fragments representing the structural surfaces of different periods or levels. Simultaneously, the texture image data is segmented into texture image groups roughly corresponding to each triangular mesh fragment based on its shooting perspective and content. A projection mapping relationship is established between each triangular mesh fragment and its corresponding texture image group. Through feature point matching and perspective transformation, the texture is accurately mapped to the corresponding mesh surface. In this process, partial overlap or gaps in space are allowed for triangular mesh fragments belonging to different periods or levels. Their relative positional relationships are constrained by the logic of the architectural evolution path, ultimately generating layered aligned architectural spatial data. This data contains multiple structural surfaces and their textures that are spatially and temporally separated but logically related.

[0094] In practice, the process of cross-validating and revising the set of structural change hypotheses by combining historical drawing scan data is illustrated by the facade evolution analysis of a brick-and-wood mixed structure building from the Republican era. The acquired historical drawing scan data undergoes vectorization and layer separation processing to extract the building outline, structural component lines, and annotation information. The drawing scan data includes a 1947 architectural survey drawing and a 1965 partial reconstruction approval drawing. The vectorization process transforms the line drawings and text annotations in the drawings into independent layer objects with geometric attributes. The extracted historical drawing elements are then coarsely matched in spatial position and scale with the current building's point cloud projection outline to establish a coordinate system correspondence. The coarse matching method utilizes relatively fixed feature points present in both the drawings and the point cloud, such as building corners and permanent foundation corners, to calculate a preliminary coordinate transformation matrix through similarity transformation.

[0095] In practice, under the coordinate system correspondence, the structural state change corresponding to each event hypothesis model in the structural change hypothesis set is compared with the architectural form or component information of different periods marked in historical drawings. The structural change hypothesis set includes three event models: "main structure built in the early 1930s", "west balcony added in 1952", and "enclosed balcony converted into a room in 1978". The drawing information includes the original balcony outline marked on the 1947 survey map and the "proposed conversion into a room" annotation area marked on the 1965 drawing. When there is a conflict between the event hypothesis model and the historical drawing information, the credibility of the event hypothesis model is downgraded or a correction process is initiated according to the authority weight of the drawing information. For example, if the hypothesis model speculates that the balcony was built in 1958, but the balcony is clearly drawn on the 1947 drawing, the confidence of the hypothesis model will be significantly reduced, and a re-evaluation process of the relevant trace area will be triggered. When the event hypothesis model is supported by historical drawing information, its credibility is enhanced. The precise dimensions and shape information from the drawings are incorporated into the event hypothesis model. For example, the boundary range of the "1978 balcony enclosure" event model closely matches the annotation area of ​​the 1965 drawings, thus increasing the confidence of the model. At the same time, the precise window opening dimensions and wall thickness obtained from the drawings are updated into the model parameters. After comparison and adjustment, a time-marked architectural structure evolution path that integrates trace reasoning and drawing evidence is output. The architectural structure evolution path is finally described as "Phase 1 (1930-1947): Initial construction of the main structure and the west balcony; Phase 2 (1965-1978): Balcony enclosure, window opening dimensions corrected to 1.2 meters wide and 1.5 meters high according to the 1965 drawings."

[0096] In some embodiments, the currently acquired point cloud data and texture image data are subjected to layered analysis and asynchronous alignment based on the building structure evolution path. According to the structural characteristics of different periods in the building structure evolution path, the point cloud data is decomposed into point cloud subsets corresponding to different construction periods or structural levels based on spatial location and geometric features. For the aforementioned building, the point cloud is segmented into "point cloud subset of the main wall and roof constructed in the early 1930s," "point cloud subset of the west balcony," and "point cloud subset of the sealing wall of the enclosed balcony in 1978." Each point cloud subset is independently reconstructed using 3D surface reconstruction to generate a series of triangular mesh fragments representing the surface of different periods or levels of structure. The point cloud subset representing the balcony railing is reconstructed as an independent triangular mesh fragment, spatially adjacent but separated from the mesh fragment representing the later enclosed wall.

[0097] In some embodiments, the texture image data is synchronously segmented into texture image groups that roughly correspond to each triangular mesh segment based on its shooting perspective and content. Photos taken from multiple angles of the west-facing balcony are grouped into the "balcony texture group," while photos of the inner and outer surfaces of the enclosed wall are grouped into the "enclosed wall texture group." A projection mapping relationship is established between each triangular mesh segment and its corresponding texture image group. Through feature point matching and perspective transformation, the texture is accurately mapped to the corresponding mesh surface. In specific implementations, the homography matrix is ​​calculated to complete the texture mapping by identifying the corner points on the triangular mesh segments of the balcony railing and the corresponding pixels in the photos of the "balcony texture group." In this process, it is permissible for triangular mesh segments belonging to different periods or levels to have partial overlap or gaps in space. The mesh representing the original balcony floor and the mesh representing the later added floor tiles have gaps in the vertical direction. Their relative positional relationship is constrained by the logic of the building structure evolution path, which indicates that the added layer is located on top of the original layer. The final result is layered aligned architectural space data, which contains multiple structural surfaces and their textures that are spatially and temporally separated but logically related. The data format can be a multi-layered data structure, with each layer associated with a period label, a triangular mesh, and a set of mapped texture coordinates.

[0098] It is understandable that evaluating the accuracy of coordinate alignment is a crucial step in the hierarchical parsing and asynchronous alignment process. Optionally, a quantitative method for evaluating the alignment accuracy between a subset of point clouds and historical drawing features can employ residual statistics. This method defines an average projection residual metric:

[0099]

[0100] Where: symbol This represents the average projection residual, used to measure the average positional deviation between the point cloud and the drawing. (Symbol) Indicates the number of common feature point pairs used for coordinate matching. (Symbol) This represents the first point extracted from the current building point cloud. A three-dimensional coordinate vector of a feature point. (Symbol) This indicates elements extracted from historical drawings, and... The corresponding number The two-dimensional coordinate vectors of a feature point on the drawing plane. (Symbol) This represents the projection transformation function from 3D point cloud coordinates to 2D drawing coordinates, calculated through coarse matching. (Symbol) This represents the calculation of the Euclidean norm of a two-dimensional vector, i.e., distance, with the symbol […]. It is an introduced positive correction factor used to ensure the divisor Always positive to prevent the number of feature point pairs from increasing in extreme cases. When the value is zero, division by zero occurs, resulting in invalid calculations.

[0101] Optionally, feature point matching during texture mapping can adopt an optimization strategy based on multi-view geometric constraints. When establishing projection mapping relationships, if a single texture image cannot completely cover a complex triangular mesh segment, the system selects multiple images with overlapping views from the texture image group corresponding to that segment. Through a multi-view stereo vision algorithm, the precise geometric relationships between these images and between the images and the mesh are calculated, thereby generating a seamless, high-resolution texture map that avoids seams and distortions. The specific implementation of the multi-view stereo vision algorithm involves the system automatically selecting multiple images with overlapping views from the texture image group corresponding to the triangular mesh segment, establishing the correspondence between these images through feature point detection and matching technology, and optimizing the calculation of the camera intrinsic and extrinsic parameters and relative spatial pose of each image based on the principle of multi-view geometric constraints. The algorithm further integrates the 3D geometric data of the images and the triangular mesh segment, accurately calibrates the projection mapping relationship, and achieves seamless fusion of texture information from multiple perspectives, ultimately generating a high-resolution, seamless, and distortion-free texture map, ensuring the realistic restoration and visual consistency of the surface details of historical buildings.

[0102] Example 3: A parametric generative model is driven by hierarchically aligned architectural space data. Each triangular mesh segment and its spatiotemporal label from the hierarchically aligned architectural space data are fed into the parametric generative model as an input sequence. The core of the parametric generative model is a voxel generation network with a spatiotemporal attention mechanism. The voxel generation network progressively parses the input sequence and, according to the architectural structure evolution path, generates voxel structures added or changed in subsequent periods within the three-dimensional voxel space, starting from the earliest basic form. In each generation step, the network outputs an intermediate voxel model for the current step and compares the intermediate voxel model with the corresponding hierarchically aligned architectural space data. The difference signal generated by the comparison is fed back to the network's attention mechanism to adjust the parameters of the next generation step. Thus, in the iteration, the generated voxel structure continuously approximates the real, hierarchical architectural space data, eventually converging to generate a preliminary three-dimensional voxel model.

[0103] A rule base based on historical construction principles is introduced to verify the logical rationality of the initial 3D voxel model. This rule base contains constraints related to the culture, region, and era of the target historical building, including structural, material mechanics, and aesthetic constraints. The initial 3D voxel model is converted into a symbolic structural relationship diagram suitable for rule-based reasoning. Nodes in the diagram represent building components, and edges represent connections, supports, or spatial relationships between components. The symbolic structural relationship diagram is then matched against the constraints in the rule base one by one. When a relationship violating the constraints is detected in the structural relationship diagram—for example, a cantilevered component lacking undersupport or an impossible overlapping relationship between components from different periods—a logical verification mark is added to the violating component or relationship. Marking types include logical conflict, suspected missing elements, and inconsistency in era. After traversing the entire structural relationship diagram, a 3D model with logical verification marks is output. This model visually or numerically identifies all areas with logical problems.

[0104] In practical implementation, the process of using layered aligned architectural spatial data to drive a parametric generative model takes an ancient pagoda that underwent two major additions during the Qing Dynasty and the Republic of China as an example. Each triangular mesh segment and its spatiotemporal label in the layered aligned architectural spatial data are fed into the parametric generative model as an input sequence. The input sequence can be represented as [M_Qing base, T_Qing base], [M_Qing body, T_Qing body], [M_Republic top, T_Republic top], where M represents the triangular mesh segment and T represents the corresponding period label. The core of the parametric generative model is a voxel generation network with a spatiotemporal attention mechanism. The voxel generation network gradually parses the input sequence and, according to the architectural structure evolution path, sequentially generates voxel structures added or changed in subsequent periods in the three-dimensional voxel space, starting from the earliest basic form. The network first generates voxel blocks representing the Qing Dynasty base based on [M_Qing base, T_Qing base]. Then, combining the logic of "a bell tower was added to the Qing Dynasty pagoda during the Republic of China period" in the architectural structure evolution path, it incrementally generates voxel structures representing the Republic of China bell tower on the already generated basic voxels.

[0105] In practice, during each generation step, the network outputs an intermediate voxel model for the current step and compares it with the corresponding layered architectural space data. When the network generates an intermediate voxel model representing the Qing Dynasty pagoda body, the system calculates the spatial position and shape differences between the surface of this voxel model and the [M_Qing body] triangular mesh segment. The resulting difference signal is fed back to the network's attention mechanism to adjust the parameters for the next generation step. The difference signal indicates a deviation between the generated pagoda body tapering curve and the real triangular mesh segment. The attention mechanism adjusts the network parameters to focus more on the geometric features of the pagoda body outline in the next generation. Thus, through iteration, the generated voxel structure continuously approximates the real, layered architectural space data, eventually converging to generate a preliminary 3D voxel model. This preliminary 3D voxel model is a complete 3D mesh containing voxel labels from different periods, where each voxel carries its inferred construction period attribute.

[0106] In some embodiments, a rule base based on historical construction principles is introduced to verify the logical rationality of the preliminary 3D voxel model. The rule base contains constraints related to the culture, region, and era of the target historical building, including structural logic, material mechanics logic, and morphological aesthetics logic. For the ancient pagoda, the rule base includes specific entries such as "the tapering ratio of the body of a Qing Dynasty brick pagoda is usually between 10% and 15%" and "wooden brackets should not appear below a brick dome." The preliminary 3D voxel model is converted into a symbolic structural relationship diagram suitable for rule-based reasoning. Nodes in the diagram represent building components, and edges represent connections, supports, or spatial relationships between components. For example, there is an edge "supported on" between the "tower body-foundation" node, and an edge "located on" between the "bell tower-tower body" node. The conversion process is as follows: First, the geometric core information of each voxel cluster in the preliminary 3D voxel model is extracted, including the spatial coordinate range, overall size, morphological outline, and spatiotemporal label of the voxel cluster, while simultaneously associating the already labeled building component identifiers in the model. By combining a pre-built historical construction method trace feature library with the construction logic in the rule base, each voxel cluster is semantically categorized. Based on its geometric features and spatiotemporal attributes, corresponding component type semantic tags are matched to clarify the building component category corresponding to each voxel cluster. By analyzing the spatial positional relationships and geometric overlap states of each voxel cluster, and combining this with the material mechanics logic in the rule base, the relationships between components are determined. For example, based on the vertical distribution relationship and force transmission logic of voxel clusters, a "supported by" relationship is identified; based on the adjacent splicing state of voxel clusters, a "connected by" system is identified. Using semantically tagged building components as nodes in the graph, and the determined relationships between components as edges, a symbolic structural relationship graph is constructed. Nodes are synchronously associated with component identifiers, spatiotemporal tags, and other attributes, while edges clearly label the relationship type to ensure compatibility with subsequent rule reasoning and verification.

[0107] In some embodiments, the symbolic structural relationship diagram is matched one by one with the constraints in the rule base. The system traverses each rule in the rule base and searches for structural relationship combinations that violate the rule in the symbolic structural relationship diagram. When a relationship that violates the constraint is detected in the structural relationship diagram, such as a cantilevered component lacking lower support or an impossible overlapping relationship between components from different periods, a logical verification mark is added to the component or relationship that violates the rule. The mark types include logical conflict, suspected missing, and inconsistency in era. For example, in the preliminary three-dimensional voxel model, a brick block marked "Qing Dynasty" and a reinforced concrete block marked "Republic of China" are displayed as "closely embedded" in the structural relationship diagram. This triggers a rule that "materials and processes from different periods have discontinuous combinations", thus adding a "inconsistency in era" logical verification mark to these two blocks. After traversing the entire structural relationship diagram, a 3D model with logical verification marks is output. The 3D model with logical verification marks marks all areas with logical problems in a visual or data form. In the 3D visualization interface, voxel blocks with logical verification marks are highlighted and the specific violation rule number is recorded in the attribute table.

[0108] It is understandable that the computational process of the spatiotemporal attention mechanism can be formally expressed. In each generation step of the voxel generation network, the calculation of attention weights needs to consider the continuity constraint of historical periods. Optionally, one method for calculating weights in the spatiotemporal attention mechanism is as follows:

[0109]

[0110] Where: symbol Indicates the first In each generation step, network attention is assigned to historical memory units. Weights. Symbols This represents the extraction of geometric features and spatiotemporal labels from the current triangular mesh segment, which are then transformed into a fixed-dimensional input encoded vector via a feature encoding network. (Symbol) Indicates the number of internal network maintenance Each historical memory unit stores the features of previously generated voxel structures. (Function) This indicates the calculation of the cosine similarity between two vectors. (Symbol) This is a temperature parameter used to control the sharpness of the weight distribution. (Symbol) This represents the total number of historical memory units. (Function) It is a discriminant function, which determines the target period label of the current step. With historical memory unit The associated period tag When the function conforms to the temporal relationship defined in the building structure evolution path, the function value is 1; otherwise, it is a minimum value close to 0.

[0111] Optionally, the rule base matching check can be based on a graph pattern matching algorithm. When matching the symbolic structure graph with the constraints in the rule base, the constraints of each historical construction rule are pre-compiled into one or more "problem graph patterns" to be detected. During graph traversal, the system continuously attempts to find subgraph structures isomorphic to these "problem graph patterns" in the vast symbolic structure graph. Once a match is found, the system immediately triggers the addition of logical verification markers on the corresponding nodes and edges.

[0112] See Figure 4 This is an iterative optimization analysis chart of a 3D model of a historical building, showcasing the changes in two key indicators during the model iteration process. The two indicators show a negative correlation: a decrease in logical markers corresponds to an increase in completeness, consistent with the logic of "correcting contradictions and supplementing data during iterative optimization"; the fifth round is a critical node: after this round, completeness is achieved, the number of logical markers drops below 5, and subsequent iterations only involve fine-tuning. This type of chart is used for process monitoring in the 3D reconstruction of historical buildings, visually displaying the model's optimization trajectory from "initial generation" to "convergence and achievement of standards," helping to determine the timing of iteration termination and assess reconstruction efficiency.

[0113] Example 4: A new round of data collection requests is initiated for areas in the 3D model with inconsistencies or missing information. The system automatically analyzes the type and spatial distribution of logical verification markers. First, the system reads the attribute information of all logical verification markers, including marker type, the building component identifier to which the marker is attached, and the marker's coordinates in 3D space. The logical verification markers are classified according to their type, with categories including logical conflict, suspected missing information, and inconsistency in era. For each category, the system calculates the spatial density distribution of that type of marker on the surface of the 3D model and generates a density heatmap using a kernel density estimation algorithm. The system identifies high-density areas in the density heatmap and demarcates the boundaries between high-density and low-density areas. Combining the marker type and spatial density distribution, the system generates an analysis report indicating the degree of clustering and spatial distribution patterns of various marker types, providing a priority basis for subsequent data collection requests.

[0114] For areas marked as logically conflicting or suspected missing, the system generates targeted data acquisition task instructions. These instructions specify the types of data to be collected, the key areas to focus on, the recommended acquisition equipment, and the acquisition perspective. The data acquisition task instructions are encapsulated as data acquisition requests and sent to the connected data acquisition scheduling terminal or platform. This triggers a targeted supplementary data acquisition process, which, depending on the request, may include laser scanning of concealed building areas at specific angles, radar detection of suspected internal voids or material boundaries, or higher-resolution photogrammetry of specific structural nodes.

[0115] In practice, a new round of data collection requests is initiated for areas in the 3D model that contain contradictions or omissions. The system automatically analyzes the type and spatial distribution of logical verification markers. The specific scenario is set as a wooden palace that has been repeatedly repaired during the Ming and Qing dynasties, and a 3D model with logical verification markers has already been generated. The system first reads the attribute information of all logical verification markers. The attribute information of logical verification markers includes the marker type, the architectural component identifier to which the marker is attached, and the coordinates of the marker in 3D space. For example, an attribute record is "Marker ID: Conflict_0231, Type: Logical Conflict, Component ID: Dougong_45, Coordinates (x,y,z):(12.3,45.6,8.7)".

[0116] In implementation, logical check tags are categorized according to their type. These categories include logical conflicts, suspected missing tags, and outdated tags. The system groups and statistically analyzes all tags according to these three predefined categories. For each category, the system calculates the spatial density distribution of the tags on the 3D model surface, generating a density heatmap using a kernel density estimation algorithm. This algorithm uses the spatial coordinates of each logical check tag as input points, calculates the number of tags per unit area at any location on the entire model surface using a Gaussian kernel function, and renders the heatmap with different colors. The system identifies high-density areas in the density heatmap and demarcates them from low-density areas. A density threshold is set, and continuous areas with kernel density values ​​higher than this threshold are labeled "High-Density Area A," while areas lower are labeled "Low-Density Area B." Combining the tag type and spatial density distribution, the system generates an analysis report. This report indicates the degree of clustering and spatial distribution patterns of various tags, providing a priority basis for subsequent data collection requests. The report includes a description such as "Logical conflict tags show high-density clustering in the northwest corner eaves area of ​​the building."

[0117] In some embodiments, the analysis report can be further quantified into specific data tables. A quantification method for comprehensively evaluating and guiding data collection prioritization is based on weighted calculations of spatial density and label type. This method defines a regional priority weight for ranking:

[0118]

[0119] Where: symbol This represents the overall priority weight score for a specific high-density area; a higher score indicates a higher priority for that area in data collection. (Symbol) This represents the normalized average kernel density value for the region, which ranges from 0 to 1. (Symbol) This represents the average severity score of all logical check tags within the region. The score is assigned by the rule base when generating the tags based on the severity of rule violations, and is also normalized to the range of 0 to 1. (Symbol) This is a weighting coefficient related to the tag type; for example, "logical conflict" is assigned a coefficient of 1.0, "suspected missing" is assigned 0.7, and "out of date" is assigned 0.5. (Symbol) This is a balancing factor used to adjust the relative importance of spatial density and problem severity in priority assessment; its value ranges between 0 and 1. Using this formula, the system can generate a quantified list of regional priorities, as shown in Table 1.

[0120] Table 1: Results of Spatial Clustering Analysis of Logical Check Markers

[0121]

[0122] It is understandable that the system generates targeted data acquisition task instructions based on the conclusions of the analysis report. For areas marked as having logical conflicts or suspected missing data, the system generates targeted data acquisition task instructions. These instructions specify in detail the data types that need to be collected, the key areas to focus on, the recommended acquisition equipment, and the acquisition perspective. For example, for area R01, the instructions would be: "Data type: Point cloud and infrared thermal imaging data of the wooden frame inside the roof; Location: The junction of the rafters, purlins, and brackets in the third bay of the northwest corner; Equipment: Handheld laser scanner, infrared thermal imager; Perspective: Shooting from inside the attic upwards, and from the outside of the roof at an angle." The data acquisition task instructions are encapsulated into data acquisition requests and sent to the connected data acquisition scheduling terminal or platform. The data acquisition requests use a structured JSON or XML format, including the area coordinate boundaries, task instructions, and expected data accuracy.

[0123] Optionally, the generation of data acquisition task instructions can incorporate semantic information from the Building Information Modeling (BIM) component library. When the building component identifier attached to the logical verification mark can be associated with a predefined BIM component type, the data acquisition task instructions can be further refined. For example, for a "socket" component marked as "suspected missing," the system will retrieve the standard form, common dimensions, and connection methods of the "socket" from the BIM component library, and specifically indicate in the task instructions that the integrity of the socket spout and the overlapping details with the arch should be checked. This triggers a targeted supplementary data acquisition process, which, based on the request content, includes laser scanning of concealed parts of the building at specific angles, radar detection of suspected internal hollow areas or boundaries between different materials, or higher-resolution photogrammetry of specific structural nodes. For example, to verify a wall plaster layer that is "out of date," the process will dispatch ground-penetrating radar equipment to perform a planar scan of the wall to detect the internal brick masonry structure.

[0124] Optionally, the data acquisition scheduling platform can optimize and merge tasks after receiving requests. Upon receiving multiple independent data acquisition requests, the platform analyzes the spatial proximity of the requested targets and the overlap of required equipment. Requests that are spatially close and require the same equipment are merged into a single comprehensive acquisition work order to improve the efficiency of field data acquisition. For example, photogrammetry requests targeting three "suspected missing" areas on different floors of the same facade can be merged into a single, continuous shooting task covering the entire facade from multiple angles.

[0125] Example 5: Integrating new data from the targeted supplementary data acquisition process with existing data, performing multi-round iterative model optimization and refinement operations. Upon receiving new data from the targeted supplementary data acquisition process, this new data is first fused with the previously corresponding layered aligned architectural space data to update the layered aligned architectural space data. Using the updated layered aligned architectural space data, the steps of driving parametric model generation and performing logical rationality checks are repeated to generate a new version of the 3D digital model. The number and severity of residual logical check markers on the new version model are checked. If the logical check markers are not reduced to below a preset threshold, or new contradictions are discovered, a new round of data acquisition requests is generated based on the new version model and residual markers, initiating the next iteration. If the logical check markers meet the preset integrity and consistency thresholds, the model is considered converged, and the current version of the 3D digital model is used as the final output. During the iteration process, each model update and re-evaluation of historical traces fine-tunes the earlier architectural structure evolution path, forming a closed-loop optimization from data to inference.

[0126] In practice, new data from the targeted supplementary data acquisition process is integrated with existing data. Multiple rounds of iterative model optimization and refinement are performed. The scenario continues with a 3D model of Ming and Qing dynasty wooden palace buildings using proprietary logical verification markers. The model triggers its first targeted supplementary data acquisition in the northwest corner eaves area due to dense logical conflict markers. Upon receiving new data from the targeted supplementary data acquisition process, this new data is first fused with the previously corresponding layered aligned architectural space data to update the layered aligned architectural space data. For example, high-precision point clouds of the concealed wooden structure inside the roof are acquired through supplementary acquisition. This point cloud data is registered and stitched with the original triangular mesh segment "M_Roof_Clear" representing the exterior of the roof, generating an updated "M_Roof_Clear_v2" triangular mesh segment containing both inner and outer surfaces. This updated segment is then integrated into the layered aligned architectural space dataset to replace the old version.

[0127] In practice, the updated, layered, aligned architectural space data is used to repeatedly drive the parametric model generation process and perform logical rationality checks, generating a new version of the 3D digital model. The system takes the updated data sequence containing "M_Roof_Clear_v2" as input, re-runs the parametric model generation process, and generates a preliminary 3D voxel model with corrected internal structure. Then, based on a rule base of historical construction principles, the newly generated model undergoes another logical rationality check. The number and severity of remaining logical check markers on the new version model are examined. The system counts the total number of all logical check markers on the new version of the 3D digital model and calculates its average severity score, comparing it with the results of the previous iteration. If the number of logical check markers fails to decrease below a preset threshold, or if new contradictions are discovered, a new round of data collection requests is generated based on the new version model and remaining markers, initiating the next iteration. For example, after resolving the eaves construction conflict, if new "suspected missing" markers are found in the column base area of ​​the new model, the system will generate data collection requests for the column base and foundation areas accordingly.

[0128] In practice, if the logical verification markers meet the preset integrity and consistency thresholds, the model is considered converged, and the current version of the 3D digital model is taken as the final output. The preset thresholds can be set to a total number of logical verification markers not exceeding 10 and no markers with a severity level of "fatal." When the model meets this condition, the iteration loop terminates. During the iteration process, each model update and reassessment of historical traces fine-tunes the earlier architectural structure evolution path, forming a closed-loop optimization from data to inference. For example, column base data obtained in subsequent iterations reveals an unrecorded early renovation. This information will be fed back to the historical trace identification and analysis step, correcting the division of construction periods in the architectural structure evolution path.

[0129] In some embodiments, iterative convergence can be determined based on a comprehensive evaluation function. It can be understood that a function for quantifying the optimization effect of a single iteration considers the reduction of logic check marks and the improvement in model geometric consistency. This iterative effect evaluation function is expressed as follows:

[0130]

[0131] Where: symbol This represents the optimization performance score for a single iteration; a higher score indicates a more significant improvement in that iteration. (Symbol) This indicates the total number of logical checkpoints remaining on the new version of the 3D digital model after this iteration. (Symbol) This represents the total number of logic check tags on the model after the previous iteration. (Ratio) It measures the relative change in the number of markers. (Symbol) This represents the improvement in overall geometric consistency between the 3D digital model and the hierarchically aligned architectural space data before and after this iteration. This value can be quantified by calculating the reduction in the average distance between the model surface and the corresponding point cloud. (Symbol) It is a weighting factor between 0 and 1, used to balance the proportion of reduced marker count versus improved geometric consistency in the effectiveness evaluation. (Symbol) It is an introduced positive correction factor used to ensure that the denominator Always greater than zero, to avoid problems in the early stages of iteration. A division-by-zero error may occur when the result is zero, ensuring the continuity of function calculation. When Below a certain set threshold and When the preset integrity and consistency thresholds have been reached, the optimization can be considered to have converged.

[0132] In some embodiments, the reassessment of historical traces and the fine-tuning of the building structure evolution path are automated processes. When newly acquired data shows irreconcilable contradictions with the current assumptions of the building structure evolution path, the system creates a path correction proposal. This proposal is associated with the affected trace areas and event hypothesis models and submitted to a collaborative verification process consisting of a historical trace identification and parsing module and a path reasoning module for automatic evaluation and adjustment, ultimately outputting an updated version of the building structure evolution path. Optionally, simple reinforcement learning logic can be introduced into the decision-making process in closed-loop optimization. When deciding whether to initiate a new round of data acquisition requests, in addition to checking logical verification markers, the system can also refer to the "benefit-cost ratio" of historical iterations. For example, if the supplementary acquisition for a certain type of marker in previous rounds has failed to significantly reduce the number of markers, the system will reduce the priority of acquiring such areas in subsequent iterations or allocate resources to other types of contradictory areas. Optionally, model version management adopts a tree structure. Each iteration generates a new version of the 3D digital model, along with its corresponding layered aligned architectural space data and architectural structure evolution path, which are all saved as a branch node. The parent version identifier and the reason for triggering the iteration are also recorded. This management method allows for backtracking to any historical version for analysis during the optimization process, or starting a new optimization branch from a specific node when new research clues emerge.

[0133] See Figure 5This is a bar chart showing the statistical distribution of events related to the structural evolution of a historical building, illustrating the distribution of such events across different historical periods. The Qing Dynasty saw the highest number of events: construction events (12), structural changes (7), and material replacements (6) were all the highest, indicating the most frequent renovations to the building during this period. The Yuan Dynasty had the fewest events: the number of constructions, structural changes, and material replacements were all low, suggesting a relatively stable building condition during this period. From the Song Dynasty to the Qing Dynasty, the number of events generally increased, while in modern times it has fallen back to a lower level, reflecting the fluctuating patterns of architectural activities throughout history. This type of chart is a core tool for analyzing the evolutionary path of historical building structures. By quantifying renovation and maintenance activities in different periods, it helps to deduce the spatiotemporal logic of building changes, providing a basis for the layered analysis of 3D model reconstruction.

[0134] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0135] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A historical building three-dimensional model digital reconstruction method, characterized in that, The method includes: Identify and analyze historical traces on the surface of the target historical building, and output the evolution path of the building structure; Based on the building structure evolution path, the currently collected point cloud data and texture image data are subjected to layered parsing and asynchronous alignment to generate layered aligned building space data; The hierarchical aligned architectural space data is used to drive a parametric generation model, which gradually synthesizes a preliminary three-dimensional voxel model of the target historical building through an iterative feedback mechanism. A rule base based on historical construction principles is introduced to perform logical rationality verification on the preliminary three-dimensional voxel model, generating a three-dimensional model with logical verification marks. Based on the logical verification mark, a new round of data collection requests is initiated for areas in the 3D model that have contradictions or missing data, triggering a targeted supplementary data collection process; Integrate new data from the targeted supplementary data acquisition process with existing data, perform multiple rounds of iterative model optimization and refinement operations until the preset integrity and consistency thresholds are met, and output the optimized three-dimensional digital model; Based on the aforementioned three-dimensional digital model, the evolution of architectural forms in different historical periods is simulated, and corresponding virtual restoration plans are generated. The identification and analysis of historical traces on the surface of the target historical building, and the output of the building structure evolution path, include: Obtain the historical trace identification results, and based on the historical trace identification results, deduce the structural change hypothesis of the target historical building in the time dimension, and generate a set of structural change hypotheses; By combining historical drawing scan data, the set of structural change assumptions is cross-validated and corrected, and a validated building structure evolution path is output. The identification and analysis of historical traces on the surface of the target historical building includes: The spectral reflectance distribution map of the building surface is obtained by using a high-resolution multispectral scanning device; In the spectral reflectance characteristic distribution map, based on the differences in reflectance characteristics of different materials in different bands, areas suspected of being artificially modified, naturally weathered, or damaged are segmented and marked as initial trace areas; For each of the initial trace regions, perform a micro-scale three-dimensional topography scan to extract the micro-undulation data and texture direction data of its surface geometry; The micro-undulation data and texture direction data are matched and similarity calculated item by item with a pre-built historical construction method trace feature library. Based on the matching and calculation results, one or more possible historical cause labels and corresponding confidence levels are assigned to each initial trace area. The historical cause labels include changes in masonry methods, material repairs, and structural crack propagation. All tagged and confidence-rated trace regions are aggregated to form the historical trace identification results, which record the spatial location, geometric features, and inferred causes of the traces.

2. The historical building three-dimensional model digital reconstruction method according to claim 1, characterized in that, The assumptions derived from the structural changes of the target historical building over time include: The spatiotemporal distribution patterns and superposition order of various types of traces in the historical trace identification results are analyzed. Based on the superposition order of the traces, the relative time sequence of the construction, renovation or damage of different parts of the building can be inferred; By combining the historical causal tags, traces with the same or similar causes and spatiotemporal correlations are clustered to form several sets of trace events. For each set of trace events, a descriptive event hypothesis model is constructed, which includes the event type, scope of impact, relative timing of occurrence, and possible structural state changes. All event hypothesis models are sorted and logically connected according to the inferred relative time series to construct a chain of hypotheses describing the entire process of a building's evolution from its original state through various events to its current state, which is the set of structural change hypotheses.

3. The method for digital reconstruction of three-dimensional models of historical buildings according to claim 2, characterized in that, The cross-validation and correction of the set of structural change hypotheses by combining historical drawing scan data includes: The acquired historical drawing scan data is vectorized and layered to extract building outlines, structural component lines, and annotation information. The extracted historical drawing elements are roughly matched with the point cloud projection outline of the current building in terms of spatial location and scale to establish a coordinate system correspondence. Under the coordinate system correspondence, the structural state change corresponding to each event hypothesis model in the set of structural change hypotheses is compared with the building form or component information of different periods marked in the historical drawings; When there is a conflict between the event hypothesis model and historical drawing information, the credibility of the event hypothesis model will be downgraded or a correction process will be initiated based on the authority weight of the drawing information. When the event hypothesis model is supported by historical drawing information, its credibility is increased, and the precise dimensions and shape information in the drawings are incorporated into the event hypothesis model. After comparison and adjustment, a time-stamped evolution path of the building structure is output, which integrates trace reasoning and drawing evidence.

4. The historical building three-dimensional model digital reconstruction method according to claim 3, characterized in that, The process of performing layered parsing and asynchronous alignment of the currently acquired point cloud data and texture image data includes: Based on the structural characteristics of different periods in the building structure evolution path, the point cloud data is decomposed into point cloud subsets corresponding to different construction periods or structural levels according to spatial location and geometric features. Each subset of point clouds is independently reconstructed into a three-dimensional surface to generate a series of triangular mesh fragments representing surfaces of different periods or hierarchical structures. Simultaneously, the texture image data is divided into texture image groups that roughly correspond to each triangular mesh segment based on its shooting angle and content; Establish the projection mapping relationship between each triangular mesh fragment and its corresponding texture image group, and accurately map the texture to the corresponding mesh surface through feature point matching and perspective transformation; In this process, triangular mesh fragments belonging to different periods or levels are allowed to partially overlap or have gaps in space. Their relative positional relationship is constrained by the logic of the architectural structure evolution path, and finally the layered aligned architectural space data is generated. The architectural space data contains multiple structural surfaces and their textures that are separated in time and space but logically related.

5. The historical building three-dimensional model digital reconstruction method according to claim 4, characterized in that, The parametric generation model is driven by the hierarchically aligned architectural space data, including: Each triangular mesh segment and its spatiotemporal label in the hierarchically aligned architectural spatial data are fed into the parameterized generation model as an input sequence; The core of the parametric generation model is a voxel generation network with a spatiotemporal attention mechanism. The voxel generation network parses the input sequence step by step and, according to the building structure evolution path, generates voxel structures that are added or changed in subsequent periods in the three-dimensional voxel space, starting from the basic form in the earliest period. In each generation step, the network outputs an intermediate voxel model for the current step and compares the intermediate voxel model with the corresponding hierarchically aligned architectural space data. The difference signals generated by the comparison are fed back to the network's attention mechanism to adjust the parameters of the next generation step, so that the generated voxel structure continuously approximates the real, layered architectural space data during the iteration, and finally converges to generate the preliminary three-dimensional voxel model.

6. The historical building three-dimensional model digital reconstruction method according to claim 5, characterized in that, The introduction of a rule base based on historical construction principles to perform logical rationality verification on the preliminary 3D voxel model includes: The rule base based on historical construction principles includes constraints related to the construction logic, material mechanics logic, and morphological aesthetic logic of the target historical building, which are related to its culture, region, and era. The preliminary three-dimensional voxel model is converted into a symbolic structural relationship diagram suitable for rule-based reasoning. In the diagram, nodes represent building components, and edges represent connections, supports, or spatial relationships between components. The symbolic structural relationship diagram is matched one by one with the constraints in the rule base; When a relationship that violates the constraints is detected in the structural relationship diagram, the logical verification mark is added to the component or relationship that violates the rules. The relationship that violates the constraints includes a cantilever component lacking a lower support and an impossible overlapping relationship between components from different periods. The types of logical verification marks include logical conflict, suspected missing, and inconsistency in time. After traversing the entire structural relationship diagram, the three-dimensional model with logical verification marks is output. The three-dimensional model marks all areas where logical problems may exist in a visual or data form.

7. The historical building three-dimensional model digital reconstruction method according to claim 6, characterized in that, The step of initiating a new round of data collection requests for areas in the 3D model that contain contradictions or missing data includes: The system automatically analyzes the type and spatial distribution of the logical check tags; For areas marked as logically conflicting or suspected missing, the system generates targeted data acquisition task instructions. These instructions specify in detail the types of data that need to be collected, the key areas to focus on, the recommended acquisition equipment, and the acquisition perspective. The data acquisition task instruction is encapsulated into the data acquisition request and sent to the connected data acquisition scheduling terminal or platform; The targeted supplementary data acquisition process is triggered. The data acquisition process, based on the request content, includes laser scanning of concealed parts of the building at specific angles, radar detection of suspected internal hollow areas or boundaries between different materials, or higher-resolution photogrammetry of specific structural nodes.

8. The historical building three-dimensional model digital reconstruction method according to claim 7, characterized in that, The system automatically analyzes the type and spatial distribution of the logical check tags, including: The system first reads the attribute information of all logical verification tags, including the tag type, the building component identifier to which the tag is attached, and the coordinates of the tag in three-dimensional space; Logical check tags are categorized according to their tag type. The categories include logical conflict, suspected missing tags, and outdated tags. For each classification category, the system calculates the spatial density distribution of the label of that category on the surface of the 3D model and generates a density heatmap using a kernel density estimation algorithm; The system identifies high-density regions in the density heatmap and delineates the boundaries between high-density and low-density regions. By combining the marker type and spatial density distribution, the system generates an analysis report, which indicates the degree of aggregation and spatial distribution pattern of various markers, providing a priority basis for subsequent data collection requests.

9. The historical building three-dimensional model digital reconstruction method according to claim 8, characterized in that, The multi-round iterative model optimization and refinement operations include: After receiving new data from the targeted supplementary data acquisition process, the new data is first merged with the previously corresponding layered aligned building space data to update the layered aligned building space data. Using the updated hierarchically aligned architectural space data, repeat the steps of driving parametric model generation and performing logical rationality verification to generate a new version of the 3D digital model. Check the number and severity of logic check tags left on the new version of the model; If the logical verification flags fail to decrease to below the preset threshold, or if new contradictions are discovered, a new round of data collection requests will be generated based on the new version model and the legacy flags, and the next iteration will be initiated. If the logical verification flags have met the preset integrity and consistency thresholds, the model is determined to have converged, and the current version of the 3D digital model is used as the final output. During the iteration process, each model update and reassessment of historical traces may fine-tune the earlier architectural evolution path, forming a closed-loop optimization from data to reasoning.

Images