Three-dimensional geographic scene construction method and system

Abstract

The invention provides a three-dimensional geographic scene construction method and system, and the method comprises the steps: screening building geometric data, texture data and vegetation distribution data corresponding to a current visual angle, obtaining a high-similarity sight line path through a historical browsing path of a user, and carrying out the correlation analysis of the high-similarity sight line path and a preliminarily correlated building; the method comprises the following steps of: fusing and optimizing data matching precision through texture analysis details; carrying out grid segmentation on a sight line path and associating corresponding scene data to generate a pre-cache data block; grouping building and vegetation data to obtain a complete loading sequence; distributing cache resources in combination with a memory state through a dynamic cache mechanism, and updating a scene element index in real time. The method comprises the following steps of: constructing a three-dimensional geographic scene construction framework on the basis of a traditional three-dimensional geographic scene system, dynamically updating a complete texture analysis path, verifying a complete loading sequence through potential nodes, and realizing low-delay loading, high-fidelity rendering and multi-view switching of a three-dimensional geographic scene. And the high efficiency and interactivity in the three-dimensional geographic scene construction process are improved.

Description

Technical Field

[0001] This invention relates to the field of 3D scene construction technology, and more specifically, to a method and system for constructing 3D geographic scenes. Background Technology

[0002] As digital cities, smart navigation, virtual reality and other fields develop towards refinement and real-time, how to achieve low-latency loading, high-fidelity rendering and multi-view switching under dense spatial data has become a core technical challenge to ensure the application value of 3D geographic scenes.

[0003] Traditional methods for constructing 3D geographic scenes have significant limitations. For example, traditional 3D geographic scene systems often use a full data loading mode without dynamically filtering data based on the user's real-time perspective. When the scene coverage expands to the city level, loading takes a long time and has a high memory consumption, which can easily lead to problems such as interface lag and rendering crashes. At the same time, traditional 3D geographic scene systems lack coordinated optimization of the loading sequence of building geometry data, road model data, and vegetation texture data. This often results in problems such as building geometry being displayed but textures not being loaded, and spatial misalignment between road models and surrounding vegetation, leading to scene distortion and failing to meet the high-precision requirements of scenarios such as urban governance and emergency rescue.

[0004] Therefore, there is an urgent need to develop a perspective-driven dynamic prediction and multi-dimensional data collaborative optimization method. Based on the traditional 3D geographic scene system, a 3D geographic scene construction framework integrating hierarchical indexing, path prediction, temporal control, and cache scheduling should be built to realize on-demand filtering, accurate preloading, and temporal collaboration of scene data, so as to achieve an efficient and highly interactive 3D geographic scene construction process. Summary of the Invention

[0005] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for constructing a three-dimensional geographic scene, the method comprising: Obtain geometric and texture data of buildings from the user's perspective to generate a preliminary list of associated buildings; Historical data is acquired, and similarity is calculated based on a preliminary list of associated buildings to obtain a high-similarity line-of-sight path prediction sequence. Correlation analysis is performed on highly similar line-of-sight path prediction sequences to obtain extended loading sequences, and a set of pre-cached data blocks is obtained based on the extended loading sequences; Optimize the pre-cached data block set to obtain an optimized time-series mismatch correction structure; The view rotation event is processed based on the temporal mismatch correction structure to obtain the complete texture parsing path. Potential nodes are extracted from the complete texture parsing path to obtain the complete loading sequence. Memory is allocated for the complete loading sequence, real-time scene feature index is obtained, and a real-time 3D geographic scene is constructed based on the real-time scene feature index.

[0006] As a further aspect of the present invention, the geometric and texture data of buildings from the user's perspective are obtained, and a preliminary list of associated buildings is obtained, including: Construct a multi-level index hierarchy structure by hierarchical levels of geographic regions, building types, and data precision. The geographic region level is used to obtain the region to which the current view belongs, the building type level is used to obtain the target type of building, and the corresponding precision of the building's geometric and texture data is extracted through the data precision level. Obtain vegetation distribution data for the area to which the current viewpoint belongs, including vegetation type, vegetation density, and vegetation spatial coordinates; The building geometric data, texture data and vegetation distribution data are spatially correlated and integrated to obtain building elements, vegetation elements and spatial relationships and form an initial set of scene elements. Perform preliminary matching on the buildings in the initial set of scene elements to obtain a preliminary list of matched and associated buildings.

[0007] As a further aspect of the present invention, historical data is acquired, and similarity calculation is performed based on a preliminary list of associated buildings to obtain a highly similar line-of-sight path prediction sequence, including: The system acquires historical data consisting of multiple historical browsing paths, the dwell time for each path, the frequency of perspective switching, and the sequence of buildings of interest. Historical feature vectors are composed of path orientation angle, average viewing height, building attention weight, and path length extracted from historical data. Obtain the feature vector of the current viewpoint, and obtain the similarity score between the feature vector of the current viewpoint and the historical feature vector based on the k-nearest neighbor algorithm; Set a similarity score threshold, and filter the line-of-sight paths corresponding to historical feature vectors with similarity scores greater than the similarity score threshold as high similarity line-of-sight paths; The similarity scores of highly similar line-of-sight paths are sorted in descending order to obtain the predicted sequence of highly similar line-of-sight paths.

[0008] As a further aspect of the present invention, correlation analysis is performed on the highly similar line-of-sight path prediction sequences to obtain an extended loading sequence, and a pre-cached data block set is obtained based on the extended loading sequence, including: Road model data is composed of road geometry data, road type attributes, and road-related building identifiers extracted from high-similarity line-of-sight path prediction sequences. Correlation analysis is then performed on the road model data to obtain the correlation analysis similarity. The initial loading sequence is obtained based on the highly similar line-of-sight path prediction sequence and the preliminary list of associated buildings; Set a similarity threshold for association analysis. If the similarity of association analysis is greater than the similarity threshold, perform texture parsing detail fusion to obtain scene elements corresponding to buildings and road extensions associated with the road, and integrate them into the initial loading sequence to obtain the extended loading sequence. The texture parsing detail fusion includes aligning the resolution of the road texture with the resolution of the building texture, correcting the coordinate offset between the road texture and the building texture, and supplementing the transition details of the vegetation texture around the road. The system iterates through the distances between the current viewpoint and the associated buildings on the road, and then segments the line-of-sight path using a grid partitioning algorithm to obtain a set of pre-cached data blocks.

[0009] As a further aspect of the present invention, the line-of-sight path is segmented based on a grid partitioning algorithm to obtain a set of pre-cached data blocks, including: The line-of-sight path is divided into multiple sub-regions based on a grid partitioning algorithm. Each sub-region corresponds to a grid cell, and the spatial coordinates of each grid cell are recorded. The size of each grid is adjusted by the curvature of the line-of-sight path and the data density. Associate the building geometry data, texture data, road model data, and vegetation distribution data within the coverage area of ​​each grid cell to generate corresponding pre-cache data blocks; Obtain the set of pre-cached data blocks based on the pre-cached data blocks corresponding to all grid cells.

[0010] As a further aspect of the present invention, the pre-cached data block set is optimized to obtain an optimized timing mismatch correction structure, including: Unify the spatial coordinate system of the data in the pre-cached data block set; Extract vegetation distribution data from the pre-cached data block set to obtain the vegetation distribution area; Obtain building geometric models based on a preliminary list of associated buildings; The hierarchical clustering algorithm is used to group adjacent vegetation distribution areas and building geometric models, and the temporal differences in data loading within each group are analyzed to identify temporal mismatch issues. Based on the aforementioned timing mismatch problem, the loading order and duration allocation of data within each group are adjusted to obtain an optimized timing mismatch correction structure.

[0011] As a further aspect of the present invention, the view rotation event is processed based on the temporal mismatch correction structure to obtain the complete texture parsing path, and potential nodes are extracted from the complete texture parsing path to obtain the complete loading sequence, including: Obtain the real-time view rotation angle and set the rotation angle threshold; If the real-time view rotation angle is greater than the rotation angle threshold, then update the building geometry and texture data corresponding to the current view. Identify associated buildings based on updated building geometry and texture data; For each associated building, trace its storage path, resolution level, and loading dependencies to form a complete texture parsing path; Potential nodes are extracted from the complete texture parsing path, and the complete loading sequence is obtained by combining historical data.

[0012] As a further aspect of the present invention, potential nodes are extracted from the complete texture parsing path, and a complete loading sequence is obtained by combining historical data, including: Analyze the extension direction of the complete texture parsing path and extract potential nodes of the extension path; Obtain historical potential nodes from historical data, and obtain the access frequency and dwell time of historical potential nodes. Calculate the attention weight of historical potential nodes based on the access frequency and dwell time. Based on the attention weight, similar matching data is obtained, and the extended loading sequence is verified based on the similar matching data to obtain the complete loading sequence.

[0013] As a further aspect of the present invention, memory allocation is performed on the complete loading sequence, real-time scene feature index is obtained, and a real-time 3D geographic scene is constructed based on the real-time scene feature index, including: Get the current memory usage status of the device; Memory allocation is performed on the complete loading sequence based on the memory usage status; Building geometry and texture data are stored using a dynamic caching mechanism; When memory is insufficient, the least recently accessed data block with the lowest priority is evicted. Real-time detection of data in the cache is performed to obtain the cache address, status identifier and access timestamp of the data block and form a real-time scene element index. The real-time scene element index is then input into the original system's 3D rendering module for real-time updates to obtain the real-time 3D geographic scene.

[0014] Furthermore, embodiments of the present invention also provide a three-dimensional geographic scene construction system, comprising: The acquisition module is used to acquire building geometric data and texture data from the user's operating perspective, acquire a preliminary list of associated buildings, acquire historical data, acquire a set of pre-cached data blocks based on the extended loading sequence, and construct a real-time three-dimensional geographic scene based on the real-time scene element index. The calculation module performs similarity calculations on the preliminary list of associated buildings based on historical data to obtain a high-similarity line-of-sight path prediction sequence; The analysis module is used to perform correlation analysis on highly similar line-of-sight path prediction sequences to obtain extended loading sequences; The optimization module is used to optimize the pre-cached data block set and obtain an optimized timing mismatch correction structure. The processing module processes the view rotation event based on the timing mismatch correction structure to obtain the complete texture parsing path; The extraction module extracts potential nodes from the complete texture parsing path to obtain the complete loading sequence; The allocation module is used to allocate memory for the complete loading sequence and obtain the real-time scene element index.

[0015] Based on the above, this application embodiment achieves dynamic filtering of building geometry data, texture data, and vegetation distribution data corresponding to the current viewpoint through hierarchical indexing technology. It combines the k-nearest neighbor algorithm to perform similarity matching on the user's historical browsing path to predict high-similarity view paths, overcoming the difficulty of traditional full-data loading modes that do not consider real-time viewpoint and user habits, resulting in excessively long loading times and ensuring timely data loading. It performs spatial and attribute association analysis between road model data in high-similarity view paths and initially associated buildings, sets association similarity thresholds to filter suitable data, optimizes data matching accuracy through texture parsing detail fusion, and divides the view path into adaptive-sized grid cells based on a grid partitioning algorithm and associates corresponding grid cells. Scene data is generated into pre-cached data blocks. A hierarchical clustering algorithm is constructed to group building and vegetation data to obtain a complete loading sequence, which improves the scheduling and matching efficiency of scene data, thereby obtaining a temporally coordinated pre-cached data set. A dynamic caching mechanism is used to allocate cache resources based on memory state, update scene element indexes in real time and input them into the 3D rendering module. Combined with view rotation events, associated building data and complete texture parsing paths are dynamically updated. The complete loading sequence is supplemented through potential node verification. On the basis of the traditional 3D geographic scene system, a 3D geographic scene construction framework is built to achieve low-latency loading, high-fidelity rendering and multi-view switching of 3D geographic scenes, thereby improving the efficiency and interactivity of the 3D geographic scene construction process. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the execution flow of a three-dimensional geographic scene construction method provided in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of a three-dimensional geographic scene construction system provided in an embodiment of the present invention. Detailed Implementation

[0018] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a schematic diagram of the execution flow of a three-dimensional geographic scene construction method provided in one embodiment of the present invention. The following is a detailed description of this three-dimensional geographic scene construction method.

[0019] Specifically, a method for constructing a three-dimensional geographic scene includes: It should be noted that the 3D scene system is a pre-built spatial information processing system. This system is built using 3D geographic information acquisition and modeling software, and users can directly browse geographic 3D scenes through the 3D scene system.

[0020] Step S1: Obtain the geometric and texture data of the buildings from the user's perspective, and obtain a preliminary list of associated buildings.

[0021] In this embodiment, step S1 includes: Step S11: Construct a multi-level index structure consisting of a geographic region level, a building type level, and a data precision level.

[0022] In some possible implementations, the geographic region level can be divided by spatial range, typically into four levels: city-administrative region-block-plot, thereby enabling precise location of the user's spatial range; the building type level is classified according to functional attributes, including residential buildings, commercial buildings, public buildings, industrial buildings, etc.; the data precision level is divided according to the level of detail in geometry and texture, and can be divided into three levels: low precision, medium precision, and high precision.

[0023] It should be noted that if the viewing distance to the target is within 50 meters, high-precision data is extracted; if the viewing distance to the target is between 50 and 1000 meters, medium-precision data is extracted; and if the viewing distance to the target is more than 1000 meters, low-precision data is extracted. The high-precision data has an accuracy of 0.1 meters, the medium-precision data has an accuracy of 1 meter, and the low-precision data has an accuracy of 10 meters.

[0024] Step S12: Obtain the region to which the current view belongs through the geographic region level, obtain the target type building through the building type level, and extract the building geometric data and texture data of the corresponding precision through the data precision level.

[0025] In some possible embodiments, when a user's perspective is 30 meters away from a CBD area and browsing Tower A of the China World Trade Center through a 3D scene system, the location is determined by the geographical region level as XX City-XX District-CBD core block-China World Trade Center plot, the building type level is defined as commercial office building, and the data is determined to be high-precision data by the data accuracy level. The building geometric data may include the building's vertex coordinates, roof outline, window position and size, etc., and the texture data includes glass reflection texture, stone exterior wall texture, etc.

[0026] Step S13: Obtain vegetation distribution data of the area to which the current view belongs. The vegetation distribution data includes vegetation type, vegetation density and vegetation spatial coordinates.

[0027] Specifically, the system retrieves vegetation databases associated with geographic regions to obtain vegetation distribution data for the area to which the current viewpoint belongs. The vegetation types can be classified into street trees, lawns, shrubs, etc. through biological classification, and the vegetation density can be expressed as trees per square meter or coverage percentage.

[0028] In some possible embodiments, the vegetation distribution data around the China World Trade Center is obtained: street trees and lawns; there are 2 street trees every 10 meters, and the lawn coverage is 90%; the coordinates of the center point of the trunk of a certain street tree and the boundary coordinates of the lawn.

[0029] Step S14: Spatially integrate the building geometry data, texture data and vegetation distribution data to obtain building elements, vegetation elements and spatial relationships, and form an initial scene element set.

[0030] Specifically, building geometric data and texture data are combined to form building elements, which include building ID, geometric model, and texture mapping rules; vegetation type, vegetation density, and vegetation spatial coordinates are combined to form vegetation elements, which include vegetation ID, type identifier, and spatial range; the relationship between the two is calculated through spatial coordinate matching, such as the distance between building entrances and street trees, and the topological relationship between rooftop greening and the main building; building elements, vegetation elements, and spatial relationships are grouped by geographical region to form an initial scene element set.

[0031] In some possible implementations, taking a certain Guomao plot as an example, the geometric model and curtain wall texture of Tower A of the Guomao Building are combined into building elements with ID B001. The 10 surrounding ginkgo trees with IDs V001-V010 and the lawn on the east side are combined into vegetation elements with ID V011. Relationships are established through coordinate calculations, such as the distance between the main entrance of B001 and the ginkgo tree of V003 being 2 meters away, and the boundary of the east outer wall of B001 coinciding with the boundary of the lawn of V011. These elements and their relationships are combined into the initial scene element set of the Guomao plot.

[0032] Step S15: Perform preliminary matching on the buildings in the initial scene element set to obtain a preliminary list of matched associated buildings.

[0033] It should be noted that the preliminary matching conditions include spatial distance, visual visibility, importance weight, etc. Through these filtering conditions, information such as building ID, name, and location index is obtained, the importance weight of this information is obtained, and sorted to obtain a preliminary list of matched associated buildings.

[0034] In some possible embodiments, if the matching conditions are set as follows: within a 300-meter viewing range, unobstructed view, commercial office building type, and importance weight greater than 80 points, the buildings that meet the conditions after filtering are China World Trade Center Tower A (B001, weight 95), China World Trade Center Tower B (B002, weight 90), and China World Trade Center Tower D (B003, weight 85). The buildings are then sorted in descending order of importance weight to obtain a preliminary list of matched buildings [B001-China World Trade Center Tower A, B002-China World Trade Center Tower B, B003-China World Trade Center Tower D].

[0035] Step S2: Obtain historical data and perform similarity calculation based on the preliminary list of associated buildings to obtain a high-similarity line-of-sight path prediction sequence.

[0036] In this embodiment, step S2 includes: Step S21: Obtain historical data consisting of multiple historical browsing paths, the dwell time corresponding to each path, the frequency of perspective switching, and the sequence of buildings of interest.

[0037] In some possible implementations, taking a certain Guomao area as an example, the browsing data of users in that Guomao area over the past 3 months is collected. The historical browsing path is from Guomao Tower A to Guomao Tower B and then to Guomao Tower D. The dwell time is 120 seconds in Guomao Tower B and 60 seconds in Guomao Tower B. The perspective switching frequency is once every 2 seconds. The sequence of buildings of interest is B002-Guomao Tower B and B003-Guomao Tower D. Multiple user data are obtained to form the historical data of this area.

[0038] Step S22: Extract the path direction angle, average viewing height, building attention weight, and path length from the historical data to form a historical feature vector.

[0039] Specifically, the path direction angle is obtained by calculating the angle between the line connecting adjacent coordinate points in historical data and due north; the average viewing height is obtained by calculating the average vertical distance between the viewing angle and the ground during browsing; the building attention weight is calculated by the dwell time and the viewing angle switching frequency; and the path length is obtained by calculating the total spatial distance of the browsing path. The path direction angle, average viewing height, building attention weight, and path length are combined to form a historical feature vector.

[0040] In some possible embodiments, the path direction angle is extracted from the historical data of the China World Trade Center from Tower A to Tower D and back to Tower D. The direction angle is 30° from Tower A to Tower B and 60° from Tower B to Tower D, with an average of 45°. The average viewing height is 50 meters throughout the browsing process. The building attention weight is 0.8 for Tower B and 0.5 for Tower D, with an average of 0.65. The path length is a total distance of 800 meters. The historical feature vector of the path is formed as [45°, 50 meters, 0.65, 800 meters].

[0041] Step S23: Obtain the current view feature vector and obtain the similarity score between the current view feature vector and the historical feature vector based on the k-nearest neighbor algorithm.

[0042] Specifically, the current view feature vector is constructed by referring to the historical feature vectors. The path direction angle corresponds to the movement direction angle of the current view, the average view height corresponds to the current view height, the building attention weight corresponds to the average weight of the preliminary associated building list, and the path length corresponds to the length of the currently viewed path. The k nearest neighbors algorithm is used to obtain the k neighbors that are most similar to the current view feature vector in the historical feature vectors, and the similarity of each neighbor is calculated by Euclidean distance. The weighted average of the similarity of the k neighbors is obtained, which is the similarity score between the current view feature vector and the historical feature vector.

[0043] Furthermore, the weights of the k current view feature vectors are obtained through the usage frequency of the neighbors.

[0044] It should be noted that the k-nearest neighbor algorithm finds the k nearest neighbors by calculating the distance between the sample to be predicted and the samples in the training set, and predicts the target based on the category or value of these neighbors.

[0045] In some possible embodiments, assuming the current user's viewpoint feature vector is obtained as [42°, 48 meters, 0.7, 750 meters], and the k value in the k-nearest neighbor algorithm is set to 5, the five most similar neighbors are found from the historical feature vector set. Assuming their Euclidean distances are 5, 8, 10, 12, and 15, the corresponding similarities are 0.2, 0.125, 0.1, 0.083, and 0.067, respectively. The similarity is specifically expressed as 1 / distance. Assuming the usage frequencies of these five neighbors are 30, 25, 20, 15, and 10, the weights of these five neighbors are 0.3, 0.25, 0.2, 0.15, and 0.1, respectively. Then the similarity score between the current viewpoint feature vector and the historical feature vector is 0.145.

[0046] Step S24: Set a similarity score threshold and filter the line-of-sight paths corresponding to historical feature vectors with similarity scores greater than the similarity score threshold as high similarity line-of-sight paths.

[0047] In some possible embodiments, the similarity score threshold is set to 0.12. If the user's current similarity score is 0.145, which is greater than the similarity score threshold of 0.12, then the historical line-of-sight path corresponding to the user's current similarity score is taken as the high similarity line-of-sight path.

[0048] Step S25: Sort the similarity scores of the high-similarity line-of-sight paths in descending order to obtain the high-similarity line-of-sight path prediction sequence.

[0049] In some possible embodiments, it is assumed that three highly similar line-of-sight paths are obtained from step S24 as path 1, path 2, and path 3, with similarity scores of 0.145, 0.132, and 0.125 respectively. These three highly similar line-of-sight paths are then sorted in descending order according to their corresponding similarity scores to obtain a highly similar line-of-sight path prediction sequence [path 1, path 2, path 3], where path 1 is the highest priority prediction path.

[0050] Step S3: Perform correlation analysis on the high similarity line-of-sight path prediction sequence to obtain the extended loading sequence, and obtain the pre-cached data block set based on the extended loading sequence.

[0051] In this embodiment, step S3 includes: Step S31: Extract road geometry data, road type attributes, and road-related building identifiers from the high-similarity line-of-sight path prediction sequence to form road model data, perform association analysis on the road model data, and obtain the association analysis similarity.

[0052] Specifically, the road geometry data includes data such as road centerline coordinates, road width, number of lanes, and road curvature; road type attributes cover road level, such as expressway, urban arterial road, secondary road, road surface material, and traffic function, such as motor vehicle lane and non-motor vehicle lane; road-related building identifiers are the building IDs and entrance / exit location indexes directly connected to the road; and similarity calculations are performed on the road model data from two dimensions: spatial association and attribute association, to obtain association analysis similarity.

[0053] In some possible embodiments, road model data is extracted from path 1 of the high-similarity line-of-sight path prediction sequence. Specifically, the road geometry data represents the auxiliary road of the Guomao Bridge: centerline coordinates 39.9160-39.9170° N, 116.4640-116.4650° E, road width 15 meters, two-way four lanes, curvature 0.08; the road type attribute is specifically represented as urban arterial road, asphalt pavement, mixed traffic of motor vehicles and non-motor vehicles; and the road-related building identifier is specifically represented as B00. 1. Guomao Building A, entrance and exit are located 2 meters west of the road; 2. Yintai Center Guomao Building D, entrance and exit are located 5 meters north of the road. The average distance between B001 and B003 is 3.5 meters. The spatial correlation is calculated to be 0.92. The main road has a matching score of 95 with B001 and a matching score of 85 with B003. The attribute correlation is calculated to be 0.88. The weights of spatial correlation and attribute correlation are set to 0.6 and 0.4 respectively. The similarity of the correlation analysis is 0.904.

[0054] Step S32: Obtain the initial loading sequence based on the high similarity line-of-sight path prediction sequence and the preliminary list of associated buildings.

[0055] Specifically, basic data for each building is extracted from the initial list of associated buildings as the core of the sequence, including building geometry data, texture data, etc.; road model data directly associated with each building is extracted from the high similarity line-of-sight path prediction sequence, including road geometry data, type attributes, etc.; road data is mapped to relevant buildings through the spatial connection relationship between buildings and roads; and the initial loading sequence is obtained by sorting the buildings according to the rules of distance from the current viewpoint from near to far and the association strength between roads and buildings from high to low.

[0056] In some possible embodiments, based on the preliminary list of associated buildings [B001-China World Trade Center Tower A, B002-China World Trade Center Tower B, B003-China World Trade Center Tower D], low-precision geometric data and basic texture data of the three are extracted; the basic road model data of the Guomao Bridge auxiliary road is extracted from the high similarity line-of-sight path prediction sequence and associated with B001 and B003; the order is adjusted according to distance and association strength, assuming that B001 is 30 meters away from the current viewpoint, B003 is 50 meters away, B002 is 60 meters away, and the road association degree of B001 is 0.92, which is greater than the road association degree of B003 is 0.88, to obtain the initial loading sequence [B001-China World Trade Center Tower A, Guomao Bridge Auxiliary Road, B003-China World Trade Center Tower D, B002-China World Trade Center Tower B].

[0057] Step S33: Set the association analysis similarity threshold. If the association analysis similarity is greater than the association analysis similarity threshold, perform texture parsing detail fusion to obtain the scene elements corresponding to the buildings and road extensions associated with the road, and integrate them into the initial loading sequence to obtain the extended loading sequence.

[0058] The texture parsing detail fusion includes aligning the resolution of the road texture with that of the building texture, correcting the coordinate offset between the road texture and the building texture, and supplementing the transition details of the vegetation texture around the road.

[0059] In some possible embodiments, assuming the similarity threshold for association analysis is set to 0.7, and the similarity obtained through step S31 is 0.904, which is greater than the similarity threshold of 0.7, the asphalt texture resolution of the Guomao Bridge auxiliary road is increased from 1024×1024 to 4096×4096, which is consistent with the texture of the B001 curtain wall. The coordinate offset between the Guomao Bridge auxiliary road and B001 is 0.2 meters, which is corrected, and the texture transition details of the ginkgo trees on both sides of the road are added. The road extension scene elements are added: the B004-Fortune Center and the extension road model within 50 meters eastward extension of the Guomao Bridge auxiliary road, and they are integrated into the initial loading sequence to obtain the extended loading sequence [B001, Guomao Bridge auxiliary road, B003, B004, extension road, B002].

[0060] It should be noted that if the similarity of the association analysis is less than or equal to the similarity threshold of the association analysis, it means that the spatial adaptability and attribute matching degree of the road model and the initially associated buildings in the current high similarity line-of-sight path prediction sequence do not meet the requirements for scene construction. The loading ability of basic scene elements should be retained to ensure that the data is loaded normally.

[0061] Step S34: Traverse the road based on the distance between the current viewpoint and the associated buildings, segment the line-of-sight path based on the grid partitioning algorithm, and obtain a set of pre-cached data blocks.

[0062] In this embodiment, step S34 includes: Step S341: Based on the grid partitioning algorithm, the line-of-sight path is divided into multiple sub-regions, each sub-region corresponding to a grid cell, and the spatial range coordinates of each grid cell are recorded.

[0063] The size of each grid is adjusted by the curvature of the line-of-sight path and the data density.

[0064] Specifically, the curvature of each point on the line of sight is obtained, the building density around the path is obtained simultaneously, the side length of the grid is dynamically adjusted based on the curvature and building density, the line of sight is divided into multiple sub-regions, and the latitude and longitude coordinates of the diagonal of each sub-region are recorded simultaneously.

[0065] Understandably, path segments with curvature less than 0.1 are defined as straight path segments, and path segments with curvature greater than or equal to 0.1 are defined as curved path segments. Building density less than 3 buildings per thousand square meters is defined as sparse building, and building density greater than or equal to 3 buildings per thousand square meters is defined as dense building. For straight path segments with sparse buildings, a large grid with a side length of 50 meters is used for division, and for curved path segments or dense building segments, a small grid with a side length of 10 meters is used for division.

[0066] Step S342: Associate the building geometry data, texture data, road model data and vegetation distribution data within the coverage area of ​​each grid cell to generate the corresponding pre-cached data block.

[0067] Specifically, each grid cell is traversed, and the scene data within the coverage area of ​​each grid cell is filtered out, including building geometry data, texture data, road model data, and vegetation distribution data. Data tags such as data block ID, corresponding grid cell identifier, data type classification, data size, and loading priority are added to obtain the corresponding pre-cached data blocks.

[0068] It should be noted that the loading priority is specifically defined as follows: a distance of less than 50 meters between the grid and the current viewpoint is set as priority 1; a distance of 50-100 meters between the grid and the current viewpoint is set as priority 2; and a distance of more than 100 meters between the grid and the current viewpoint is set as priority 3.

[0069] In some possible embodiments, it is assumed that the scene data obtained from the mesh cell G002 includes the building geometry data as the high-precision vertex coordinates of the exterior wall of B001, the texture data as the 4096×4096 reflection texture of the glass curtain wall of B001, the road model data as the local geometry and asphalt texture of the auxiliary road of Guomao Bridge, and the vegetation data as the roadside tree information of V003-V005. A data tag with data block ID D002, mesh identifier G002, data size 180MB, and loading priority 1 is added to generate a pre-cached data block D002.

[0070] Step S343: Obtain a set of pre-cached data blocks based on the pre-cached data blocks corresponding to all grid cells.

[0071] Understandably, the pre-cached data blocks corresponding to all grid cells can be sorted by loading priority to obtain the set of pre-cached data blocks.

[0072] Step S4: Optimize the pre-cached data block set to obtain the optimized timing mismatch correction structure.

[0073] In this embodiment, step S4 includes: Step S41: Unify the spatial coordinate system of the data in the pre-cached data block set.

[0074] It should be noted that the data in the pre-cached data block set may come from different acquisition devices, resulting in inconsistent coordinate systems. Therefore, the spatial coordinate system of all data should be unified.

[0075] In some possible implementations, all data can be uniformly converted into a standard coordinate system for scene construction, such as a geodetic coordinate system.

[0076] Step S42: Extract vegetation distribution data from the pre-cached data block set to obtain the vegetation distribution area.

[0077] Specifically, based on the vegetation distribution data, and by dividing the region according to vegetation type and spatial continuity, the spatially adjacent and type-identical vegetation is grouped into the same vegetation distribution area.

[0078] In some possible embodiments, the vegetation distribution data obtained includes ginkgo street trees V001-V010 and lawns V010-V012. The ginkgo street trees V001-V010 are divided into ginkgo street trees R001 and ginkgo street trees R002 based on vegetation type and spatial continuity. R001 includes V001-V005, R002 includes V005-V010, and lawns R003 include V010-V012. A vegetation distribution area is formed based on R001, R002, and R003.

[0079] Step S43: Obtain the building geometric model based on the preliminary list of associated buildings.

[0080] Specifically, using the preliminary list of associated buildings as the filtering criteria, the corresponding building geometric data is extracted from the pre-cached data block set, and the building geometric model is constructed through a 3D modeling algorithm.

[0081] Furthermore, the 3D modeling algorithm specifically involves connecting discrete vertex coordinates to construct the basic outline of the building, while supplementing the geometric information of detailed structures such as roofs, doors, and windows, establishing the topological relationships of the model, such as the vertical relationship between the wall and the ground, and the relationship between the window and the wall, to obtain the geometric model of the building.

[0082] Step S44: Group adjacent vegetation distribution areas and building geometric models based on hierarchical clustering algorithm, and analyze the temporal differences in data loading within each group to identify temporal mismatch issues.

[0083] Specifically, based on the Euclidean distance between the center point of the building model and the center point of the vegetation area, and the matching degree between the building height and the average height of the vegetation, buildings and vegetation areas with a distance of less than or equal to 20 meters and an attribute similarity of greater than or equal to 0.7 are grouped together. The loading logs of the data in each group are retrieved, and the loading completion time of the building geometric model, vegetation data, and texture data is statistically analyzed to identify temporal differences and obtain temporal mismatch issues.

[0084] In some possible implementations, the geometric models of B001-B003 and the vegetation regions of R001-R003 are clustered. The distance between the center point of B001 and the center point of R001 is 8 meters, the height of B001 is 150 meters, the height of the ginkgo tree in R001 is 10 meters, and the attribute similarity is 0.93, so they are merged into group G1; the distance between B002 and R002 is 12 meters, and the attribute similarity is 0.91, so they are merged into group G2; the distance between B003 and R003 is 15 meters, and the attribute similarity is 0.85, so they are merged into group G3. By analyzing the loading log, it was found that the loading completion time of B001 in group G1 was 0.5 seconds, while the loading completion time of R001 was 1 second, which was determined to be a time-series mismatch problem of vegetation data loading lag.

[0085] Step S45: Based on the timing mismatch problem, adjust the loading order and duration allocation of data in each group to obtain an optimized timing mismatch correction structure.

[0086] In some possible embodiments, if there is a problem of vegetation data loading lag in G1, the loading trigger time of R001 is adjusted to be when B001 loads to 25% progress, and the loading time is extended from 0.3 seconds to 0.4 seconds. The dependency relationship R001 loading depends on B001 geometric data loading being greater than or equal to 25%. Similarly, the loading parameters of groups G2 and G3 are adjusted to obtain the optimized timing mismatch correction structure.

[0087] Step S5: Process the view rotation event based on the temporal mismatch correction structure, obtain the complete texture parsing path, extract potential nodes from the complete texture parsing path, and obtain the complete loading sequence.

[0088] In this embodiment, step S5 includes: Step S51: Obtain the real-time view rotation angle and set the rotation angle threshold.

[0089] Specifically, the real-time view rotation angle is collected in real time through the sensors of the scene browsing device. The sampling frequency can be set to 30Hz, and the rotation angle threshold can be set.

[0090] In some possible embodiments, assuming the user rotates the viewpoint by dragging with a mouse, the real-time viewpoint rotation angle can be collected by means of a mouse gyroscope or a VR headset posture sensor, and a rotation angle threshold of 30° can be set.

[0091] Step S52: If the real-time view rotation angle is greater than the rotation angle threshold, then update the building geometry data and texture data corresponding to the current view.

[0092] Specifically, the rotational field of view is recalculated, and architectural geometry and texture data within the new field of view are extracted from the multi-level index through hierarchical queries.

[0093] In some possible embodiments, assuming the real-time view rotation angle is 35°, which is greater than the set threshold of 30°, a view rotation event is triggered. After the view rotation, the new view range adds B005 in Tower E of the China World Trade Center. B003 in the original view is still visible. The medium-precision data and texture data of B005 are extracted by hierarchical query, the high-precision data of B003 is retained, the low-precision data of B002 is temporarily stored, and the data is updated.

[0094] Step S53: Obtain the associated buildings based on the updated building geometry and texture data.

[0095] In some possible embodiments, five related buildings are selected from the updated building geometry and texture data: B001 (China World Trade Center Tower A, 20 meters from New Perspective, 95% attention), B003 (China World Trade Center Tower D, 50 meters, 90%), B005 (Kerry Centre, 80 meters, 85%), B004 (Fortune Centre, 120 meters, 80%), and B006 (China Merchants Tower, 150 meters, 75%).

[0096] Step S54: Trace each associated building to obtain the storage path, resolution level, and loading dependency relationship, and form a complete texture parsing path.

[0097] Specifically, for each associated building, the storage path, resolution level, and loading dependency are traced from the multi-level index. This information is then organized into a structure of building ID-storage path-resolution level-dependency to obtain the complete texture parsing path for each building.

[0098] In some possible embodiments, it is assumed that the B005 parsing traces back to obtain the storage path / storage / texture / B005 / , the resolution levels are low (512×512, / B005_low.jpg), medium (2048×2048, / B005_mid.jpg), and high (4096×4096, / B005_high.jpg), and the loading dependency relationship is that the medium resolution depends on the low resolution to be loaded, and the high resolution depends on the medium resolution to be loaded. Similarly, the complete texture parsing path of other buildings can be obtained through the above method.

[0099] Step S55: Extract potential nodes from the complete texture parsing path and combine them with historical data to obtain the complete loading sequence.

[0100] In this embodiment, step S55 includes: Step S551: Analyze the extension direction of the complete texture parsing path and extract potential nodes of the extension path.

[0101] Specifically, the extension direction of the complete texture parsing path of each associated building is analyzed, and key spatial nodes in the path extension direction are identified, including building vertices, road intersections, and the center point of dense vegetation areas, as potential nodes for the extension path.

[0102] In some possible embodiments, the extension direction of the B005 texture parsing path is analyzed to extract potential nodes: B005 roof vertex (node ​​P1, coordinates 39.9170°, 116.4660°), the intersection of East Third Ring Road and Guanghua Road (node ​​P2, 39.9175°, 116.4670°), and the center of the lawn on the east side of Kerry Centre (node ​​P3, 39.9168°, 116.4665°).

[0103] Step S552: Obtain historical potential nodes from historical data, and obtain the access frequency and dwell time of historical potential nodes. Calculate the attention weight of historical potential nodes based on the access frequency and dwell time.

[0104] Specifically, all historical potential nodes visited by users are extracted from historical data, and the access frequency and dwell time of each historical potential node are obtained. The maximum access frequency and maximum dwell time are also obtained. The access frequency / maximum access frequency and dwell time / maximum dwell time are weighted and assigned to obtain the attention weight of the historical potential node.

[0105] In some possible embodiments, historical potential nodes related to the CBD area are obtained, wherein it is assumed that the access frequency of node P2 is 120 times per week, the maximum access frequency is 150 times per week, the dwell time of node P2 is 45 seconds, the maximum dwell time is 60 seconds, and its attention weight is calculated to be 0.78.

[0106] Step S553: ​​Obtain similar matching data based on the attention weight, verify the extended loading sequence based on the similar matching data, and obtain the complete loading sequence.

[0107] Specifically, nodes with an attention weight greater than or equal to 0.6 are obtained, and their corresponding buildings, roads, vegetation, and other scenes are extracted as similarity matching data. The similarity matching data is used to verify and identify uncovered scene elements in the extended loading sequence, and the uncovered scene elements are used to supplement the extended loading sequence to obtain a complete loading sequence.

[0108] In some possible embodiments, the attention weight of the obtained P2 node is 0.78, which is greater than 0.6. The corresponding scene data is extracted, including the surrounding building B007 Guanghua Road, the connecting road East Third Ring Road main road, and the vegetation area R005 intersection green belt. The extended loading sequence [B001, B002, B003, B004, Guomao Bridge auxiliary road] is checked. It is found that B007, East Third Ring Road main road, and R005 are not covered. These elements are added to the extended loading sequence to generate a complete loading sequence.

[0109] Step S6: Allocate memory for the complete loading sequence, obtain the real-time scene feature index, and construct a real-time 3D geographic scene based on the real-time scene feature index.

[0110] In this embodiment, step S6 includes: Step S61: Obtain the current memory usage status of the device.

[0111] Specifically, the memory usage status includes total memory capacity, remaining memory capacity, allocated memory capacity, and the type classification of allocated memory. The types of allocated memory include geometric data memory, texture data memory, and system reserved memory. Memory usage rate is also acquired synchronously. When the memory usage rate exceeds a set memory usage threshold, it indicates insufficient memory.

[0112] In some possible embodiments, the total memory of a certain 3D geographic scene construction system is obtained as 64GB. The memory status is obtained through performance detection, including memory capacity of 18GB, allocated memory of 46GB, of which geometric data memory is 15GB, texture data memory is 25GB, and system reserved memory is 6GB. The memory utilization rate is obtained simultaneously as 71.9%. The memory utilization threshold is set as 80%. If the threshold is not exceeded, memory is allocated according to the conventional strategy.

[0113] Step S62: Allocate memory for the complete loading sequence based on the memory usage status.

[0114] Specifically, memory allocation rules are formulated based on the data type and priority of each element in the complete loading sequence. The memory allocation rules are specifically expressed as follows: elements with priority 1, such as geometric data and high-resolution texture data visible from the current viewpoint, are allocated contiguous physical memory first; elements with priority 2, such as high-interest node data and medium-resolution texture data, are allocated non-contiguous physical memory; elements with priority 3, such as extension segment data and low-resolution texture data, are not allocated memory for the time being, and only disk cache space is reserved; if there is insufficient memory, priority 1 elements' memory allocation is guaranteed first, and the memory usage of priority 2 elements is reduced.

[0115] Step S63: Store building geometry data and texture data based on a dynamic caching mechanism.

[0116] When memory is insufficient, the least recently accessed data block with the lowest priority is evicted.

[0117] In some possible embodiments, the building geometry data and texture data in the cache can be managed through a dynamic caching mechanism such as LRU. For example, cache data blocks D005 (B001 texture data, last accessed 10 seconds ago, priority 1), D008 (B002 geometry data, last accessed 60 seconds ago, priority 3), and D010 (B004 texture data, last accessed 45 seconds ago, priority 3). When a new data block D015 (B005 texture data, priority 2) needs to be loaded, if there is insufficient memory, LRU eviction is triggered, the earliest recently accessed D008 is evicted, 2GB of memory is released, and D0015 is loaded.

[0118] Step S64: Real-time detection of data in the cache, obtaining the cache address, status identifier and access timestamp of the data block and forming a real-time scene element index, inputting the real-time scene element index into the original system's 3D rendering module for real-time update, and obtaining the real-time 3D geographic scene.

[0119] In some possible embodiments, the status of cached data blocks is detected in real time. For example, the cache address of D005 (B001 texture data) is 0x0000028F4A100000, the status is "loaded", and the access timestamp is 2025-08-30-15:30:20. This information is organized into a real-time scene element index and input into the original system's 3D rendering module. The 3D rendering module retrieves the data through the index and renders the 3D morphology of the China World Trade Center Tower A, China World Trade Center Tower B, and surrounding roads and vegetation in real time, ultimately forming a real-time 3D geographic scene of the CBD area.

[0120] Figure 2 The diagram illustrates a three-dimensional geographic scene construction system that can realize the ideas of this application, according to some embodiments of this application.

[0121] Specifically, a geographic information mapping data processing method system includes: The acquisition module is used to acquire building geometric data and texture data from the user's operating perspective, acquire a preliminary list of associated buildings, acquire historical data, acquire a set of pre-cached data blocks based on the extended loading sequence, and construct a real-time three-dimensional geographic scene based on the real-time scene element index. The calculation module performs similarity calculations on the preliminary list of associated buildings based on historical data to obtain a high-similarity line-of-sight path prediction sequence; The analysis module is used to perform correlation analysis on highly similar line-of-sight path prediction sequences to obtain extended loading sequences; The optimization module is used to optimize the pre-cached data block set and obtain an optimized timing mismatch correction structure. The processing module processes the view rotation event based on the timing mismatch correction structure to obtain the complete texture parsing path; The extraction module extracts potential nodes from the complete texture parsing path to obtain the complete loading sequence; The allocation module is used to allocate memory for the complete loading sequence and obtain the real-time scene element index.

[0122] The specific usage and function of this embodiment are explained below: First, geometric and texture data of buildings from the user's perspective are acquired to obtain a preliminary list of associated buildings. Next, historical data is acquired, and similarity calculations are performed on the preliminary list of associated buildings to obtain high-similarity line-of-sight path prediction sequences. This overcomes the difficulty of traditional full-data loading modes, which do not consider real-time perspective and user habits, resulting in excessively long loading times, ensuring timely data loading. Then, association analysis is performed on the high-similarity line-of-sight path prediction sequences to obtain extended loading sequences. Based on these extended loading sequences, a set of pre-cached data blocks is obtained. This pre-cached data block set is then optimized to obtain an optimized time-series mismatch correction structure. Finally, the time-series mismatch is further optimized. The sequence mismatch correction structure processes viewpoint rotation events, obtains the complete texture parsing path, extracts potential nodes from the complete texture parsing path, and obtains the complete loading sequence, improving the scheduling and matching efficiency of scene data. This leads to the acquisition of a temporally coordinated pre-cached data set. Finally, memory is allocated for the complete loading sequence, real-time scene element indexes are obtained, and real-time 3D geographic scenes are constructed based on these indexes. A 3D geographic scene construction framework is built on the basis of traditional 3D geographic scene systems, achieving low-latency loading, high-fidelity rendering, and multi-view switching of 3D geographic scenes, thus improving the efficiency and interactivity of the 3D geographic scene construction process.

[0123] Furthermore, embodiments of the present invention also provide an electronic device, comprising: At least one processor; and at least one memory communicatively connected to the processor; wherein the memory stores instructions executable by at least one processor, the instructions being executed by at least one processor to enable at least one processor to perform the method proposed in Embodiment 1 of the present invention.

[0124] The following is a detailed introduction to the various components of the electronic device: In this context, the processor is the control center of the electronic device. It can be a single processor or a collective term for multiple processing elements. For example, a processor can be one or more central processing units (CPUs), specific integrated circuits (SIs), or one or more integrated circuits configured to implement Embodiment 1 of this invention, such as one or more microprocessors (DSPs), or one or more field-programmable gate arrays (FPGs).

[0125] The processor can perform various functions of an electronic device by running or executing software programs stored in memory and by calling data stored in memory.

[0126] The memory is used to store the software program that executes the solution of the present invention, and the execution is controlled by the processor. For specific implementation methods, please refer to the above method embodiments, which will not be repeated here.

[0127] The memory can be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, a random access memory (RM) or other type of dynamic storage device capable of storing information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a read-only optical disc (D-ROM), or other optical disc storage, optical disk storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto. The memory can be integrated with the processor or exist independently and coupled to the processor through an interface circuit of an electronic device; this embodiment of the invention does not specifically limit this.

[0128] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via limited means (e.g., infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0129] It should be understood that the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, "and / or I-frame / P-frame interval offset rate" can represent three cases: existing alone, existing simultaneously with "and / or I-frame / P-frame interval offset rate", and existing alone. The I-frame / P-frame interval offset rate can be singular or plural. Additionally, the character " / " in this article generally indicates an "or" relationship between the preceding and following related objects, but it may also indicate an "and / or" relationship. Please refer to the context for a more accurate understanding.

[0130] It should be understood that, in the embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0131] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method for constructing a three-dimensional geographic scene, characterized in that, The method comprises: Obtaining building geometry data and texture data under a user operation view angle, and obtaining a preliminary associated building list; Obtaining historical data, and performing similarity calculation based on the preliminary associated building list to obtain a high-similarity line-of-sight path prediction sequence; Performing associated analysis on the high-similarity line-of-sight path prediction sequence to obtain an expanded loading sequence, and obtaining a pre-cached data block set based on the expanded loading sequence; Optimizing the pre-cached data block set to obtain an optimized timing mismatch correction structure; Processing a view angle rotation event based on the timing mismatch correction structure to obtain a complete texture analysis path, extracting potential nodes from the complete texture analysis path, and obtaining a complete loading sequence; Performing memory allocation on the complete loading sequence to obtain a real-time scene element index, and constructing a real-time three-dimensional geographic scene based on the real-time scene element index. 2.The method of claim 1, wherein, Obtaining building geometry data and texture data under a user operation view angle, and obtaining a preliminary associated building list, comprising: Building a multi-level index hierarchy structure of geographic area levels, building type levels and data precision levels; Obtaining the region to which the current view angle belongs through the geographic area levels, obtaining target type buildings through the building type levels, and extracting building geometry data and texture data of corresponding precision through the data precision levels; Obtaining vegetation distribution data of the region to which the current view angle belongs, the vegetation distribution data comprising vegetation type, vegetation density and vegetation spatial coordinates; Integrating the building geometry data, texture data and vegetation distribution data in space to obtain building elements, vegetation elements and spatial association relationships and form an initial scene element set; Performing preliminary matching on the buildings in the initial scene element set to obtain a preliminary matching associated building list. 3.The method of claim 1, wherein, Obtaining historical data, and performing similarity calculation based on the preliminary associated building list to obtain a high-similarity line-of-sight path prediction sequence, comprising: Obtaining historical data composed of multiple historical browsing paths, path corresponding stay time, view angle switching frequency and attention building sequence; Extracting path direction angle, average view angle height, building attention weight and path length from the historical data to form a historical feature vector; Obtaining a current view angle feature vector, and obtaining a similarity score of the current view angle feature vector and the historical feature vector based on a k-nearest neighbor algorithm; Setting a similarity score threshold, and screening line-of-sight paths corresponding to historical feature vectors with a similarity score greater than the similarity score threshold as high-similarity line-of-sight paths; Arranging the similarity scores of the high-similarity line-of-sight paths in descending order to obtain a high-similarity line-of-sight path prediction sequence.

4. The method of claim 1, wherein, Performing associated analysis on the high-similarity line-of-sight path prediction sequence to obtain an expanded loading sequence, and obtaining a pre-cached data block set based on the expanded loading sequence, comprising: Extracting road geometry data, road type attributes and road associated building identifiers from the high-similarity line-of-sight path prediction sequence to form road model data, performing associated analysis on the road model data to obtain an associated analysis similarity; Obtaining an initial loading sequence based on the high-similarity line-of-sight path prediction sequence and the preliminary associated building list; A correlation analysis similarity threshold is set, and if the correlation analysis similarity is greater than the correlation analysis similarity threshold, texture analysis detail fusion is performed, scene elements corresponding to buildings and road extensions associated with the road are obtained, and are integrated into the initial loading sequence to obtain an expanded loading sequence; The texture analysis detail fusion includes aligning the resolution of the road texture with the resolution of the building texture, correcting the coordinate offset of the road texture and the building texture, and supplementing the transition details of the road peripheral vegetation texture; Based on the distance between the current view angle and the associated buildings of the road, traversal is performed, and the line-of-sight path is segmented based on a grid division algorithm to obtain a pre-cached data block set.

5. The method of claim 4, wherein, The line-of-sight path is segmented based on the grid division algorithm to obtain the pre-cached data block set, including: The line-of-sight path is segmented into a plurality of sub-regions based on the grid division algorithm, each sub-region corresponds to a grid unit, and the spatial range coordinates of each grid unit are recorded; The size of each grid is adjusted based on the curvature change of the line-of-sight path and the data amount density; The building geometry data, texture data, road model data, and vegetation distribution data within the coverage range of each grid unit are associated to generate corresponding pre-cached data blocks; The pre-cached data block set is obtained based on the pre-cached data blocks corresponding to all grid units.

6. The method of claim 1, wherein, The pre-cached data block set is optimized to obtain an optimized timing mismatch correction structure, including: The spatial coordinate system of the data in the pre-cached data block set is unified; The vegetation distribution data in the pre-cached data block set is extracted to obtain vegetation distribution regions; The building geometry model is obtained based on the preliminary associated building list; The adjacent vegetation distribution regions and building geometry models are grouped based on a hierarchical clustering algorithm, and the timing difference of data loading in each group is analyzed to obtain a timing mismatch problem; The loading order and duration allocation of the data in each group are adjusted based on the timing mismatch problem to obtain an optimized timing mismatch correction structure.

7. The method of claim 1, wherein, The view angle rotation event is processed based on the timing mismatch correction structure to obtain a complete texture analysis path, potential nodes are extracted from the complete texture analysis path, and a complete loading sequence is obtained, including: The real-time view angle rotation angle is obtained, and a rotation angle threshold is set; If the real-time view angle rotation angle is greater than the rotation angle threshold, the building geometry data and texture data corresponding to the current view angle are updated; The associated buildings are obtained based on the updated building geometry data and texture data; Each associated building is traced to obtain a storage path, a resolution level, and a loading dependency relationship and to form a complete texture analysis path; Potential nodes are extracted from the complete texture analysis path, and a complete loading sequence is obtained by combining historical data.

8. The method of claim 7, wherein, Potential nodes are extracted from the complete texture analysis path, and a complete loading sequence is obtained by combining historical data, including: The extension direction of the complete texture analysis path is analyzed, and potential nodes of the extension path are extracted; The historical potential nodes of the historical data are obtained, and the access frequency and dwell time of the historical potential nodes are obtained, and the attention weight of the historical potential nodes is calculated based on the access frequency and dwell time. The similar matching data is obtained based on the attention weight, the extended loading sequence is verified based on the similar matching data, and a complete loading sequence is obtained.

9. The method of claim 1, wherein, Memory allocation is performed on the complete loading sequence, a real-time scene element index is obtained, and a real-time three-dimensional geographic scene is constructed based on the real-time scene element index, including: obtaining the current memory usage state of the device; performing memory allocation on the complete loading sequence based on the memory usage state; storing building geometry data and texture data based on a dynamic caching mechanism; wherein, when the memory is insufficient, the data block that is not accessed recently and has the lowest priority is eliminated; real-time detection is performed on the data in the cache, the cache address, state identifier and access timestamp of the data block are obtained and composed into a real-time scene element index, the real-time scene element index is input into the three-dimensional rendering module of the original system for real-time updating, and a real-time three-dimensional geographic scene is obtained.

10. A three-dimensional geographic scene construction system, characterized by, including: an obtaining module, configured to obtain building geometry data and texture data under a user operation view angle, obtain a preliminary associated building list, obtain historical data, obtain a pre-cache data block set based on an extended loading sequence, and construct a real-time three-dimensional geographic scene based on a real-time scene element index; a calculation module, configured to perform similarity calculation on the preliminary associated building list based on the historical data, and obtain a high-similarity line-of-sight path prediction sequence; an analysis module, configured to perform correlation analysis on the high-similarity line-of-sight path prediction sequence, and obtain an extended loading sequence; an optimization module, configured to optimize the pre-cache data block set, and obtain an optimized time sequence mismatch correction structure; a processing module, configured to process a view angle rotation event based on the time sequence mismatch correction structure, and obtain a complete texture analysis path; an extraction module, configured to extract potential nodes from the complete texture analysis path, and obtain a complete loading sequence; and an allocation module, configured to perform memory allocation on the complete loading sequence, and obtain a real-time scene element index.

Images