An air-ground integrated big data fusion method for urban measurement

Abstract

The application discloses an air-ground integrated big data fusion method for urban measurement, and relates to the technical field of data processing. The method comprises the following steps: acquiring air-ground integrated big data of a city; uniformly calibrating the air-ground integrated big data to a city standard coordinate system to obtain low spatio-temporal deviation data; acquiring a first conflict entity and a second conflict entity based on the low spatio-temporal deviation data; correcting the spatial positions of the first conflict entity and the second conflict entity in the low spatio-temporal deviation data to obtain a non-conflict big data model; and constructing a long short-term memory network based on the non-conflict big data model. The application integrates air-ground integrated big data such as urban spatial point clouds, sensor networks and remote sensing images, solves the problem of low fusion precision caused by the non-uniformity of multi-source data spatio-temporal benchmarks and the large format precision difference, eliminates spatio-temporal deviation and entity conflicts, improves data consistency and reliability, and avoids information faults and semantic conflicts.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically a method for air-ground integrated big data fusion for urban measurement. Background Technology

[0002] In the process of urban development and planning today, comprehensive and accurate measurement of cities is crucial for achieving refined management, precise decision-making, and efficient resource allocation. Urban measurement requires comprehensive consideration of various factors such as spatial structure, functional layout, population flow, and infrastructure. With technological advancements, data from single sources is no longer sufficient to meet the complex needs of urban measurement. Integrating data from different platforms, both air and ground-based, can provide more comprehensive and accurate urban information, helping to gain a deeper understanding of urban operational patterns and development trends, and providing strong support for sustainable urban development.

[0003] However, due to the lack of unified spatiotemporal benchmarks among air-to-ground data acquisition platforms, significant differences in data format and accuracy, and the absence of effective mechanisms for aligning features and semantic associations of multi-source heterogeneous data, the fused data suffers from information gaps and semantic conflicts. For example, discrepancies exist between aerial remote sensing data and ground sensor data in terms of time synchronization and spatial coordinate transformation, making it difficult to form complete spatiotemporal sequences of data on urban spatial structure and infrastructure. Simultaneously, insufficient real-time processing capabilities for massive amounts of data prevent the rapid integration of multi-source data to construct dynamic urban models, resulting in fragmented urban health diagnostics and hindering accurate decision-making in critical scenarios such as rainstorm and flood warnings and responses to public emergencies. Summary of the Invention

[0004] The purpose of this application is to provide a method for air-ground integrated big data fusion for urban measurement, in order to solve the technical problem that the low accuracy of air-ground integrated big data fusion makes it impossible to accurately identify special urban events.

[0005] To achieve the above objectives, this application provides the following technical solution:

[0006] A big data fusion method for urban measurement that integrates air and ground data includes:

[0007] Acquire integrated air-ground big data of the city; the integrated air-ground big data includes at least the city's spatial point cloud data, the city's sensor network data, and the city's remote sensing image data.

[0008] The integrated air-ground big data is uniformly calibrated to the urban standard coordinate system to obtain data with low spatiotemporal deviation; the urban standard coordinate system is at least used to eliminate the spatiotemporal deviation of different source data in the integrated air-ground big data.

[0009] Based on the low spatiotemporal deviation data, a first conflicting entity and a second conflicting entity are obtained; the first conflicting entity is any entity in the low spatiotemporal deviation data; the second conflicting entity is any entity in the low spatiotemporal deviation data that spatially overlaps with the first conflicting entity; the data sources of the first conflicting entity and the second conflicting entity are different.

[0010] The spatial positions of the first and second conflicting entities in the low spatiotemporal bias data are corrected to obtain a conflict-free big data model.

[0011] Based on the conflict-free big data model, a long short-term memory network is constructed; the long short-term memory network is used to identify special events in the city.

[0012] As a specific solution in the technical solution of this application, the acquisition of integrated air-ground big data of the city includes:

[0013] The city area is scanned by a drone equipped with LiDAR to obtain spatial point cloud data of the city.

[0014] Remote sensing image data of the city were acquired using low-altitude remote sensing satellites.

[0015] Sensor network data of the city is acquired based on fixed sensors deployed on major urban roads and in public areas.

[0016] As a specific solution in the technical solution of this application, the urban standard coordinate system includes a first-level reference network, a second-level structural network, a third-level coverage network, and a fourth-level dynamic network;

[0017] The primary benchmark network is a network composed of multiple benchmark points in the urban space; the benchmark points are deep-buried municipal benchmark points or anchor points of cross-river bridges; at least three benchmark points are set up in each administrative division unit in the city to form a spatial triangle constraint.

[0018] The secondary structure network is a spatial point cloud network composed of landmark buildings in the urban space.

[0019] The three-level coverage network consists of the spatial location network of buildings in the first-level reference network and the second-level structural network extracted from the remote sensing image data based on the image feature extraction algorithm.

[0020] The four-level dynamic network is a local coordinate system network based on synchronous positioning and mapping technology, with each fixed sensor as an anchor point.

[0021] As a specific solution in this application, the step of uniformly calibrating the integrated air-ground big data to the urban standard coordinate system to obtain low spatiotemporal deviation data includes:

[0022] Map each spatial point cloud in the secondary structure network to the primary reference network to obtain the first fusion network;

[0023] Based on the first fusion network, multiple elevation columns are obtained; the elevation columns are used to represent the corresponding buildings in the city;

[0024] Based on each elevation column, obtain building projection information corresponding to each elevation column; the projection direction of the building projection is vertically downward; the building projection information includes at least the horizontal distance between each building projection.

[0025] Based on the three-level coverage network, the projection offset of each building projection is obtained; the projection offset is used to characterize at least the difference in horizontal distance between the building projection and the corresponding building in the three-level coverage network.

[0026] Based on the projection offset, adjust the horizontal position of each elevation column in the first fusion network to obtain the second fusion network;

[0027] Based on the four-level dynamic network, the motion trajectory information of targets in each local coordinate system network is obtained; the targets include pedestrians and vehicles.

[0028] The motion trajectory information is back-projected onto the second fusion network to obtain the low spatiotemporal deviation data.

[0029] As a specific solution in this application, the step of mapping each spatial point cloud in the secondary structure network to the primary reference network to obtain the first fusion network includes:

[0030] Based on each spatial point cloud, multiple outliers are obtained;

[0031] Each outlier is removed from each spatial point cloud to obtain multiple corrected spatial point clouds;

[0032] Based on the iterative nearest point algorithm, each reference point and each modified spatial point cloud form multiple matching point pairs; in the matching point pairs, each reference point is the source point set and each modified spatial point cloud is the target point set.

[0033] The first fusion network is obtained based on each matching point pair.

[0034] As a specific solution in the technical solution of this application, the step of obtaining multiple elevation columns based on the first fusion network includes:

[0035] The surface curvature of each cloud point in the first fusion network is obtained based on Delaunay triangulation technology;

[0036] Based on the surface curvature of each cloud point, multiple surface fitting results are obtained;

[0037] Based on the fitting results of each surface, multiple elevation columns are obtained.

[0038] As a specific solution in this application, the step of obtaining the projection offset of each building projection based on the three-level coverage network includes:

[0039] Based on the projections of each building, the projection of the target building is obtained; the target building projection is any building projection among the various building projections for which the projection offset has not been obtained.

[0040] Based on the target building projection, a fixed offset is obtained; the fixed offset is the offset of the target building projection in the three-level coverage network.

[0041] Based on the fixed offset, the projection offset of the target building projection is obtained.

[0042] As a specific solution in this application, the step of obtaining the first conflicting entity and the second conflicting entity based on the low spatiotemporal deviation data includes:

[0043] Based on the low spatiotemporal deviation data, a first entity and a second entity are obtained; the first entity is any entity in the low spatiotemporal deviation data; the second entity is any entity in the low spatiotemporal deviation data that spatially overlaps with the first entity; and the data sources of the first entity and the second entity are different.

[0044] The first entity and the second entity are sliced ​​according to a spatial grid and a time window, and the spatial overlap area and spatial overlap duration are obtained; the spatial overlap area is the area where the first entity and the second entity overlap; the spatial overlap duration is the duration during which the first entity and the second entity overlap.

[0045] Based on the spatial overlap area and the spatial overlap duration, a spatiotemporal overlap value is obtained; the spatiotemporal overlap value is at least used to characterize the size of the area and duration of the spatial overlap between the first entity and the second entity.

[0046] If the spatiotemporal overlap value meets a preset condition, then the first entity and the second entity are respectively designated as the first candidate entity and the second candidate entity; the preset condition includes that the spatiotemporal overlap value is less than a first preset value;

[0047] Based on the first candidate entity and the second candidate entity, obtain the first conflicting entity and the second conflicting entity.

[0048] As a specific solution in this application, the step of obtaining the first conflicting entity and the second conflicting entity based on the first candidate entity and the second candidate entity includes:

[0049] Based on different data sources in the low spatiotemporal bias data, multiple confidence probabilities are obtained; the confidence probability is the probability that the first candidate entity and the second candidate entity are two conflicting entities in the corresponding data source;

[0050] Based on the trust function theory, conflict probabilities and non-conflict probabilities are obtained from various confidence probabilities; the sum of the conflict probabilities and the non-conflict probabilities equals 100%.

[0051] Based on the conflict probability and the non-conflict probability, a probability difference is obtained; the probability difference is equal to the conflict probability minus the non-conflict probability.

[0052] If the probability difference is greater than the second preset value, then the first candidate entity is determined to be the first conflicting entity and the second candidate entity is determined to be the second conflicting entity.

[0053] As a specific solution in this application, the spatial positions of the first and second conflicting entities in the low spatiotemporal bias data are corrected to obtain a conflict-free big data model, including:

[0054] Using the position parameters of the first and second conflicting entities as particle dimensions, a particle swarm is randomly generated within the allowable adjustment range, with each particle representing a potential correction position.

[0055] The three-dimensional Euclidean distance between the corrected position of each particle and the positions of the first and second conflicting entities in each data source is calculated, and the distance error is obtained by weighting and averaging the distances together with the confidence levels obtained from each data source during the evidence fusion stage.

[0056] Obtain the nearest triangle edge or triangle face in the Delaunay triangulation for the corrected position of each particle, calculate the vertical distance from the corrected position to the triangle edge or triangle face, and if the corrected position belongs to a preset category of entity and the vertical distance exceeds the allowed offset threshold of the category of entity, then the vertical distance is used as the topology violation degree.

[0057] A fitness function is constructed based on the distance error and the topology violation; the fitness function is the sum of the distance error and the topology violation.

[0058] Based on the fitness function, the optimal position is output as the correction result;

[0059] Based on the correction results, the spatial positions of the first conflicting entity and the second conflicting entity in the low spatiotemporal bias data are corrected to obtain a conflict-free big data model.

[0060] Compared with the prior art, the beneficial effects of this application are:

[0061] This application addresses the issue of low fusion accuracy caused by inconsistent spatiotemporal benchmarks and significant differences in format and precision among multi-source data by integrating urban spatial point clouds, sensor networks, and remote sensing imagery. By eliminating spatiotemporal biases and entity conflicts, it improves data consistency and reliability, avoiding information gaps and semantic conflicts. Furthermore, the big data model built upon conflict-free data can accurately identify special events, providing support for scenarios such as urban rainstorm and flood warnings and emergency public event responses. This facilitates refined urban management, precise decision-making, and efficient resource allocation, meeting the complex measurement needs of cities. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of a big data fusion method for urban measurement proposed in an embodiment of this application. Detailed Implementation

[0063] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0064] The terms "first," "second," etc., in the specification and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. For example, the first conflicting entity and the second conflicting entity mentioned below belong to different entities. It should be understood that such names can be used interchangeably where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or modules is not necessarily limited to those steps or modules explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. The division of modules in the embodiments of this application is merely a logical division. In actual applications, there may be other division methods. For example, multiple modules may be combined into or integrated into another system, or some features may be ignored or not performed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be through some interface, and the indirect coupling or communication connection between modules may be electrical or other similar forms. None of these are limited in the embodiments of this application. Furthermore, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed among multiple circuit modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiments of this application.

[0065] To address the technical problem mentioned in the background art—the low accuracy of integrated air-ground big data fusion, which prevents precise identification of special urban events—this application proposes an integrated air-ground big data fusion method oriented towards urban measurement. This integrated air-ground big data fusion method for urban measurement includes steps S100 to S500.

[0066] Step S100: Obtain integrated air and ground big data of the city.

[0067] In this embodiment, the integrated air-ground big data includes at least the city's spatial point cloud data, the city's sensor network data, and the city's remote sensing image data.

[0068] In the embodiments of this application, no restrictions are placed on the source of the integrated air-ground big data. For example, spatial point cloud data can be collected by professional teams through airborne / ground-based laser scanning, or obtained from government platforms, commercial companies, and scientific research sharing channels; sensor network data can come from municipally deployed traffic and environmental sensors, enterprise-operated IoT devices, and information opened by public data platforms; remote sensing image data includes free satellite imagery (e.g., Landsat and Sentinel), commercial satellite data, as well as drone aerial photography and aerial photogrammetry results. This data can be free-to-download open-source information, or data read from databases, obtained through cooperation, or purchased.

[0069] To accurately acquire the three core types of integrated air-ground big data (i.e., urban spatial point cloud data, urban sensor network data, and urban remote sensing image data) through clear and operable technical means, ensuring the relevance and effectiveness of data sources, and providing high-quality raw data for subsequent fusion processing, in one embodiment of this application, step S100, acquiring integrated air-ground big data of the city, includes steps S110 to S130.

[0070] Step S110: Scan the urban area using a drone equipped with lidar to obtain spatial point cloud data of the city.

[0071] In this embodiment, a multi-rotor UAV or a fixed-wing UAV equipped with a lidar (e.g., LiDAR) can be selected. The flight path (e.g., grid-like or strip-like paths) is planned based on the terrain complexity and accuracy requirements of the urban area. The flight altitude (typically 50-200 meters, reduced for higher accuracy requirements) and scanning frequency (e.g., 100-500 points / m²) are set. During flight, the lidar emits laser pulses towards the ground, calculates distances by receiving reflected signals, and generates point cloud data containing three-dimensional coordinates (X, Y, Z), covering surface features such as buildings, roads, and vegetation. After flight, point cloud processing software (e.g., Cloud Compare) is used for data stitching and noise reduction.

[0072] It is important to understand that ground images contain rich visual information (e.g., building facade textures, road markings, vegetation colors, etc.), which can be combined with spatial point cloud data collected by UAV LiDAR to help identify the semantic attributes of various entities in the point cloud data (e.g., distinguishing between buildings, roads, vegetation, etc.), thus improving the interpretability of the point cloud data. Ground images can also provide texture information for 3D models constructed from point cloud data, making the models more closely resemble real-world scenes. Furthermore, for details that may be missed by point cloud data (e.g., small facilities, surface markings, etc.), ground images can serve as a supplementary data source, enhancing the ability of integrated air-ground big data to depict local urban details and providing more comprehensive foundational data for subsequent fusion steps such as urban standard coordinate system calibration and entity conflict detection. Based on this, in this embodiment, ground images of the city can also be acquired using a UAV equipped with a camera.

[0073] Specifically, in order to acquire ground images of the city, a high-resolution (e.g., 20 megapixels or more) RGB camera can be mounted on a drone to capture ground images with a preset overlap rate (e.g., forward overlap rate of 70% or more, lateral overlap rate of 60% or more), thereby acquiring detailed information such as building facades, road signs, pedestrians and vehicles, which can be used to assist in semantic recognition or texture mapping of point cloud data.

[0074] Step S120: Acquire remote sensing image data of the city based on low-altitude remote sensing satellites.

[0075] In this embodiment, a low-altitude remote sensing satellite with an orbital altitude below 1000 kilometers (e.g., Gaofen-7 or WorldView-3) can be selected. The image resolution (e.g., 0.5 meters to 5 meters, with fine urban measurements requiring less than 1 meter) and spectral band (visible light, near-infrared, etc., used to distinguish vegetation, buildings, and water bodies) are chosen according to requirements. A data subscription request is submitted through the satellite operator (e.g., the Resource Satellite Application Center or Digital Globe), specifying the latitude and longitude range of the urban area and the shooting time (avoiding cloud cover). After the satellite takes images as instructed, it generates remote sensing images containing geographic coordinates. After preprocessing such as radiometric and geometric corrections, these images are provided to the user. The user can further process the images using professional software (e.g., ENVI, ERDAS) to extract macroscopic features such as urban land use types and building density.

[0076] Step S130: Acquire sensor network data of the city based on fixed sensors deployed on major urban roads and public areas.

[0077] In this embodiment, different types of fixed sensors can be deployed in public areas such as major urban road intersections, bridges, squares, and parks, according to functional requirements. For example, radar sensors or video analytics devices can be installed to collect data such as traffic flow, vehicle speed, and congestion status in real time; PM2.5 sensors, noise sensors, and temperature and humidity sensors can be deployed to periodically upload monitoring data; and IoT sensors can be embedded in devices such as streetlights and trash cans to collect information such as energy consumption and usage status. The sensors can be connected to the city's data platform via wired or wireless networks, and the data is stored in real time with timestamps to form a structured database (e.g., MySQL) or a time-series database (e.g., InfluxDB) for subsequent fusion processing to reflect the dynamic operating status of the city.

[0078] Step S200: The integrated air-ground big data is uniformly calibrated to the urban standard coordinate system to obtain data with low spatiotemporal deviation.

[0079] In this embodiment, the urban standard coordinate system is used at least to eliminate spatiotemporal deviations between different source data in the integrated air-ground big data. Based on this, in this embodiment, the urban standard coordinate system includes a primary reference network, a secondary structure network, a tertiary coverage network, and a quaternary dynamic network.

[0080] In this embodiment, the primary reference network is a network composed of multiple reference points in the urban space. These reference points are either deeply buried municipal reference points or anchorage points of cross-river bridges. At least three reference points are established within each administrative division unit of the city to form a spatial triangle constraint. It is important to understand that spatial triangle constraints are a technical method in surveying, geographic information systems, or engineering surveying that utilizes the geometric properties of triangles to locate and control the stability of spatial points. Specifically, in three-dimensional space, three non-collinear points can form a triangle, which has a uniquely determined spatial location and geometric shape (e.g., side length, angle, and area). Utilizing this property, by accurately measuring the three reference points (i.e., the three-dimensional coordinates of each reference point), a stable "spatial reference frame," or spatial triangle constraint, can be formed.

[0081] In this embodiment, the secondary structure network is a spatial point cloud network composed of landmark buildings in the urban space.

[0082] In this embodiment, the three-level coverage network consists of the first-level reference network extracted from the remote sensing image data based on the image feature extraction algorithm and the spatial location network of buildings in the second-level structural network.

[0083] In this embodiment, the four-level dynamic network is a local coordinate system network based on simultaneous localization and mapping (SMR) technology, with each fixed sensor as an anchor point.

[0084] In this embodiment, step S200 aims to uniformly calibrate the integrated air-ground big data to the urban standard coordinate system, thereby obtaining data with low spatiotemporal bias. The core objective is to eliminate spatiotemporal bias caused by differences in acquisition platforms (e.g., drones, near-Earth satellites) from different sources, laying the foundation for subsequent data fusion, conflict resolution, and event identification. Based on this, step S200, which involves uniformly calibrating the integrated air-ground big data to the urban standard coordinate system to obtain data with low spatiotemporal bias, includes steps S210 to S270.

[0085] Step S210: Map each spatial point cloud in the secondary structure network to the primary reference network to obtain the first fusion network.

[0086] In this embodiment, the first fusion network is formed by mapping the spatial point clouds in the secondary structure network to the primary reference network. The core function of step S210 is to achieve spatial alignment between local high-precision data and the global reference framework, providing a unified spatial foundation for subsequent data fusion. Based on this, step S210 maps each spatial point cloud in the secondary structure network to the primary reference network to obtain the first fusion network, including steps S211 to S214.

[0087] Step S211: Based on each spatial point cloud, obtain multiple outlier points.

[0088] In this embodiment, outliers refer to points in each spatial point cloud that differ significantly from the distribution of the surrounding point clouds and do not conform to normal spatial characteristics. These outliers usually originate from unstable features (e.g., antennas on building rooftops, temporarily erected objects, etc.) or data acquisition errors, which can interfere with the accurate matching between the secondary structure network and the primary reference network. Therefore, they need to be deleted in step S212 to ensure the reliability of subsequent corrections to the spatial point cloud.

[0089] In this embodiment, outliers can be identified using any reasonable method. For example, when processing spatial point clouds, a circle with a radius of 10 meters (or 5 meters or 15 meters in other embodiments) can be used as the neighborhood, centered on any cloud point, and the average distance between each cloud point and other cloud points in its neighborhood can be calculated. If the average distance of a certain cloud point is greater than three times the average of the average distances of all cloud points, then the cloud point can be determined to be an outlier.

[0090] Step S212: Remove each outlier from each spatial point cloud to obtain multiple corrected spatial point clouds.

[0091] It is important to understand that removing certain cloud points (i.e., outliers) from the corresponding spatial point cloud is a mature technology, which will not be elaborated here.

[0092] Step S213: Based on the iterative nearest point algorithm, form multiple matching point pairs between each reference point and each corrected spatial point cloud.

[0093] In this embodiment, each reference point in the matching point pair is a source point set, and each corrected spatial point cloud is a target point set.

[0094] It is important to understand that the Iterative Closest Point (ICP) algorithm is a classic algorithm for 3D point cloud registration. Its core objective is to find the optimal rigid body transformation (i.e., rotation matrix and translation vector) between two point clouds from different viewpoints through iterative optimization, so that they coincide as much as possible after alignment. Based on this, in this embodiment, step S213: Based on the ICP algorithm, multiple matching point pairs are formed between each reference point and each corrected spatial point cloud, including steps S213a to S213c.

[0095] Step S213a: Based on each reference point and each corrected spatial point cloud, obtain multiple iterative matching point pairs.

[0096] In this embodiment, the iterative matching point pair is the point pair formed by the reference point and the cloud point that is closest to each corrected spatial point cloud.

[0097] Step S213b: Based on all iteratively matched point pairs, solve the objective function using the least squares method to obtain the rotation matrix and translation vector that minimizes the sum of squared distances between the source point set and the target point set.

[0098] In this embodiment, the objective function formula is:

[0099]

[0100] Where n represents the number of iteratively matched point pairs; This represents the coordinates of the points in the target point set that are matched in the i-th iteration. This represents the coordinates of the points in the source point set that are matched in the i-th iteration. Represents the rotation matrix; Represents the translation vector; This expresses the expression for the squared distance between two coordinate points; This represents the rotation matrix and translation vector that minimize the sum of the squared distances between the source point set and the target point set.

[0101] Step S213c: Transform the source point set according to the calculated rotation matrix and translation vector, find the nearest point again and update the iterative matching point pair. Repeat the calculation process of step S213b above until the change of the sum of squared distances between two iterations is less than the set threshold or the iteration reaches the maximum number of iterations. At this time, a stable matching point pair is obtained.

[0102] In the embodiments of this application, the above threshold can be set according to requirements, for example, the above threshold can be 0.1 or 0.2, etc.

[0103] Step S214: Obtain the first fusion network based on each matching point pair.

[0104] In this embodiment, through iterative optimization of the ICP algorithm, multiple precise matching point pairs are ultimately formed between the source point set (i.e., each reference point) and the target point set (i.e., each corrected spatial point cloud). This provides a spatial transformation basis for subsequently mapping the secondary structure network to the primary reference network to form the first fusion network. In other words, in this embodiment, aligning each reference point in the primary reference network with each matching point in the secondary structure network yields the first fusion network.

[0105] Step S220: Based on the first fusion network, obtain multiple elevation columns.

[0106] In this embodiment, the elevation pillars are used to represent the corresponding buildings in the city. It should be noted that the purpose of obtaining the elevation pillars is to provide a precise three-dimensional reference for correcting the projection distortion (i.e., projection offset) of the satellite imagery in step S240. In embodiments of this application, multiple elevation pillars can be obtained based on the first fusion network using any reasonable method. For example, step S220, based on the first fusion network, obtains multiple elevation pillars, including steps S221 to S223.

[0107] Step S221: Obtain the surface curvature of each cloud point in the first fusion network based on Delaunay triangulation technology.

[0108] It's important to understand that Delaunay Triangulation (DT) is a point-set-based geometric partitioning technique widely used in Geographic Information Systems (GIS), Computer Graphics, and Finite Element Analysis. Its core principle is to efficiently represent the topological relationships of a spatial point set using a network of triangles that satisfy specific geometric criteria. Obtaining the surface curvature of each cloud point in the first fused network using Delaunay Triangulation is a mature technology and will not be elaborated upon here.

[0109] Step S222: Based on the surface curvature of each cloud point, obtain multiple surface fitting results.

[0110] In this embodiment, after the triangular mesh is constructed, the minimum curvature surface fitting method is used to calculate the surface curvature at each feature point by weighted averaging of the curvatures of all triangles in the neighborhood of that point, ultimately obtaining the surface fitting result of the urban framework surface. This process ensures that the elevation and spatial position of the elevation columns are consistent with the actual building.

[0111] Step S223: Based on the fitting results of each surface, obtain multiple elevation columns.

[0112] In this embodiment, the center point coordinates of the elevation column correspond one-to-one with the geometric center coordinates of the bottom of the building, so that each elevation column uniquely represents the corresponding building in the city.

[0113] Step S230: Based on each elevation column, obtain the building projection information that corresponds one-to-one with each elevation column.

[0114] In this embodiment, the projection direction of the building projection is vertically downward. The building projection information includes at least the horizontal distance between the various building projections.

[0115] Step S240: Based on the three-level coverage network, obtain the projection offset of each building projection.

[0116] In this embodiment, the projection offset is used at least to characterize the difference in horizontal distance between the building projection and the corresponding building in the three-level coverage network. That is, in this embodiment, the purpose of step S240 is to obtain the difference in horizontal distance between the building projection and the corresponding building in the three-level coverage network, thereby providing crucial data support for subsequent adjustments to the building projection position and achieving more accurate data fusion. In the process of air-ground integrated big data fusion for urban measurement, this step plays a significant role in calibrating satellite imagery and ensuring spatial consistency of data from different data sources. Based on this, step S240, based on the three-level coverage network, obtains the projection offset of each building projection, including steps S241 to S243.

[0117] Step S241: Obtain the target building projection based on the projections of each building.

[0118] In this embodiment, the target building projection is any building projection among the various building projections for which no projection offset has been obtained. That is, in this embodiment, the method for obtaining the projection offset of any building projection among the various building projections is the same as the method for obtaining the projection offset of the target building projection.

[0119] Step S242: Obtain a fixed offset based on the target building projection.

[0120] In this embodiment, the fixed offset is the offset of the target building's projection within the three-level coverage network. Specifically, the formula for calculating the fixed offset is as follows:

[0121]

[0122] in, This represents the fixed offset corresponding to the target building projection; The elevation of the column corresponding to the projection of the target building; Indicates the satellite image angle; Represents the tangent function; Indicates the scale of satellite imagery.

[0123] Step S243: Based on the fixed offset, obtain the projection offset of the target building projection.

[0124] In this embodiment, step S243, obtaining the projection offset of the target building projection based on the fixed offset, includes steps S243a and S243b.

[0125] Step S243a: Obtain the projection offset direction.

[0126] In this embodiment, the projection offset direction is along the radial direction from the building outline point in the satellite image to the center point of the corresponding elevation column. This direction is determined by the central projection principle of satellite image imaging, ensuring that the offset compensation is consistent with the physical cause of the actual projection distortion.

[0127] Step S243b: Based on the fixed offset and the projection offset direction, obtain the projection offset of the target building projection. Specifically, the calculation formula for obtaining the projection offset of the target building projection based on the fixed offset and the projection offset direction is as follows:

[0128]

[0129] in, This represents the projection offset corresponding to the target building's projection; This represents the fixed offset corresponding to the target building projection; This represents the vector from the building outline point in the satellite image to the center point of the elevation column in three-dimensional space (hereinafter referred to as the projection vector). The magnitude of the projection vector; Indicates the zero-prevention coefficient. It can be any positive number close to zero, for example, It can be 0.01 or 0.001, etc.

[0130] Step S250: Based on the projection offset, adjust the horizontal position of each elevation column in the first fusion network to obtain the second fusion network.

[0131] In this embodiment, the second fusion network can be obtained by adjusting the horizontal position of each elevation column in the first fusion network according to the projection offset.

[0132] Step S260: Based on the four-level dynamic network, obtain the motion trajectory information of the target in each local coordinate system network.

[0133] In this embodiment, the targets include pedestrians and vehicles.

[0134] Step S270: Project the motion trajectory information backward onto the second fusion network to obtain the low spatiotemporal deviation data.

[0135] It is important to note that while low spatiotemporal bias data eliminates the spatiotemporal bias in integrated air-ground big data, it is still insufficient to support the semantic understanding of city models. This is because the same entity may have conflicting records in different data sources, requiring the elimination of entity conflicts to enhance the event interpretability of the model. Therefore, step S300 needs to be executed.

[0136] Step S300: Based on the low spatiotemporal deviation data, obtain the first conflict entity and the second conflict entity.

[0137] In this embodiment, the first conflicting entity is any entity in the low spatiotemporal skew data. The second conflicting entity is any entity in the low spatiotemporal skew data that spatially overlaps with the first conflicting entity. The first and second conflicting entities originate from different data sources.

[0138] It should be clear that, in the embodiments of this application, the first conflicting entity and the second conflicting entity can be obtained based on the low spatiotemporal deviation data in any reasonable manner. For example, any two spatially overlapping entities in the low spatiotemporal deviation data whose spatial overlap area or spatial overlap duration is less than a third preset value can be used as the first conflicting entity and the second conflicting entity. For example, if the determination of whether any two spatially overlapping entities are the first conflicting entity and the second conflicting entity is based on the spatial overlap area, the third preset value can be 0.3 times or 0.2 times the maximum spatial overlap area (that is, the maximum spatial overlap area that the two spatially overlapping entities can form), etc.; in this embodiment, if the determination of whether any two spatially overlapping entities are the first conflicting entity and the second conflicting entity is based on the spatial overlap duration, the third preset value can be 10 minutes or 20 minutes, etc.

[0139] In order to accurately obtain the first conflicting entity and the second conflicting entity, in one embodiment of this application, step S300, based on the low spatiotemporal deviation data, obtains the first conflicting entity and the second conflicting entity, including steps S310 to S350.

[0140] Step S310: Based on the low spatiotemporal deviation data, obtain the first entity and the second entity.

[0141] In this embodiment, the first entity is any entity in the low spatiotemporal skew data. The second entity is any entity in the low spatiotemporal skew data that spatially overlaps with the first entity, and the first entity and the second entity have different data sources.

[0142] Step S320: Slice the first entity and the second entity according to the spatial grid and time window, and obtain the spatial overlap area and spatial overlap duration.

[0143] In this embodiment, the spatial overlap area is the area where the first entity and the second entity overlap spatially. The spatial overlap duration is the duration during which the first entity and the second entity overlap spatially.

[0144] It is important to note that for the same entity within a city (e.g., a building, a road crack, or flooding in an area), records from different data sources (e.g., data acquired from UAV point clouds, satellite imagery, and ground sensors) should theoretically have a high degree of overlap in terms of spatiotemporal scope. This is because they describe the same objectively existing urban element. For example, a building scanned by a UAV and the same building captured in satellite imagery at the same time should have roughly matching spatial locations and times, resulting in a high degree of overlap. When different data sources show low overlap in records of what should theoretically be the same entity, it indicates a significant discrepancy in the descriptions of that entity across multiple data sources. This could be due to an identification error by one data source (e.g., satellite imagery misinterpreting building outlines), semantic bias (e.g., different data sources defining "road" differently), or different entities being incorrectly associated. This low spatiotemporal consistency violates the fundamental assumption that "multi-source data for the same entity should converge," and therefore can be preliminarily identified as an anomaly.

[0145] Step S330: Obtain the spatiotemporal overlap value based on the spatial overlap area and the spatial overlap duration.

[0146] In this embodiment, the spatiotemporal overlap value is used at least to characterize the area and duration of the spatial overlap between the first entity and the second entity.

[0147] In this embodiment, the spatiotemporal overlap value can be obtained based on the spatial overlap area and the spatial overlap duration using any reasonable method. For example, the spatiotemporal overlap value can be the sum or product of the spatial overlap area and the spatial overlap duration. In a specific embodiment of this application, step S330: the calculation formula for obtaining the spatiotemporal overlap value based on the spatial overlap area and the spatial overlap duration is as follows:

[0148]

[0149] in, This represents the spatiotemporal overlap value corresponding to the first entity and the second entity; This represents the average spatial overlap area of ​​the first and second entities in each spatial slice; This represents the maximum spatial overlap area of ​​the first and second entities in each spatial slice; Indicates the duration of spatial overlap between the first and second entities; This indicates the total duration of the low spatiotemporal deviation data. Generally, and There is no possibility that the value is zero. If there is an extreme case, a non-zero constant, such as 0.001, is added to the denominator of the fraction.

[0150] Step S340: If the spatiotemporal overlap value meets the preset conditions, then the first entity and the second entity are respectively regarded as the first candidate entity and the second candidate entity.

[0151] In this embodiment, a larger spatiotemporal overlap value between the first entity and the second entity indicates that the first entity and the second entity are more likely to be the same entity, and there is no anomaly between them; conversely, a smaller spatiotemporal overlap value indicates that the first entity and the second entity are more likely not to be the same entity, and there is an anomaly between them. That is, in this embodiment, the preset condition can be set according to requirements. For example, the preset condition can be that the spatiotemporal overlap value is less than a first preset value. In the embodiments of this application, the first preset value can be selected according to requirements; for example, the first preset value can be 0.3, 0.4, or 0.5, etc.

[0152] Step S350: Based on the first candidate entity and the second candidate entity, obtain the first conflicting entity and the second conflicting entity.

[0153] As discussed earlier, even if the first and second candidate entities meet the preset conditions, it does not necessarily mean that they conflict. Specifically, if either the first or second candidate entity is not a real entity but a virtual entity formed by data anomalies, then the first and second candidate entities do not necessarily conflict. For example, during drone LiDAR scanning, false point clouds may be generated due to fog, bird interference, or equipment vibration, which may be misidentified by the algorithm as small buildings or other entities. Satellite imagery may be extracted as abnormal areas (e.g., road cracks) due to cloud cover or sensor noise artifacts (e.g., light spots and stripes). These candidate entities are merely products of data noise and do not correspond to real city entities. During feature extraction or clustering, the algorithm may misclassify meaningless spatial point sets as entities due to unreasonable parameters. For example, randomly distributed ground point clouds may be clustered as road cracks, or meaningless pixel blocks in satellite imagery may be misclassified as small facilities. These candidate entities are the result of algorithmic processing errors and do not correspond to real entities. In other words, the purpose of step S350 is to accurately determine the real conflicting entity (i.e., the first conflicting entity and the second conflicting entity) from the candidate entities (i.e., the first candidate entity and the second candidate entity). Based on this, step S350, based on the first candidate entity and the second candidate entity, obtains the first conflicting entity and the second conflicting entity, including steps S351 to S354.

[0154] Step S351: Based on different data sources in the low spatiotemporal bias data, obtain multiple confidence probabilities.

[0155] In this embodiment, the confidence probability can be the probability that the first candidate entity is a real entity, and / or the probability that the second candidate entity is a real entity. Alternatively, the confidence probability can be the probability that the first candidate entity and the second candidate entity are two conflicting entities in their respective data sources. For any candidate entity, UAV point clouds can determine the probability of its existence based on point cloud density and geometric features; satellite imagery can determine the probability of its existence based on spectral features and texture analysis; and ground sensors can determine the probability of its existence by combining data from nearby devices for cross-validation.

[0156] Step S352: Based on the trust function theory, obtain the conflict probability and the non-conflict probability from each confidence probability.

[0157] In this embodiment, the sum of the conflict probability and the non-conflict probability is equal to 100%.

[0158] It's important to note that trust function theory quantifies the degree of support different pieces of evidence for a proposition through basic probability assignment, then synthesizes multi-source evidence using orthogonal sums and rules, and finally judges the credibility of the proposition based on the fusion result. Its core is dealing with "uncertainty" rather than "randomness," allowing for an unknown range of support levels for the proposition, making it suitable for conflict fusion scenarios involving multi-source heterogeneous data. For example, suppose a road section in a city may have a crack, and three data sources—drones, satellites, and ground sensors—provide observational information. Trust function theory is needed to determine whether the crack actually exists.

[0159] First, extract evidence from each data source:

[0160] Drone point cloud: Scanning revealed a point cloud density of 20 points / m² (>15 points / m²). Based on the rules, the probability of "the existence of cracks" is 0.8 (i.e., as stated below). (The remaining 0.2 is uncertain).

[0161] Satellite imagery analysis revealed anomalies in reflectance at specific wavelengths (consistent with crack characteristics), leading to a confidence level of 0.7 for the presence of cracks (i.e., as stated below). (The remaining 0.3 is uncertain).

[0162] Ground sensors: Only a single sensor detected weak vibrations, resulting in insufficient cross-validation. The probability of determining "the existence of a crack" is 0.6 (i.e., as mentioned below). (The remaining 0.4 is uncertain).

[0163] At this point, the evidence set from the three data sources is as follows: , , .

[0164] Then, the evidence is fused using the trust function theory:

[0165] The first step is to integrate drone and satellite evidence:

[0166] Overlap coefficient Both support the existence of a crack, their intersection is significant, and the calculation yields... =0.8×0.7+0.8×0.3+0.2×0.7=0.56+0.24+0.14=0.94.

[0167] Conflict coefficient There is no contradiction between the two (neither denies the existence of "cracks"), =0.

[0168] Fusion probability : = / (1- =0.94 / 1=0.94.

[0169] Step 2: Integrate the above results with evidence from ground sensors:

[0170] Overlap coefficient : =0.94×0.6+0.94×0.4+0.06×0.6=0.98.

[0171] Conflict coefficient There is still no conflict. =0.

[0172] Final fusion probability : = / (1- =0.98.

[0173] In this embodiment, the probability of a road having cracks is 0.98 (equivalent to the conflict probability mentioned above), and the probability of a road not having cracks is 0.02 (equivalent to the non-conflict probability mentioned above).

[0174] Step S353: Obtain the probability difference based on the conflict probability and the non-conflict probability.

[0175] In this embodiment, the probability difference is equal to the conflict probability minus the non-conflict probability. In the computer field, obtaining the difference between two values ​​(i.e., the conflict probability and the non-conflict probability) is a mature technique, and will not be elaborated upon here.

[0176] Step S354: If the probability difference is greater than the second preset value, then the first candidate entity is determined to be the first conflicting entity and the second candidate entity is determined to be the second conflicting entity.

[0177] It should be noted that in this embodiment, a larger probability difference indicates that the first candidate entity and the second candidate entity are more likely to be conflicting entities; conversely, a smaller probability difference indicates that the first candidate entity and the second candidate entity are more likely to be non-conflicting entities. In this embodiment, the second preset value can be set according to requirements; for example, the second preset value can be 0.2 or 0.3, etc.

[0178] Step S400: Correct the spatial positions of the first conflicting entity and the second conflicting entity in the low spatiotemporal deviation data to obtain a conflict-free big data model.

[0179] In this embodiment, if the first conflicting entity and the second conflicting entity are the same entity, then the first conflicting entity and the second conflicting entity need to be brought closer to each other to increase the spatiotemporal overlap value of the first conflicting entity and the second conflicting entity; if the first conflicting entity and the second conflicting entity are not the same entity, then the first conflicting entity and the second conflicting entity need to be moved away from each other to avoid the first conflicting entity and the second conflicting entity overlapping each other.

[0180] In a specific embodiment of this application, step S400 involves correcting the spatial positions of the first conflicting entity and the second conflicting entity in the low spatiotemporal deviation data to obtain a conflict-free big data model, including steps S410 to S460.

[0181] Step S410: Using the position parameters of the first conflicting entity and the second conflicting entity as the particle dimension, a particle swarm is randomly generated within the allowable adjustment range, with each particle representing a potential correction position.

[0182] In this embodiment, the corrected position is generated as follows:

[0183] First, determine the particle dimension: use the spatial position parameters (e.g., three-dimensional coordinates (x, y, z)) of the first and second conflicting entities as the particle dimension, i.e., the dimension of each particle is 3.

[0184] Then, set the adjustment range: based on the location distribution of the conflicting entities (i.e., the first conflicting entity or the second conflicting entity) in the multi-source data, determine the allowable adjustment range for each dimension (e.g., [ , ]、[ , ]、[ , ],in, This represents the minimum range of values ​​for the X-axis in three-dimensional space. The maximum range of values ​​for the X-axis in three-dimensional space; Represents the minimum range of values ​​for the Y-axis in three-dimensional space; The maximum range of values ​​for the Y-axis in three-dimensional space; Represents the minimum range of values ​​for the Z-axis in three-dimensional space; The maximum range of the Z-axis in three-dimensional space. The above range is 3 times the standard deviation of the entity's position in each data source (in other embodiments, it may be 2 or 4 times).

[0185] Finally, generate a particle swarm: randomly generate a number of particles within the above adjustment range (e.g., 40 or 50, etc.), with each particle representing a potential correction position.

[0186] Step S420: Calculate the three-dimensional Euclidean distance between the corrected position of each particle and the positions of the first conflicting entity and the second conflicting entity in each data source, and then calculate the weighted average based on the credibility obtained from each data source during the evidence fusion stage to obtain the distance error.

[0187] In this embodiment, the method for obtaining the distance error is as follows:

[0188] First, calculate the three-dimensional Euclidean distance: for the corrected position corresponding to each particle, calculate the three-dimensional Euclidean distance between it and the original positions of the first and second conflicting entities in each data source (e.g., UAV point cloud, satellite imagery, ground sensor network data, etc.).

[0189] Finally, a weighted average is calculated: the confidence level of each data source obtained in the evidence fusion stage of step S300 is obtained (e.g., the confidence level of UAV point cloud is 0.8, the confidence level of satellite imagery is 0.7, and the confidence level of ground sensor is 0.6), and the distance error is calculated using the formula:

[0190]

[0191] in, Indicates distance error; Indicates the number of data sources; This represents the credibility obtained from the evidence fusion stage corresponding to the j-th data source; This represents the three-dimensional Euclidean distance between the original positions of the first and second conflicting entities in the j-th data source.

[0192] Step S430: Obtain the nearest triangle edge or triangle face in the Delaunay triangulation for the corrected position of each particle, calculate the vertical distance from the corrected position to the triangle edge or triangle face, and if the corrected position belongs to a preset category entity and the vertical distance exceeds the allowed offset threshold of the category entity, then the vertical distance is used as the topology violation degree.

[0193] In this embodiment, the allowable offset threshold can be preset as needed, for example, the allowable offset threshold can be 0.5 meters or 0.6 meters, etc.

[0194] Step S440: Construct a fitness function based on the distance error and the topology violation.

[0195] In this embodiment, the fitness function is the sum of the distance error and the topology violation.

[0196] Step S450: Based on the fitness function, output the optimal position as the correction result.

[0197] In this embodiment, the correction position that minimizes the fitness function is found through iterative search using the Particle Swarm Optimization (PSO) algorithm. It's important to understand that PSO is a swarm intelligence-based optimization algorithm that searches for the optimal solution in the solution space by simulating the cooperative behavior of a flock of birds foraging. It is a mature technology and will not be elaborated upon here.

[0198] Step S460: Based on the correction result, correct the spatial positions of the first conflicting entity and the second conflicting entity in the low spatiotemporal deviation data to obtain a conflict-free big data model.

[0199] In this embodiment, the original positions of the first conflicting entity and the second conflicting entity are updated to the optimal corrected positions output in step S450, thereby eliminating the conflict caused by the spatial position deviation between the two.

[0200] Step S500: Based on the conflict-free big data model, construct a long short-term memory network.

[0201] In this embodiment, the Long Short-Term Memory (LSTM) network is used to identify special events in the city. LSTM is a recurrent neural network suitable for processing time-series data, effectively capturing long-term dependencies in the data. The fused and corrected data is divided into multiple sequence samples according to time windows and input into the LSTM model for training. During training, the model learns the characteristic patterns of data under normal urban conditions. When the input sample data deviates significantly from the normal pattern, it is determined that a special urban event may occur. For example, in a rainstorm and flood warning scenario, when the fused features such as monitored rainfall data, drainage system flow data, and surface water level data show abnormal changes, the LSTM model identifies a possible flooding event and labels the relevant data. The labeling information includes key information such as the time, location, and type of the event.

[0202] The proposed air-ground integrated big data fusion method for urban measurement addresses the low fusion accuracy issues caused by inconsistent spatiotemporal benchmarks and significant differences in format and precision among multi-source data by integrating urban spatial point clouds, sensor networks, and remote sensing imagery. It improves data consistency and reliability by eliminating spatiotemporal biases and entity conflicts, avoiding information gaps and semantic conflicts. Furthermore, the big data model built based on conflict-free data can accurately identify special events, supporting scenarios such as urban rainstorm and flood warnings and emergency public event responses. This facilitates refined urban management, precise decision-making, and efficient resource allocation, meeting the complex needs of urban measurement.

[0203] It is important to understand that the computer-readable storage medium in this application includes both permanent and non-permanent, removable and non-removable media, and information storage can be achieved by any method or technology. Information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to: phase-change memory, static random access memory, dynamic random access memory, other types of random access memory, read-only memory, electrically erasable programmable read-only memory, flash memory or other memory technologies, optical disc read-only memory, digital versatile optical disc or other optical storage, magnetic tape, disk storage or other magnetic storage devices, or any other non-transfer medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include temporary computer-readable media, such as modulated data signals and carrier waves.

[0204] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0205] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the methods, apparatuses, and devices described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0206] In the embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or modules may be electrical, mechanical, or other forms.

[0207] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; that is, they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0208] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0209] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0210] The computer program product includes one or more computer instructions. When the computer program is loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or data center that integrates one or more available media. The available medium may be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., digital video optical disc), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0211] Although embodiments of this application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles of this application.

Claims

1. A method for air-ground integrated big data fusion for urban measurement, characterized in that, include: Acquire integrated air-ground big data of the city; the integrated air-ground big data includes at least the city's spatial point cloud data, the city's sensor network data, and the city's remote sensing image data. The integrated air-ground big data is uniformly calibrated to the urban standard coordinate system, which includes a first-level reference network, a second-level structural network, a third-level coverage network, and a fourth-level dynamic network. The primary benchmark network is a network composed of multiple benchmark points in the urban space; the benchmark points are deep-buried municipal benchmark points or anchor points of cross-river bridges; at least three benchmark points are set up in each administrative division unit in the city to form a spatial triangle constraint. The secondary structure network is a spatial point cloud network composed of landmark buildings in the urban space. The three-level coverage network consists of the spatial location network of buildings in the first-level reference network and the second-level structural network extracted from the remote sensing image data based on the image feature extraction algorithm; The four-level dynamic network is a local coordinate system network based on synchronous positioning and mapping technology, with each fixed sensor as an anchor point. Map each spatial point cloud in the secondary structure network to the primary reference network to obtain the first fusion network; Based on the first fusion network, multiple elevation columns are obtained; the elevation columns are used to represent the corresponding buildings in the city; Based on each elevation column, obtain the building projection information that corresponds one-to-one with each elevation column; The projection direction of the building projection is vertically downward; The building projection information includes at least the horizontal distance between the various building projections; Based on the three-level coverage network, the projection offset of each building projection is obtained; the projection offset is used to characterize at least the difference in horizontal distance between the building projection and the corresponding building in the three-level coverage network. Based on the projection offset, adjust the horizontal position of each elevation column in the first fusion network to obtain the second fusion network; Based on the four-level dynamic network, the motion trajectory information of targets in each local coordinate system network is obtained; the targets include pedestrians and vehicles. The motion trajectory information is back-projected onto the second fusion network to obtain low spatiotemporal bias data; the urban standard coordinate system is at least used to eliminate spatiotemporal bias of different source data in the integrated air-ground big data. Based on the low spatiotemporal deviation data, a first conflicting entity and a second conflicting entity are obtained; the first conflicting entity is any entity in the low spatiotemporal deviation data; the second conflicting entity is any entity in the low spatiotemporal deviation data that spatially overlaps with the first conflicting entity; the data sources of the first conflicting entity and the second conflicting entity are different. The spatial positions of the first and second conflicting entities in the low spatiotemporal bias data are corrected to obtain a conflict-free big data model. Based on the aforementioned conflict-free big data model, a long short-term memory network is constructed. The Long Short-Term Memory (LSTM) network is used to identify special events in the city.

2. The air-ground integrated big data fusion method for urban measurement according to claim 1, characterized in that, The acquisition of integrated air-ground big data of the city includes: The city area is scanned by a drone equipped with LiDAR to obtain spatial point cloud data of the city. Remote sensing image data of the city were acquired using low-altitude remote sensing satellites. Sensor network data of the city is acquired based on fixed sensors deployed on major urban roads and in public areas.

3. The air-ground integrated big data fusion method for urban measurement according to claim 1, characterized in that, The step of mapping each spatial point cloud in the secondary structure network to the primary reference network to obtain the first fusion network includes: Based on each spatial point cloud, multiple outliers are obtained; Each outlier is removed from each spatial point cloud to obtain multiple corrected spatial point clouds; Based on the iterative nearest point algorithm, each reference point and each modified spatial point cloud form multiple matching point pairs; in the matching point pairs, each reference point is the source point set and each modified spatial point cloud is the target point set. The first fusion network is obtained based on each matching point pair.

4. The air-ground integrated big data fusion method for urban measurement according to claim 1, characterized in that, The process of obtaining multiple elevation columns based on the first fusion network includes: The surface curvature of each cloud point in the first fusion network is obtained based on Delaunay triangulation technology; Based on the surface curvature of each cloud point, multiple surface fitting results are obtained; Based on the fitting results of each surface, multiple elevation columns are obtained.

5. The air-ground integrated big data fusion method for urban measurement according to claim 1, characterized in that, The process of obtaining the projection offset of each building projection based on the three-level coverage network includes: Based on the projections of each building, the projection of the target building is obtained; the target building projection is any building projection among the various building projections for which the projection offset has not been obtained. Based on the target building projection, a fixed offset is obtained; the fixed offset is the offset of the target building projection in the three-level coverage network. Based on the fixed offset, the projection offset of the target building projection is obtained.

6. The air-ground integrated big data fusion method for urban measurement according to any one of claims 1 to 5, characterized in that, The step of obtaining the first conflicting entity and the second conflicting entity based on the low spatiotemporal deviation data includes: Based on the low spatiotemporal deviation data, a first entity and a second entity are obtained; the first entity is any entity in the low spatiotemporal deviation data; the second entity is any entity in the low spatiotemporal deviation data that spatially overlaps with the first entity; and the data sources of the first entity and the second entity are different. The first entity and the second entity are sliced ​​according to a spatial grid and a time window, and the spatial overlap area and spatial overlap duration are obtained; the spatial overlap area is the area where the first entity and the second entity overlap spatially; the spatial overlap duration is the duration during which the first entity and the second entity overlap spatially. Based on the spatial overlap area and the spatial overlap duration, a spatiotemporal overlap value is obtained; the spatiotemporal overlap value is at least used to characterize the size of the area and duration of the spatial overlap between the first entity and the second entity. If the spatiotemporal overlap value meets a preset condition, then the first entity and the second entity are respectively designated as the first candidate entity and the second candidate entity; the preset condition includes that the spatiotemporal overlap value is less than a first preset value; Based on the first candidate entity and the second candidate entity, obtain the first conflicting entity and the second conflicting entity.

7. The air-ground integrated big data fusion method for urban measurement according to claim 6, characterized in that, The step of obtaining the first conflicting entity and the second conflicting entity based on the first candidate entity and the second candidate entity includes: Based on different data sources in the low spatiotemporal bias data, multiple confidence probabilities are obtained; the confidence probability is the probability that the first candidate entity and the second candidate entity are two conflicting entities in the corresponding data source; Based on the trust function theory, conflict probabilities and non-conflict probabilities are obtained from various confidence probabilities; the sum of the conflict probabilities and the non-conflict probabilities equals 100%. Based on the conflict probability and the non-conflict probability, a probability difference is obtained; the probability difference is equal to the conflict probability minus the non-conflict probability. If the probability difference is greater than the second preset value, then the first candidate entity is determined to be the first conflicting entity and the second candidate entity is determined to be the second conflicting entity.

8. The air-ground integrated big data fusion method for urban measurement according to claim 7, characterized in that, Correcting the spatial positions of the first and second conflicting entities in the low spatiotemporal bias data to obtain a conflict-free big data model includes: Using the position parameters of the first and second conflicting entities as particle dimensions, a particle swarm is randomly generated within the allowable adjustment range, with each particle representing a potential correction position. The three-dimensional Euclidean distance between the corrected position of each particle and the positions of the first and second conflicting entities in each data source is calculated, and the distance error is obtained by weighting and averaging the distances together with the confidence levels obtained from each data source during the evidence fusion stage. Obtain the nearest triangle edge or triangle face in the Delaunay triangulation for the corrected position of each particle, calculate the vertical distance from the corrected position to the triangle edge or triangle face, and if the corrected position belongs to a preset category of entity and the vertical distance exceeds the allowed offset threshold of the category of entity, then the vertical distance is used as the topology violation degree. A fitness function is constructed based on the distance error and the topology violation; the fitness function is the sum of the distance error and the topology violation. Based on the fitness function, the optimal position is output as the correction result; Based on the correction results, the spatial positions of the first conflicting entity and the second conflicting entity in the low spatiotemporal bias data are corrected to obtain a conflict-free big data model.

Images