A method for integrated space-ground-air mapping based on oblique photography 3D mapping

Abstract

The present application discloses an integrated air-space-ground mapping method based on oblique photography three-dimensional mapping, which belongs to the field of geographic mapping technology. The method includes: step 1, formulating a comprehensive oblique photography mapping plan to ensure complete coverage of complex terrain; step 2, deploying industrial-grade RTK drones and consumer-grade drones to work together to achieve a combination of large-scale coverage and local fine collection, and obtain high-quality oblique photography image data; step 3, integrating RTK positioning data and ground control point data as control information, and performing aerial triangulation encryption processing on oblique photography images; step 4, constructing a high-quality, high-precision multi-view three-dimensional model; step 5, based on the multi-view three-dimensional model, automating the extraction and identification of ground object features, as well as automatic calculation of terrain element parameters; step 6, integrating the multi-view three-dimensional model, ground object features and terrain element parameters with the natural resource management system to generate integrated air-space-ground mapping.

Description

Technical Field

[0001] The present application relates to the field of geographic surveying and mapping technology, and more specifically, to an integrated air-space-ground-land mapping method based on oblique photography three-dimensional mapping. Background Art

[0002] With the continuous development of surveying and mapping technology, traditional mapping methods can no longer meet the needs of complex terrain and diversified application scenarios. In recent years, oblique photography 3D mapping technology has been widely used. It uses multi-angle aerial images to generate high-precision 3D models, providing strong technical support for urban planning, land surveying and mapping, disaster monitoring and other fields. However, existing oblique photography 3D mapping methods mainly rely on aerial platforms such as drones or manned aircraft. Although these methods perform well in large-area mapping, they still face many challenges in areas with high precision requirements, complex terrain or where images cannot be obtained by aerial means.

[0003] In order to overcome these limitations, the integrated space-air-ground-integrated mapping method has gradually become a research hotspot. This method can not only improve the accuracy of mapping, but also expand the scope of application of mapping by integrating multi-source data such as satellite remote sensing, aerial photography, and ground measurement. However, the current integrated space-air-ground-integrated mapping technology still has problems such as difficulty in data fusion, incomplete information acquisition, and low processing efficiency. In particular, in the effective combination of oblique photography and other mapping methods, there is a lack of unified technical standards and efficient algorithm support, resulting in low comprehensive utilization efficiency of multi-source data.

[0004] Based on this, this application proposes an integrated air-space-ground mapping method based on oblique photography three-dimensional mapping. By comprehensively utilizing oblique photography, satellite images, ground radar scanning and other surveying and mapping data sources, a high-precision three-dimensional model is constructed to improve the accuracy and efficiency of mapping, thereby meeting diverse application needs. At the same time, by optimizing data fusion and processing procedures, the collaborative work efficiency of multi-source data is significantly improved, solving the shortcomings of the existing technology. Summary of the invention

[0005] In order to overcome a series of defects in the prior art, the purpose of the present application is to provide an integrated air-space-ground mapping method based on oblique photography three-dimensional mapping for the above-mentioned problems, including the following steps.

[0006] Step 1: Develop a comprehensive oblique photogrammetry program to ensure complete coverage of complex terrain.

[0007] Step 2: Deploy industrial-grade RTK drones and consumer-grade drones to work together to achieve a combination of large-scale coverage and local fine collection to obtain high-quality oblique photography image data.

[0008] Step 3: Integrate RTK positioning data and ground control point data as control information to perform aerial triangulation encryption processing on the oblique photography images.

[0009] Step 4: Build a high-quality, high-precision multi-view 3D model.

[0010] Step 5: Automatically extract and identify terrain features and automatically calculate terrain element parameters based on the multi-view three-dimensional model.

[0011] Step 6: Integrate the multi-view 3D model, land feature and terrain element parameters with the natural resource management system to generate integrated air-space-ground mapping.

[0012] Furthermore, step 1 includes the following steps.

[0013] Collect existing geographic information data of the survey area and conduct preliminary analysis to understand the terrain characteristics, distribution of land objects and their complexity in the survey area.

[0014] GIS software was used to conduct a detailed analysis of terrain features, calculate and visualize slope, aspect and elevation change factors, and identify flat areas, steep areas and complex terrain areas within the survey area.

[0015] According to the results of terrain feature analysis, the survey area is divided into several blocks with similar features.

[0016] For each block, flight routes with different altitudes and overlaps are designed according to its terrain features to ensure the integrity and accuracy of data collection.

[0017] Determine the number and distribution of ground control points based on the survey area and the complexity of the terrain to improve surveying and mapping accuracy.

[0018] According to the characteristics of each block and the mapping requirements, the camera parameters and shooting parameters are determined to obtain high-quality oblique photography data.

[0019] The flight route, control point layout and shooting parameters are evaluated through simulation software, and fine-tuning and optimization are carried out based on the evaluation results to ensure the scientific nature and operability of the final surveying and mapping plan.

[0020] Furthermore, step 2 includes the following steps.

[0021] Select industrial-grade RTK drones and consumer-grade drones based on the characteristics of the survey area and mission requirements, and ensure that the performance parameters of the selected equipment meet the mission requirements.

[0022] Based on block division and flight route design, plan the operating range and flight routes of industrial-grade RTK drones and consumer-grade drones.

[0023] An RTK base station is set up in the survey area to ensure good communication with the drone; at the same time, a mobile station of an industrial-grade RTK drone is configured to obtain high-precision location information.

[0024] According to the predetermined flight route and shooting parameters, the selected industrial-grade RTK UAV is used to collect oblique photography data; at the same time, the flight status, image acquisition and RTK signal quality are monitored in real time to ensure the integrity and accuracy of data collection.

[0025] Based on preliminary results from industrial-grade RTK drone acquisition and on-site observations, identify areas that require detailed acquisition.

[0026] Use consumer-grade drones to conduct supplementary collection of areas that require detailed collection, including close-up photography of complex terrain, multi-angle collection of special objects, or additional coverage of shadow areas; during the flight, the acquired image data is checked in real time for quality, including image clarity, overlap, and color uniformity. If problems are found, adjust the flight parameters or re-collect in a timely manner.

[0027] Perform cross-validation flights at the interface of data collected by industrial RTK drones and consumer drones to detect and eliminate data gaps or inconsistent areas.

[0028] Furthermore, step 3 includes the following steps.

[0029] Step 3.1: Extract the precise RTK positioning information of each image from the flight data collected by the industrial RTK UAV, including the longitude, latitude, elevation, and attitude angle data at the time of shooting.

[0030] Step 3.2, integrate the RTK positioning data and ground control point data into the same coordinate system to establish a unified control point database.

[0031] Step 3.3, pre-process the collected oblique photography images, including image evenness, noise removal and blur operations to improve image quality; at the same time, extract the EXIF ​​information of each image, match it with the RTK positioning data, and establish a preliminary correspondence between the image and the position.

[0032] In step 3.4, computer vision algorithms are used to automatically identify and match feature points between adjacent images, generate a large number of connection points, and construct an image network for the entire survey area.

[0033] In step 3.5, the RTK positioning data, ground control point data and the image network of the entire survey area are used as input to perform aerial triangulation calculations using professional photogrammetry software.

[0034] In step 3.6, a comprehensive error analysis is performed on the aerial triangulation results, including control point residuals, tie point residuals, and image network adjustment accuracy.

[0035] Step 3.7, based on the error analysis results, remove gross error points, adjust weights, and perform multiple iterative calculations until the predetermined accuracy requirements are met.

[0036] Step 3.8, based on the optimized aerial triangulation results, update the precise position and attitude information of each image to obtain the optimized image data.

[0037] Step 3.9, use the optimized image data and dense matching algorithm to generate high-density three-dimensional point cloud data.

[0038] Furthermore, a computer vision algorithm is used to automatically identify and match feature points between adjacent images, generate a large number of connection points and construct an image network of the entire survey area, including the following steps.

[0039] Computer vision algorithms are applied to each image to identify and extract significant and stable feature points.

[0040] For each extracted feature point, its feature descriptor is calculated to describe the unique properties of the area around the feature point.

[0041] The nearest neighbor search algorithm is used to perform preliminary matching of feature points between adjacent images.

[0042] For the initially matched feature point pairs, a random sample consistency algorithm is applied to perform geometric verification to effectively remove false matches and only retain matching point pairs that meet the expected geometric relationship.

[0043] Based on the verified matching points, an image connection network of the entire survey area is constructed, and the relative orientation between each pair of adjacent images is calculated to provide initial values ​​for subsequent aerial triangulation.

[0044] A global optimization of the entire image network is performed, including joint optimization of all tie points and camera parameters using a bundle adjustment algorithm to minimize the reprojection error to further improve the accuracy and consistency of tie points.

[0045] Furthermore, a computer vision algorithm is applied to each image to identify and extract feature points with significance and stability, including the following steps.

[0046] The gradient of each image is calculated to obtain the gradient amplitude and direction. The specific calculation formula is: M(x,y)=√{(I x (x,y) 2 +(I y (x,y) 2},θ(x,y)=arctan(I y (x,y) / I x (x,y)), where I x (x,y) and I y (x,y) represents the partial derivatives of the image in the x and y directions respectively; θ(x,y) represents the gradient direction at the image position (x,y); M(x,y) represents the gradient amplitude at the image position (x,y).

[0047] Images of different scales are generated by Gaussian blur technology. The specific formula is: L(x,y,σ)=G(x,y,σ)*I(x,y), where I(x,y) represents the pixel value at the image position (x,y); L(x,y,σ) represents the Gaussian blurred image at scale σ; G(x,y,σ) is the Gaussian kernel function: G(x,y,σ)=1 / (2πσ 2 )·exp(-(x 2 +y 2 ) / (2σ 2 )), where * represents the convolution operation.

[0048] The differential Gaussian image is calculated between images of different scales. The specific formula is: D(x,y,σ)=L(x,y,kσ)-L(x,y,σ), where D(x,y,σ) represents the differential Gaussian image at scale σ; L(x,y,kσ) represents the Gaussian blurred image at scale kσ; k is a multiple of the scale and is used to generate adjacent scale images; L(x,y,σ) represents the Gaussian blurred image at different scales.

[0049] In the differential Gaussian images of different scales, local extreme points, i.e. feature points, are detected.

[0050] Optimize the sub-pixel accuracy of the detected feature points to improve the stability of the feature points. The formula is: x^=-H -1 ▽D, where H is the Hessian matrix, ▽D is the gradient of the differential Gaussian image; x^ represents the optimized feature point position.

[0051] Furthermore, for each extracted feature point, its feature descriptor is calculated, including the following steps.

[0052] For each feature point, a fixed-size neighborhood window is selected.

[0053] Calculate the gradient magnitude and gradient direction of each pixel in the neighborhood window.

[0054] The neighborhood is divided into sub-regions. In each sub-region, the gradient direction histogram is calculated and the gradient direction is quantized into n direction buckets. The value of each bucket is the accumulation of the gradient amplitude in that direction. The specific formula is: i,j =∑ k=1 n Bin i,j,k , where H i,j is the total gradient amplitude of the k-th direction bucket in the (i, j)-th sub-region; Bin i,j,k It is the cumulative value of the gradient magnitude of all pixels in the sub-region projected into the k-th direction bucket.

[0055] The histograms of all sub-regions are concatenated into a feature descriptor.

[0056] Normalize the feature descriptor to improve robustness to lighting changes: f i ′=f i / ‖f i ‖, where ‖f i ‖ is the feature descriptor f i The Euclidean norm of i ′ is the normalized feature descriptor vector.

[0057] Furthermore, the relative orientation between each pair of adjacent images is calculated through the following steps.

[0058] From the verified feature point matches, collect all matching point pairs between adjacent images.

[0059] By matching point pairs, estimate the fundamental matrix F AB , the specific formula is: F AB =argmin F ∑ g=1 N ρ(x A,g T Fx B,g ), where x A,g and x B,g is the g-th pair of matching points; ρ is the reprojection error; F AB =argmin F It means to find a basic matrix F so that the subsequent error function is minimized; ∑ g=1 N ρ(x A,g T Fx B,g ) is the cost function, which represents the sum of the reprojection errors for all matching point pairs; ρ(·) is the squared error metric.

[0060] Using the known camera intrinsic parameter matrix K A and K B , the fundamental matrix FAB Converted to the essential matrix, the formula is: E AB :E AB =K B T F AB K A , where K B T Represents the transpose of the camera intrinsic parameter matrix of image B.

[0061] The essential matrix E AB Decomposed into the rotation matrix R AB and the translation vector t AB , the formula is: E AB =R AB [t AB ] × , where [t AB ] × is the translation vector t AB The antisymmetric matrix of .

[0062] Get the relative position, specifically expressed as: T AB =(R AB ,t AB ).

[0063] Furthermore, the RTK positioning data, ground control point data and the image network of the entire survey area are used as inputs to perform aerial triangulation calculations using professional photogrammetry software, including the following steps.

[0064] Import RTK positioning data, ground control point data and the image network of the entire survey area into professional photogrammetry software.

[0065] Based on the RTK positioning data and image EXIF ​​information, the initial exterior orientation elements are calculated for each image, including position (X, Y, Z) and attitude angles (ω, φ, μ).

[0066] The automatic aerial triangulation program is started, and the global adjustment operation is performed using the tie point information and initial exterior orientation elements in the image network to minimize the reprojection error of the entire image network through iterative optimization.

[0067] The software automatically identifies and precisely locates known ground control points on the corresponding images and matches them to the images.

[0068] The software calculates the residuals of each connection point and control point, and automatically identifies and processes possible gross error points based on preset thresholds.

[0069] Establish a dynamic weight allocation strategy based on the accuracy characteristics of different data sources.

[0070] Perform the final adjustment calculation, integrate all RTK positioning data, ground control point data and constraints, and solve the optimal exterior orientation elements and ground point coordinates through an iterative algorithm; at the same time, generate a comprehensive accuracy assessment report including control point residuals, tie point residuals and image point residuals.

[0071] Systematically analyze the adjustment results and accuracy reports, verify whether various indicators meet the technical specifications, make optimization adjustments when necessary, and finally output high-quality exterior orientation elements and ground point coordinate data.

[0072] Furthermore, step 4 includes the following steps.

[0073] Perform comprehensive quality checks on high-density 3D point cloud data and optimized image data, including point cloud density, distribution uniformity, and image clarity.

[0074] Noise filtering is performed on high-density 3D point cloud data to remove abnormal points and low-quality points.

[0075] Based on geometric features and spectral information, high-density 3D point cloud data are preliminarily classified into ground points, building points and vegetation points.

[0076] A high-precision digital terrain model is constructed using ground points, and Kriging is used to achieve smoothness and detail fidelity of the terrain surface.

[0077] Based on the classified high-density 3D point cloud data and original images, multi-view stereo matching technology is used to reconstruct the 3D models of buildings and major landforms.

[0078] The high-resolution original image is projected onto the surface of the 3D model to generate realistic textures, and multi-view texture fusion technology is used to process texture differences shot at different angles, eliminate seams and shadow problems, and improve texture quality.

[0079] For trees and shrubs, point cloud clustering method is used for 3D reconstruction; for bridges and tower structures, professional knowledge base and parametric modeling technology are combined to improve modeling accuracy and realism.

[0080] The overall 3D model is topologically checked and repaired, and the geometric accuracy and texture quality of the overall 3D model are evaluated by comparison with control points and field measurement data. Local adjustments and optimizations are performed when necessary, ultimately forming a high-quality, high-precision multi-view 3D model.

[0081] Furthermore, a high-precision digital terrain model is constructed using the ground points, and the Kriging method is used to achieve smoothness and detail fidelity of the terrain surface, including the following steps.

[0082] The classified ground points are processed by outlier removal and density optimization.

[0083] The processed ground points are preliminarily interpolated using the triangular irregular network algorithm to construct the initial digital terrain model.

[0084] The spatial autocorrelation analysis of the initial terrain model was carried out to establish a variogram model describing the terrain characteristics.

[0085] Based on the established variogram model, the ordinary Kriging algorithm is applied for high-precision terrain interpolation.

[0086] Combined with the original high-density point cloud data, key terrain features are locally optimized to improve the expression accuracy of key terrain features and generate a digital terrain model.

[0087] Perform a comprehensive quality assessment and validation of the digital terrain model, making local adjustments where necessary to achieve the best balance of smoothness and detail fidelity.

[0088] Furthermore, based on the classified high-density three-dimensional point cloud data and the original image, a multi-view stereo matching technology is used to reconstruct the three-dimensional models of buildings and main landforms, including the following steps.

[0089] Using the known exterior orientation elements and control point information, the classified high-density 3D point cloud data and the original image are accurately aligned to the same coordinate system.

[0090] Based on the classified point cloud data, the region growing algorithm is applied to extract the contours of buildings and major objects; and the image texture and color information are combined to optimize the boundary definition and generate a mask of the region of interest.

[0091] For each target to be reconstructed, the best multi-view image combination is selected based on the spatial position and camera parameters, and the feature description algorithm is applied to extract and match feature points between multi-view images to establish dense image correspondence.

[0092] Based on the established correspondence between feature points, considering multi-view geometric constraints and local smoothness constraints, a semi-global matching multi-view stereo algorithm is used for dense matching, and a depth map corresponding to each view is generated to describe the three-dimensional structure of the scene.

[0093] The multi-view depth map is projected into three-dimensional space to form a dense point cloud. The dense point cloud is registered and fused with the original high-density three-dimensional point cloud using an iterative closest point algorithm. Statistical filtering and moving least squares method are then used to optimize the fused point cloud to improve its quality.

[0094] Based on the optimized point cloud, plane fitting and contour extraction algorithms are used to reconstruct the three-dimensional models of buildings and major landforms. For regular buildings, a model-based reconstruction method is adopted; for complex landforms, a Poisson surface reconstruction algorithm is used to generate a fine triangular mesh model.

[0095] Furthermore, step 5 includes the following steps.

[0096] Preprocess the multi-view 3D model, including noise filtering and data simplification.

[0097] A semantic segmentation-based algorithm is used to segment the multi-view 3D model into different semantic regions.

[0098] Use deep learning algorithms to identify and extract building outlines, and automatically calculate the building's height, area, volume, and roof type parameters, while analyzing the building's orientation and density distribution.

[0099] The principal component analysis method is applied to identify and extract the road network, automatically calculate the road width, road length and road slope, and analyze the connectivity and density of the road network.

[0100] Use point cloud classification algorithms to distinguish different types of vegetation, calculate vegetation coverage, tree height and crown size, and perform vegetation density analysis.

[0101] Based on elevation and texture features, water system elements are identified and extracted, water area, river length and flow direction are automatically calculated, and the catchment area is analyzed.

[0102] Digital elevation models are used to analyze terrain features, including the calculation of slope, aspect, and terrain undulation; at the same time, special landforms are identified and extracted, and terrain roughness and surface curvature are calculated.

[0103] Identify and extract artificial facilities including bridges, tunnels and transmission lines, calculate the geometric parameters of artificial facilities, and analyze their spatial distribution characteristics.

[0104] Furthermore, the building height H 建筑 Calculated by the following formula: 建筑 =Z roof -Z ground , where Z roof is the average elevation of the building roof, extracted from the point cloud data; Z ground is the average elevation of the building ground, extracted from the digital terrain model.

[0105] Building area A 建筑 It is calculated through binary mask, that is, the sum of the pixel areas of all buildings. The specific calculation formula is: A 建筑 =∑ (x＾,y＾)∈M Area(x^,y^), where M is the binary mask of the building, (x^,y^)∈M represents all pixels with a value of 1 in the mask; Area(x^,y^) is the actual ground area corresponding to the pixel (x^,y^).

[0106] The volume of the building V 建筑 It is obtained by integrating the height of the building in the mask area. The specific calculation formula is expressed as: V 建筑 =∑ (x＾,y＾)∈M H(x^,y^)·Area(x^,y^), where H(x^,y^) is the height of the buildings at each point in the mask area.

[0107] Roof type parameters T 屋顶 The deep learning classification algorithm is used to classify the extracted building roof features, which can be expressed as: T 屋顶 =f roof (I roof ), where I roof is the extracted building roof image or feature map; f roof is a deep learning based classifier for determining roof type.

[0108] Building orientation θ 建筑 It is calculated by extracting the main axis direction of the building outline, and the formula is: θ 建筑 =atan2(v y ,v x ), where v y and v x It is the direction vector component of the contour principal axis, which is calculated by principal component analysis method.

[0109] Building density ρ 建筑 It is calculated by the number of buildings per unit area. The specific formula is: 建筑 =A total / A region , where A total is the total area of ​​all buildings in the survey area; A region is the total area of ​​the survey area.

[0110] Road width W 道路 It is obtained by calculating the distance between road boundaries along the orthogonal direction of the road direction. The specific calculation formula is: W 道路 =1 / NB·∑ h=1 NB d h , where NB is the number of points sampled along the road direction; d h is the road width at the hth sampling point, calculated by the distance between road boundary points.

[0111] Road length L 道路 It is calculated by integrating along the center line of the road, and the formula is: L 道路 =∑ o=1 m √((x o -xo-1 ) 2 +(y o -y o-1 ) 2 ), where m is the number of sampling points on the road centerline; (x o ,y o ) is the coordinate of the oth sampling point; (x o-1 ,y o-1 ) are the coordinates of the o-1th sampling point.

[0112] Road slope S 道路 It is calculated by the ratio of the elevation difference between the starting point and the end point of the road to the horizontal distance. The formula is: S 道路 =(Z end -Z start ) / L 道路 ×100%, where Z star , Z end are the elevations of the start and end points of the road respectively.

[0113] Road network connectivity C 道路 It is represented by calculating the connected components of the road network: C 道路 =2E 道路 / (N 道路 (N 道路 -1)), where E 道路 is the number of edges in the road network; N 道路 is the number of intersections in the road network.

[0114] Road density D 道路 Calculated by the total length of roads within a unit area: D 道路 =L total / A region , where L total is the total length of all roads within the specified area.

[0115] The vegetation coverage rate is the ratio of the area covered by vegetation in the survey area to the total area of ​​the survey area.

[0116] Tree height H 树木 It is the difference between the maximum height of vegetation points and the minimum height of ground points in the point cloud data.

[0117] Tree crown size C size It is expressed by the horizontal projection area of ​​the tree crown: C size =(D max / 2)×(D min / 2), where D max and D min are the maximum and minimum diameters of the crown, obtained by analyzing the horizontal projection of the crown point cloud.

[0118] Vegetation density D veg It is the amount of vegetation per unit area or the proportion of vegetation coverage.

[0119] Water area A water The calculation formula is: water =∑ (p,q)∈R I(W(p,q))·A pixel , where R is the coordinate set of the survey area; I(W(p,q)) is an indicator function. When W(p,q) is "water body", I(W(p,q))=1, otherwise I(W(p,q))=0; A pixel is the actual geographic area corresponding to each pixel.

[0120] River length L river The calculation formula is: L river =∑ i=1 NR-1 √((x e+1 -x e ) 2 +(y e+1 -y e ) 2 ), where (x e ,y e ) represents the coordinates of the e-th point on the river path; NR is the number of all points on the river path.

[0121] The formula for calculating the river flow direction is: D flow (α,β)=arctan((E(α2,β2)-E(α1,β1)) / √((α2-α1) 2 +(β2-β1) 2 )), where D flow (α, β) is the direction of the river, that is, the direction angle of the water flow from point (α1, β1) to point (α2, β2); E(α1, β1) and E(α2, β2) are the elevations of point (α1, β1) and point (α2, β2) respectively.

[0122] Furthermore, step 6 includes the following steps.

[0123] The data formats of multi-view 3D models, extracted land features and calculated terrain parameters are unified and the coordinate system is converted to ensure that all data are in the same coordinate system and comply with the data standards of the natural resource management system.

[0124] The multi-view three-dimensional model is imported into the natural resource management system, and is registered and fused with the existing two-dimensional map data to form a registered and fused model.

[0125] The automatically extracted ground features are superimposed on the registration and fusion model in the form of layers, and are associated and attribute matched with the corresponding elements in the natural resource management system to form a feature superposition model.

[0126] The calculated terrain parameters are integrated into the feature overlay model to form a comprehensive geographic information model.

[0127] The data processing function of the natural resource management system is utilized to conduct quality checks and topological relationship verification on the comprehensive geographic information model to ensure the consistency and integrity of the spatial data.

[0128] Combine historical data from natural resource management systems for spatiotemporal analysis and change detection in integrated geographic information models.

[0129] Utilize the visualization function of the natural resource management system to generate diverse results including 3D topographic maps, thematic layers and statistical reports.

[0130] The comprehensive geographic information model and generated results are packaged to form a complete integrated air-space-ground mapping database, and ensure a real-time synchronization update mechanism with the natural resource management system.

[0131] Compared with the prior art, the present application has the following beneficial effects.

[0132] This application uses industrial-grade RTK drones, consumer-grade drones and ground control points to collaboratively collect high-quality oblique photography data, and generates high-precision, multi-perspective three-dimensional models through computer vision algorithms and multi-perspective three-dimensional modeling technology, thereby realizing the fusion and detailed mapping of multi-dimensional data of air, space, and land, improving the coverage and mapping accuracy of complex terrain, and being able to automatically extract and analyze land features, providing comprehensive data support for natural resource management systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0133] Figure 1 A schematic flow chart of a method for integrating air-space-ground and land-surface mapping based on oblique photography three-dimensional mapping disclosed in an embodiment of the present application. DETAILED DESCRIPTION

[0134] In order to make the purpose, technical scheme and advantages of the implementation of the present invention clearer, the technical scheme in the embodiment of the present invention will be described in more detail below in conjunction with the drawings in the embodiment of the present invention. In the drawings, the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The described embodiments are part of the embodiments of the present invention, not all of the embodiments.

[0135] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making any creative work shall fall within the scope of protection of the present invention.

[0136] The embodiments and directional terms described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be construed as limiting the present invention.

[0137] like Figure 1 As shown, a method for integrated air-space-ground mapping based on oblique photography three-dimensional mapping includes the following steps.

[0138] Step 1: Develop a comprehensive oblique photogrammetry program to ensure complete coverage of complex terrain. The purpose of this step includes planning the flight route of the survey area, selecting the appropriate flight altitude and angle, and determining the frequency and scope of data collection. Through an accurate surveying and mapping program, complex terrain can be systematically covered, including various types of landforms such as mountains, canyons, and cities. This not only avoids blind spots in data collection, but also improves the efficiency of subsequent data processing. Scientific and reasonable planning can effectively reduce overlapping areas and blank areas, ensure that each area can be fully supported by image data, and provide accurate basic data for subsequent 3D reconstruction and ground feature analysis.

[0139] Step 2: Deploy industrial RTK drones and consumer drones to work together to achieve a combination of large-scale coverage and local fine collection, and obtain high-quality oblique photography image data. In this step, the collaborative operation of industrial RTK drones and consumer drones can achieve dual data collection of large-scale and local fine. Industrial RTK drones have high-precision positioning systems and stable flight performance, which are suitable for rapid coverage of a large area and can obtain relatively wide-ranging image data at high altitudes. Consumer drones can perform refined operations in complex environments, enter areas that are difficult for industrial drones to cover, and obtain high-resolution local image data. The combined use of these two drones not only improves the coverage and accuracy of data collection, but also optimizes resource utilization, ensures high-quality image data collection, and provides comprehensive and detailed image data for subsequent processing.

[0140] Step 3, integrate RTK positioning data and ground control point data as control information, and perform aerial triangulation encryption processing on the oblique photography images. In this step, the integration of RTK positioning data and ground control point data can effectively improve the spatial accuracy of image data. The RTK positioning system provides high-precision geographic positioning information through real-time satellite data, while the ground control point data provides the actual coordinate points on the ground. Combining the two, the oblique photography images can be encrypted by aerial triangulation, that is, by establishing a spatial relationship between images, the geometric accuracy and positioning accuracy of the images can be improved. This process can effectively reduce the impact caused by image distortion or position error, improve overall accuracy and reliability, and lay a solid foundation for subsequent three-dimensional modeling and data analysis.

[0141] Step 4: Build a high-quality, high-precision multi-view 3D model. Building a high-quality, high-precision multi-view 3D model is one of the core steps of the entire mapping method. By integrating multi-angle oblique image data and using computer vision and image processing technology, a detailed and accurate 3D model can be generated. This process involves technical links such as 3D reconstruction, texture mapping and detail optimization of images. High-quality 3D models can truly reflect the spatial layout of terrain and objects, and provide rich visual information and spatial data support. High-precision models not only highly restore reality in terms of geometric shape, but also support various subsequent analyses and applications, such as urban planning and environmental monitoring.

[0142] Step 5: Automated feature extraction and identification of objects based on multi-view 3D models, as well as automatic calculation of terrain element parameters. In this step, automated feature extraction and identification of objects using multi-view 3D models is an important means to improve data processing efficiency and accuracy. Through image recognition and machine learning technology, features of objects such as buildings, roads, and vegetation can be automatically extracted from the 3D model, and parameters of terrain elements such as elevation and slope can be calculated. This automated processing method not only reduces the need for manual intervention, but also improves the speed and accuracy of data processing, providing efficient support for terrain analysis, resource assessment, etc.

[0143] Step 6: Integrate the multi-view 3D model, land feature and terrain element parameters with the natural resource management system to generate integrated air-space-ground mapping. Such mapping not only integrates high-precision 3D spatial data, but also combines land feature and terrain information to provide complete geographic information support. This integration can achieve real-time resource monitoring, dynamic change analysis and accurate decision support, improve the efficiency and scientificity of natural resource management, and has important practical application value and significance.

[0144] Furthermore, step 1 includes the following steps.

[0145] Collect the existing geographic information data of the survey area and conduct preliminary analysis to understand the terrain characteristics, distribution of objects and their complexity in the survey area. First of all, collecting the existing geographic information data of the survey area, including satellite images, topographic maps, geological data, etc., is the basis for formulating a comprehensive surveying and mapping plan. Through these data, we can have a preliminary understanding of the terrain characteristics (such as mountains, hills, plains, etc.), the distribution of objects (such as buildings, roads, vegetation, etc.) and their complexity in the survey area. This information can help analyze the overall topography of the survey area, potential challenge areas (such as steep slopes, densely built-up areas, etc.) and the difficulty of data collection. This preliminary analysis provides basic data support for subsequent detailed planning, making the entire surveying and mapping plan more targeted and scientific. Specifically, understanding the terrain characteristics and complexity can help identify areas that need to be focused on and reasonably allocate surveying and mapping resources, thereby optimizing data collection strategies, reducing resource waste, and ensuring the comprehensiveness and accuracy of data collection.

[0146] GIS software is used to analyze terrain features in detail, calculate and visualize factors such as slope, aspect, and elevation changes, and thus identify flat areas, steep areas, and complex terrain areas within the survey area. The application of GIS software plays an important role in the detailed analysis of terrain features. Through GIS technology, the terrain data of the survey area can be accurately analyzed and visualized, including key factors such as slope, aspect, and elevation changes. Slope and aspect analysis can help identify flat areas and steep areas, which is important for planning drone flight routes and adjusting shooting angles. Visualization of elevation changes helps identify complex terrain areas that may pose challenges to data collection. The results of GIS analysis can be intuitively displayed in the form of layers, allowing planners to clearly see the distribution of different terrain areas, so as to formulate targeted collection strategies to ensure comprehensive coverage and accuracy of data.

[0147] According to the results of terrain feature analysis, the survey area is divided into several blocks with similar features. Partitioning enables the development of corresponding surveying and mapping plans based on the specific terrain features of each block. For example, flat areas and steep areas require different flight altitudes and angle settings, while complex terrain areas require special shooting strategies. By dividing the survey area into different blocks, the flight route and data collection strategy can be better optimized, thereby improving the efficiency and accuracy of surveying and mapping. In addition, this partitioning method can also help identify special needs or challenges within each block, making the subsequent flight route design and control point layout more scientific and reasonable.

[0148] For each block, flight routes with different heights and overlaps are designed according to its terrain features to ensure the integrity and accuracy of data collection. Designing a suitable flight route for each terrain block is the key to ensuring the quality of oblique photography data collection. When planning the flight route, different flight heights and overlaps need to be designed according to the terrain characteristics of each block (such as flat, steep, and complex). For example, a higher flight height and lower overlap can be used in flat areas, while a lower flight height and higher overlap are required in steep or complex terrain areas to ensure image clarity and data integrity. This personalized flight route design can effectively cover all feature areas in the survey area and avoid omissions or data overlap, thereby improving the overall data quality and the accuracy of subsequent three-dimensional models.

[0149] According to the area of ​​the survey area and the complexity of the terrain, the number and distribution of ground control points are determined to improve the accuracy of surveying and mapping. The layout of ground control points is crucial to improving the accuracy of surveying and mapping. According to the area of ​​the survey area and the complexity of the terrain, the number and distribution of ground control points can be reasonably determined. Ground control points are used to provide accurate geographic coordinate information, which is helpful for subsequent image matching and three-dimensional reconstruction. For large areas or areas with complex terrain, more control points need to be laid out to ensure the accuracy and completeness of the data. Reasonable layout of control points can not only reduce errors, but also improve the geometric accuracy of the three-dimensional model. This process usually involves technical operations such as the selection, calibration and measurement of control points, which need to be scientifically planned and optimized in combination with the actual situation of the survey area.

[0150] According to the characteristics of each block and the requirements of mapping, the camera parameters and shooting parameters are determined to obtain high-quality oblique photography data. Determining the camera parameters and shooting parameters is a key step in obtaining high-quality oblique photography data. Different terrain areas require different camera settings, such as different focal lengths, resolutions, and exposure parameters. According to the characteristics of each block and the requirements of mapping, the reasonable setting of these parameters can ensure the clarity and detail of the image. The selection of shooting parameters also needs to consider factors such as lighting conditions, ground reflectivity, and weather conditions to obtain the best data quality. High-quality oblique photography data can provide more detailed information, thereby improving the accuracy of subsequent 3D modeling and analysis, and providing more reliable data support for natural resource management and other applications.

[0151] The flight route, control point layout and shooting parameters are evaluated through simulation software, and fine-tuned and optimized according to the evaluation results to ensure the scientificity and operability of the final mapping plan. Through simulation, potential problems can be foreseen, such as whether there are overlaps or blind spots in the flight route, whether the control points are evenly distributed, and whether the shooting parameters meet the requirements. Fine-tuning and optimization based on the simulation results can further improve the rationality and effectiveness of the mapping plan. This process can help discover and solve difficulties that may arise in actual operations and ensure the smooth progress of mapping work. The optimized plan can better meet the requirements of data collection in actual operations, reduce unnecessary adjustments and corrections, and improve mapping efficiency and data quality.

[0152] Furthermore, step 2 includes the following steps.

[0153] According to the characteristics of the survey area and the mission requirements, select industrial RTK drones and consumer drones, and ensure that the performance parameters of the selected equipment meet the mission requirements. Selecting the right drone is the first step to ensure the quality of data collection. Industrial RTK drones and consumer drones each have their own advantages. The former has a high-precision positioning system and stronger stability, which is suitable for large-scale, high-precision mapping tasks; the latter usually has higher flexibility and lower cost, which is suitable for detailed collection and supplementary data collection in specific areas. According to the characteristics of the survey area (such as terrain complexity, survey area, etc.) and mission requirements (such as data resolution, flight time, etc.), selecting a drone that meets the requirements can ensure that its performance parameters (such as flight time, load capacity, resolution, etc.) meet the mission requirements. By reasonably selecting a drone, you can maximize the performance of the equipment, ensure that the collected data has sufficient accuracy and coverage, and reduce potential operational risks and costs.

[0154] Based on the block division and flight route design, the operating range and flight routes of industrial RTK drones and consumer drones are planned. Industrial RTK drones are responsible for efficient data collection over a large area, including flat areas and large areas of simple terrain; while consumer drones are used to supplement details in complex terrain and hard-to-reach areas. Through detailed flight route planning, it can be ensured that each drone can effectively complete its mission in its designated area while avoiding data duplication or omission. Reasonable flight route design can not only improve the comprehensiveness of data collection, but also optimize flight time and path, thereby improving the efficiency and effectiveness of the overall surveying and mapping work.

[0155] An RTK base station is set up in the survey area to ensure good communication with the drone; at the same time, a mobile station for an industrial-grade RTK drone is configured to obtain high-precision location information. The RTK base station can greatly improve the positioning accuracy of the drone and reduce positioning errors by providing real-time positioning information. The mobile station cooperates with the RTK base station to provide the drone with real-time and accurate location information, allowing the drone to maintain high-precision positioning during flight. It is crucial to ensure good communication between the RTK base station and the drone, which can not only update the positioning data in real time, but also correct the positioning deviation in real time during the data collection process, thereby improving the spatial accuracy and consistency of the final image data. The optimization of this process can significantly improve the quality of surveying and mapping data and ensure the reliability of subsequent processing and analysis.

[0156] According to the predetermined flight route and shooting parameters, the selected industrial RTK drone is used to collect oblique photography data; at the same time, the flight status, image acquisition and RTK signal quality are monitored in real time to ensure the integrity and accuracy of data collection. In the actual data collection process, the core task is to use the industrial RTK drone to collect oblique photography data according to the predetermined flight route and shooting parameters. At this time, it is necessary to monitor the flight status of the drone (such as flight altitude, speed, etc.), image acquisition (such as whether the image is clear, whether it is shot as planned) and RTK signal quality (such as signal strength, stability, etc.) in real time. Real-time monitoring can detect and solve problems in a timely manner, such as adjusting the flight path to avoid interference, reshooting blurry images, etc. Through these monitoring measures, the integrity and accuracy of data collection can be ensured, the risk of data loss or substandard quality can be reduced, and the reliability and effectiveness of the overall surveying and mapping results can be improved.

[0157] Based on the preliminary results and on-site observations collected by the industrial RTK drone, areas that require detailed collection are identified. After the industrial RTK drone completes the initial data collection, it is necessary to conduct a preliminary analysis of the collected image data and identify areas that require detailed collection based on on-site observations. These areas include complex terrain, important objects, areas with insufficient data coverage, etc. By analyzing the preliminary data, it is possible to determine which areas have insufficient details and which areas need to improve data quality. After identifying these areas, further data collection tasks can be arranged in a targeted manner to ensure that all important areas can obtain high-quality data. This process can improve the detail accuracy of the data and provide more accurate basic data for subsequent 3D reconstruction and analysis.

[0158] Use consumer drones to perform additional collection of areas that require fine collection, including close-up photography of complex terrain, multi-angle collection of special objects, or additional coverage of shadow areas; during the flight, the acquired image data is checked for quality in real time, including image clarity, overlap, and color uniformity. If problems are found, adjust the flight parameters or re-collect in time. Due to their flexibility and low cost, consumer drones are suitable for close-up photography of complex terrain, multi-angle collection of special objects, or additional coverage of shadow areas. During the flight, the acquired image data needs to be checked for quality in real time, including checking the clarity, overlap, and color uniformity of the image. Through real-time inspection, image quality problems can be discovered in a timely manner, and corresponding adjustments or re-collection can be made to ensure that the final data meets the quality requirements. This process can supplement and improve the data, improve the detail accuracy and practical application effect of the 3D model.

[0159] Cross-validation flights are performed at the interface between data collected by industrial RTK drones and consumer drones to detect and eliminate data gaps or inconsistent areas. Cross-validation flights can detect whether there are gaps or inconsistent areas in the data collected by two different types of drones at the interface. By performing additional flight data collection in these interface areas, gaps between data can be filled and data inconsistencies can be corrected. This step helps ensure seamless data coverage across the survey area, avoids data omissions or conflicts, and thus improves the overall consistency and reliability of the final survey and mapping results.

[0160] Furthermore, step 3 includes the following steps.

[0161] Step 3.1, extract the precise RTK positioning information of each image from the flight data collected by the industrial RTK UAV, including the longitude, latitude, elevation and attitude angle data at the time of shooting. The RTK (Real-Time Kinematic) positioning system can provide high-precision geographic coordinate data, including longitude, latitude, elevation and attitude angle (such as pitch angle, roll angle and heading angle). These data can accurately describe the spatial position and direction of each image at the time of shooting, and become the key to establishing the correspondence between the image and the ground true position. Accurate positioning information helps to reduce errors in subsequent processing and improve the accuracy of image geometric correction. By accurately positioning each image, the spatial consistency of the 3D model and map can be ensured, providing a reliable data basis for subsequent image matching and 3D reconstruction.

[0162] Step 3.2, integrate the RTK positioning data and ground control point data into the same coordinate system to establish a unified control point database. Ground control points refer to landmarks with known positions on the ground, whose coordinates are obtained through precise measurement. These control points together with the RTK positioning data constitute a unified control point database for subsequent data processing and correction. The integration process requires coordinate conversion of data from different sources to ensure that all data is processed in the same coordinate system. This ensures high-precision matching between images and ground control points, reduces errors caused by inconsistent coordinate systems, and improves the accuracy and reliability of subsequent 3D modeling and analysis.

[0163] Step 3.3, pre-process the acquired oblique photography images, including image light and color grading, noise removal and blur operations to improve image quality; at the same time, extract the EXIF ​​information of each image, match it with the RTK positioning data, and establish a preliminary image-position correspondence. Image light and color grading helps solve the problem of uneven image brightness and color caused by lighting changes; removing noise and blur can improve the clarity and detail of the image. In addition, extracting the EXIF ​​information of the image (such as shooting time, camera settings, etc.) and matching it with the RTK positioning data can establish a preliminary image-position correspondence and provide basic data for geometric correction and matching of the image. This process can improve the quality of image data and ensure the accuracy and effect of subsequent processing.

[0164] Step 3.4, using computer vision algorithms, automatically identify and match feature points between adjacent images, generate a large number of connection points and construct an image network for the entire survey area. By automatically identifying and matching feature points between adjacent images, a large number of connection points can be generated, which are used to establish an image network (spatial relationship network of images). Feature point matching algorithms can identify common features in images (such as corners, edges, etc.) and establish corresponding relationships between different images. The generated image network can help stitch different images into a complete 3D model and provide a panoramic view. Accurate feature point matching can ensure the accuracy of the image network, thereby improving the quality of subsequent 3D reconstruction and analysis.

[0165] In step 3.5, the RTK positioning data, ground control point data and the image network of the entire survey area are used as inputs to perform aerial triangulation calculations using professional photogrammetry software. Aerial triangulation can determine the precise spatial position and posture of each image by performing geometric calculations on the connection points in the image network, thereby achieving geometric correction of the image. Using RTK positioning data and ground control point data as inputs can improve the accuracy of the calculations, allowing the three-dimensional model to accurately reflect the actual terrain and objects. This process involves complex mathematical models and calculations, which can convert the two-dimensional information in the image into three-dimensional spatial data, providing accurate basic data for subsequent three-dimensional modeling and analysis.

[0166] Step 3.6, conduct a comprehensive error analysis of the aerial triangulation results, including control point residuals, tie point residuals, and image network adjustment accuracy. Error analysis is a key step to ensure the accuracy of measurement results. A comprehensive error analysis of the aerial triangulation results, including control point residuals (differences between actual and calculated positions), tie point residuals (errors between matching points), and image network adjustment accuracy (adjustment accuracy of the overall image network), can identify and quantify errors in the data. Through error analysis, the overall accuracy of the image network can be evaluated and potential problems and deficiencies can be discovered. Accurate error analysis can help identify inconsistencies and deviations in data processing, provide a basis for subsequent optimization and adjustment, and thus improve the accuracy and reliability of the final surveying and mapping results.

[0167] Step 3.7, based on the error analysis results, remove gross error points, adjust weights, and perform multiple iterative calculations until the predetermined accuracy requirements are met. The accuracy of the 3D model can be continuously improved by removing gross error points (i.e. points that deviate significantly from normal values), adjusting weights (rationally allocating the contributions of different data sources), and performing multiple iterative calculations. During the iterative calculation process, each calculation is adjusted according to the results of the error analysis until the predetermined accuracy requirements are met. This process can effectively reduce errors in the data, improve the geometric accuracy and consistency of the image network, and thus ensure the accuracy and reliability of the final surveying and mapping results.

[0168] Step 3.8, based on the optimized aerial triangulation results, update the precise position and attitude information of each image to obtain the optimized image data. After completing the error optimization and iterative calculation, it is necessary to update the precise position and attitude information of each image based on the optimized aerial triangulation results. This process can ensure that the spatial position and direction of each image are consistent with the actual terrain, thereby improving the accuracy and consistency of the image data. The optimized image data can more accurately reflect the actual terrain and objects, providing a reliable foundation for subsequent 3D reconstruction and data analysis. The result of this process is high-precision image data, which provides key support for the generation of high-quality 3D models and maps.

[0169] Step 3.9, use the optimized image data and dense matching algorithm to generate high-density 3D point cloud data. High-density 3D point cloud data can not only provide rich spatial information, but also support accurate terrain analysis and data visualization. The technical effect of this process is to significantly improve the level of detail and accuracy of the 3D model, providing detailed geographic information support for practical applications.

[0170] Furthermore, a computer vision algorithm is used to automatically identify and match feature points between adjacent images, generate a large number of connection points and construct an image network of the entire survey area, including the following steps.

[0171] Computer vision algorithms are applied to each image to identify and extract significant and stable feature points. Feature point extraction is an important part of computer vision, which can identify significant and stable feature points from images. These feature points are local areas in the image with significant brightness or color contrast, which can be reliably detected in different images. Common feature point extraction algorithms include SIFT (Scale Invariant Feature Transform), SURF (Speeded Up Robust Features), and ORB (Oriented FAST and Rotated BRIEF). These algorithms can stably detect feature points at different scales and rotation angles, thereby improving the robustness of image matching. Identifying significant and stable feature points can lay the foundation for subsequent matching and geometric analysis, so that image data can be effectively registered and reconstructed.

[0172] For each extracted feature point, its feature descriptor is calculated to describe the unique properties of the area around the feature point. The feature descriptor is a vector used to describe the unique properties of the area around the feature point. Its purpose is to accurately match features between images. Commonly used feature descriptor algorithms include SIFT descriptor, SURF descriptor and BRIEF (Binary Robust Independent Elementary Features) descriptor. The feature descriptor encodes the local texture of the area around the feature point to generate a feature vector, which is used to distinguish and match feature points in different images. Calculating the feature descriptor can convert the local features in the image into a numerical form that is easy to compare, thereby improving the accuracy and robustness of feature point matching. The uniqueness and stability of the descriptor directly affect the quality of feature point matching, thereby affecting the accuracy of 3D reconstruction.

[0173] The nearest neighbor search algorithm is used to perform preliminary matching of feature points between adjacent images. Between adjacent images, the nearest neighbor search can quickly find point pairs with similar feature descriptors, which serve as candidates for preliminary matching. Commonly used nearest neighbor search algorithms include Brute-Force Matcher and KD tree-based matching methods. In this way, feature points in adjacent images can be effectively paired, laying the foundation for subsequent geometric verification and optimization. The accuracy of the preliminary matching has an important impact on the results of subsequent steps, so choosing the right search algorithm and parameter settings is the key to ensuring high matching quality.

[0174] For the initially matched feature point pairs, the random sample consensus algorithm is applied for geometric verification to effectively remove false matches and retain only the matching point pairs that meet the expected geometric relationship. The random sample consensus algorithm (RANSAC) is a powerful tool for dealing with errors and outliers in feature point matching. The RANSAC algorithm removes false matches by randomly selecting point pairs and estimating a model (such as a base matrix or homography matrix), and then verifying whether other point pairs conform to the model. In an iterative manner, RANSAC can identify the correct geometric relationship from a large number of matching point pairs and eliminate false matches that do not meet geometric constraints. This process can significantly improve the accuracy and reliability of feature point matching and provide clean matching data for subsequent image network construction. Accurate geometric verification can reduce errors and ensure that the final 3D model has a high spatial consistency.

[0175] Based on the verified matching points, the image connection network of the entire survey area is constructed, and the relative orientation between each pair of adjacent images is calculated to provide initial values ​​for subsequent aerial triangulation. Based on the geometrically verified matching points, an image connection network of the entire survey area can be constructed to describe the relative position and direction between images. This network provides initial values ​​for subsequent aerial triangulation by calculating the relative orientation (such as rotation and translation) between each pair of adjacent images. An accurate image connection network can ensure the correct spatial relationship between images and provide a reliable starting point for further optimization and correction. The accuracy of constructing the image network directly affects the effect of 3D reconstruction, and an accurate network structure can effectively support subsequent data processing steps.

[0176] The entire image network is globally optimized, including the use of the bundle adjustment algorithm to jointly optimize all tie points and camera parameters to minimize reprojection errors to further improve the accuracy and consistency of tie points. The Bundle Adjustment algorithm is an optimization technique that improves the accuracy of the 3D model by jointly optimizing the tie points and camera parameters of all images to minimize reprojection errors. Reprojection error refers to the difference between the position of a 3D point and the actual position of the point in the image after it is projected back to the image plane. Bundle Adjustment optimizes the overall consistency of the image network by iteratively adjusting the camera's internal and external parameters and the position of the 3D points. Through this optimization, the accuracy of tie points can be significantly improved, and the inconsistency between images can be reduced, resulting in more accurate and reliable 3D point cloud data. This process is the basis for the final generation of a high-quality 3D model and has a significant impact on the accuracy and reliability of surveying and mapping results.

[0177] Furthermore, a computer vision algorithm is applied to each image to identify and extract feature points with significance and stability, including the following steps.

[0178] The gradient of each image is calculated to obtain the gradient amplitude and direction. The specific calculation formula is: M(x,y)=√{(I x (x,y) 2 +(I y (x,y) 2},θ(x,y)=arctan(I y (x,y) / I x (x,y)), where I x (x,y) and I y (x,y) represents the partial derivatives of the image in the x and y directions respectively; θ(x,y) represents the gradient direction at the image position (x,y); M(x,y) represents the gradient amplitude at the image position (x,y).

[0179] Images of different scales are generated by Gaussian blur technology. The specific formula is: L(x,y,σ)=G(x,y,σ)*I(x,y), where I(x,y) represents the pixel value at the image position (x,y); L(x,y,σ) represents the Gaussian blurred image at scale σ; G(x,y,σ) is the Gaussian kernel function: G(x,y,σ)=1 / (2πσ 2 )·exp(-(x 2 +y 2 ) / (2σ 2 )), where * represents the convolution operation.

[0180] The differential Gaussian image is calculated between images of different scales. The specific formula is: D(x,y,σ)=L(x,y,kσ)-L(x,y,σ), where D(x,y,σ) represents the differential Gaussian image at scale σ; L(x,y,kσ) represents the Gaussian blurred image at scale kσ; k is a multiple of the scale and is used to generate adjacent scale images; L(x,y,σ) represents the Gaussian blurred image at different scales.

[0181] In the differential Gaussian images of different scales, local extreme points, i.e. feature points, are detected.

[0182] Optimize the sub-pixel accuracy of the detected feature points to improve the stability of the feature points. The formula is: x^=-H -1 ▽D, where H is the Hessian matrix, ▽D is the gradient of the differential Gaussian image; x^ represents the optimized feature point position.

[0183] Through the precise calculation and optimization of the above steps, the significant and stable feature points in the image can be effectively extracted, providing a solid data foundation for subsequent image registration, feature matching and 3D reconstruction. These technical steps work together to improve the quality and stability of feature point extraction, thereby improving the performance of the entire surveying and modeling system.

[0184] Furthermore, for each extracted feature point, its feature descriptor is calculated, including the following steps.

[0185] For each feature point, a fixed-size neighborhood window is selected. The purpose of this step is to define an area around the feature point, which contains the local information of the feature point. The size of the neighborhood window directly affects the quality of the feature descriptor and the stability of the feature point. Generally, the size of the neighborhood window should be adjusted according to the resolution of the image and the scale of the feature point. A fixed-size neighborhood window can effectively localize the information of the feature point, so that the calculated feature descriptor can better reflect the texture and structural characteristics around the feature point. By defining an appropriate window size, it can be ensured that the key local information of the feature point is captured, thereby improving the accuracy of subsequent matching.

[0186] Calculate the gradient magnitude and gradient direction of each pixel in the neighborhood window.

[0187] The neighborhood is divided into sub-regions. In each sub-region, the gradient direction histogram is calculated and the gradient direction is quantized into n direction buckets. The value of each bucket is the accumulation of the gradient amplitude in that direction. The specific formula is: i,j =∑ k=1 n Bin i,j,k , where H i,j is the total gradient amplitude of the k-th direction bucket in the (i, j)-th sub-region; Bin i,j,k It is the cumulative value of the gradient magnitude of all pixels in the sub-region projected into the k-th direction bucket.

[0188] The histograms of all sub-regions are concatenated into a feature descriptor.

[0189] Normalize the feature descriptor to improve robustness to lighting changes: f i ′=f i / ‖f i ‖, where ‖f i ‖ is the feature descriptor f i The Euclidean norm of i ′ is the normalized feature descriptor vector.

[0190] Through the above steps, the descriptor of the feature point can fully reflect the local texture information around the feature point and has good robustness to illumination changes. These feature descriptors play a vital role in feature matching and image analysis, and can support high-precision feature point matching and 3D reconstruction.

[0191] Furthermore, the relative orientation between each pair of adjacent images is calculated through the following steps.

[0192] From the verified feature point matches, collect all matching point pairs between adjacent images.

[0193] By matching point pairs, estimate the fundamental matrix F AB , the specific formula is: F AB =argmin F ∑ g=1 N ρ(x A,g T Fx B,g ), where x A,g and x B,g is the g-th pair of matching points; ρ is the reprojection error; F AB =argmin F It means to find a basic matrix F so that the subsequent error function is minimized; ∑ g=1 N ρ(x A,g T Fx B,g ) is the cost function, which represents the sum of the reprojection errors for all matching point pairs; ρ(·) is the squared error metric.

[0194] Using the known camera intrinsic parameter matrix K A and K B , the fundamental matrix F AB Converted to the essential matrix, the formula is: E AB :E AB =K B T F AB K A , where K B T Represents the transpose of the camera intrinsic parameter matrix of image B.

[0195] The essential matrix E AB Decomposed into the rotation matrix R AB and the translation vector t AB , the formula is: E AB =R AB [t AB ] × , where [t AB ] × is the translation vector tAB The antisymmetric matrix of .

[0196] Get the relative position, specifically expressed as: T AB =(R AB ,t AB ).

[0197] Through detailed calculation and optimization of the above steps, the relative orientation between adjacent images can be accurately obtained, providing a solid geometric foundation for subsequent image registration, three-dimensional reconstruction and other computer vision tasks.

[0198] Furthermore, the RTK positioning data, ground control point data and the image network of the entire survey area are used as inputs to perform aerial triangulation calculations using professional photogrammetry software, including the following steps.

[0199] Import RTK positioning data, ground control point data and the image network of the entire survey area into professional photogrammetry software. The accuracy and completeness of the imported data directly affect the accuracy and reliability of subsequent measurement calculations. Professional photogrammetry software can integrate these data through efficient data management and processing functions, establish a relationship model between images and control points, and provide a basis for subsequent triangulation and optimization.

[0200] Based on RTK positioning data and image EXIF ​​information, the initial exterior orientation elements are calculated for each image, including position (X, Y, Z) and attitude angles (ω, φ, μ). TK positioning data provides the precise position of the image, while image EXIF ​​information provides the internal parameters of the camera and the shooting angle. With these data, the software can calculate the initial exterior orientation elements of the image, providing an initial estimate for subsequent global adjustment and optimization. Accurate calculation of the initial exterior orientation elements is crucial for subsequent global adjustment, because it provides a starting point for the optimization algorithm, affecting the accuracy and convergence speed of the final result.

[0201] Start the automatic aerial triangulation program, use the tie point information and initial exterior orientation elements in the image network, and perform a global adjustment operation to minimize the reprojection error of the entire image network through iterative optimization. The goal of global adjustment is to accurately determine the exterior orientation elements of each image by minimizing the reprojection error of the entire image network through an iterative optimization process. The adjustment algorithm adjusts the exterior orientation elements of the image so that the feature points in the image are more accurately aligned in three-dimensional space. The global adjustment process improves the accuracy and consistency of the overall measurement results by comprehensively considering all images and tie points. The process involves complex mathematical operations and optimization techniques to ensure that the geometric relationships between all images are optimally matched.

[0202] The software automatically identifies and accurately locates known ground control points on the corresponding image and matches them to the image. By matching these control points with their actual locations in the image, the software can further correct the geometric relationships in the image. The process of automatically identifying and locating control points requires high-precision image processing and feature matching algorithms to ensure that the positions of the control points on the image are accurately determined. Accurately matching control points helps to further optimize the exterior orientation elements of the image and improve the accuracy of the overall measurement results.

[0203] The software calculates the residuals for each tie point and control point, and automatically identifies and processes possible gross error points based on preset thresholds. Residuals refer to the difference between actual observed values ​​and theoretical values. By analyzing these residuals, anomalies or errors in the data can be found. Based on the preset thresholds, the software automatically identifies and processes possible gross error points, which are usually caused by mismatching or measurement errors. The processing of gross error points includes removing abnormal data or re-adjusting weights to improve the accuracy and reliability of the overall adjustment results. Automated gross error processing can effectively reduce human intervention and improve the efficiency and accuracy of the adjustment process.

[0204] According to the accuracy characteristics of different data sources, a dynamic weight allocation strategy is established. Different data sources (such as RTK positioning data, ground control point data, and image data) have different accuracy levels, so in the adjustment process, it is necessary to assign appropriate weights to each data source. The weight allocation strategy optimizes the overall measurement results by considering the reliability and accuracy of the data and increasing the emphasis on high-precision data. Dynamic weight allocation can be flexibly adjusted during the adjustment process, so that data from different sources can be reasonably integrated in the final result, improving measurement accuracy and consistency.

[0205] Perform the final adjustment calculation, integrate all RTK positioning data, ground control point data and constraints, and solve the optimal exterior orientation elements and ground point coordinates through an iterative algorithm; at the same time, generate a comprehensive accuracy assessment report including control point residuals, tie point residuals and image point residuals. The iterative algorithm will continuously optimize the exterior orientation elements and ground point coordinates based on the previous adjustment results and residual analysis to minimize the overall reprojection error. The final adjustment calculation can provide accurate image exterior orientation information and ground point coordinates, providing high-quality data support for 3D reconstruction and surveying. The results of the adjustment calculation will include a comprehensive accuracy assessment report of control point residuals, tie point residuals and image point residuals.

[0206] Systematically analyze the adjustment results and accuracy reports, verify whether the indicators meet the technical specifications, optimize and adjust them when necessary, and finally output high-quality exterior orientation elements and ground point coordinate data. This process includes checking indicators such as control point residuals, tie point residuals, and image point residuals to ensure that they are within the specified range. If problems are found, optimization adjustments may be required to improve the accuracy of the adjustment results. The high-quality exterior orientation elements and ground point coordinate data finally output will be used to generate accurate 3D models or other surveying and mapping results to ensure the accuracy and reliability of the project.

[0207] Through the above detailed steps, the photogrammetry software can effectively perform aerial triangulation calculations and accurately determine the exterior orientation elements of the image and the coordinates of the ground points, thus providing a solid data foundation for high-precision three-dimensional reconstruction and mapping.

[0208] Furthermore, step 4 includes the following steps.

[0209] Perform a comprehensive quality check on the high-density 3D point cloud data and optimized image data, including point cloud density, distribution uniformity, and image clarity. Quality inspection is a key step in ensuring the accuracy of 3D model construction. First, check the high-density 3D point cloud data to confirm the density, distribution uniformity, and quality of the point cloud. This includes checking whether the point cloud covers the entire survey area and whether the density of the point cloud is sufficient to support the capture of details and accurate modeling. High-density point cloud data helps to generate more refined models, but it may also bring processing challenges such as noise and data redundancy. Therefore, it is necessary to ensure the uniformity and quality of the point cloud data so that subsequent processing can obtain accurate results. At the same time, perform a clarity check on the optimized image data to ensure that the image has no blur, shadows, or other quality issues. This step ensures that the image can provide sufficient details and realism when mapped onto the surface of the 3D model. The quality of the image directly affects the effect of texture mapping, so high-quality image data is the basis for building a realistic 3D model.

[0210] Noise filtering is performed on high-density 3D point cloud data to remove outliers and low-quality points. Point cloud data often contains noise and outliers, which may be caused by environmental factors, equipment failures, or other error sources. Noise filtering is an important step to improve the quality of point cloud data. These outliers and low-quality points can be removed by applying filtering algorithms (such as statistical outlier removal, voxel grid method, etc.). This step not only improves the clarity of the data, but also reduces the computational burden of subsequent modeling and analysis. Noise filtering makes the point cloud data cleaner and ensures the accuracy and reliability of the generated model.

[0211] Based on geometric features and spectral information, high-density three-dimensional point cloud data is preliminarily classified into ground points, building points, and vegetation points. Classification is an important part of point cloud data processing, which can classify data into different categories for more targeted processing. By analyzing the geometric features (such as height, shape) and spectral information (such as reflectivity, color) of the point cloud, the point cloud data can be preliminarily classified into ground points, building points, and vegetation points. Ground points are used to build digital terrain models, building points are used to generate building models, and vegetation points are used to reconstruct vegetation areas. The accuracy of classification affects the accuracy of subsequent model construction, so advanced classification algorithms, such as deep learning-based classifiers, are needed to improve the classification effect.

[0212] Use ground points to build a high-precision digital terrain model, and use Kriging to achieve smoothness and detail fidelity of the terrain surface. Kriging is a common geostatistical method used for interpolation and surface modeling. Through Kriging, discrete ground point data can be converted into a continuous terrain surface, and the smoothness and detail fidelity of the terrain can be achieved. This method can effectively handle the spatial variability in point cloud data and generate a high-precision terrain model. The application of Kriging ensures the smoothness of the terrain model while retaining the real details and characteristics of the terrain.

[0213] Based on the classified high-density 3D point cloud data and the original image, multi-view stereo matching technology is used to reconstruct the 3D models of buildings and major landforms. By combining the classified point cloud data with the original image, the 3D models of buildings and major landforms can be accurately reconstructed using stereo matching technology. This process involves acquiring images from multiple viewpoints and generating detailed 3D models through stereo matching algorithms. Multi-view stereo matching can capture complex structures and details, making the generated 3D model more accurate and realistic.

[0214] Project high-resolution original images onto the surface of a 3D model to generate realistic textures, and use multi-view texture fusion technology to process texture differences shot at different angles, eliminate seams and shadow problems, and improve texture quality. Projecting high-resolution original images onto the surface of a 3D model can generate realistic texture effects. However, images shot at different angles may have texture differences, such as lighting changes and color inconsistencies. Multi-view texture fusion technology can handle these differences, eliminate seams and shadow problems, and improve texture quality. This includes smoothing the texture and adjusting the color, making the final texture effect more consistent and natural.

[0215] For trees and shrubs, point cloud clustering method is used for 3D reconstruction; for bridges and tower structures, professional knowledge base and parametric modeling technology are combined to improve modeling accuracy and realism.

[0216] The overall 3D model is topologically checked and repaired, and the geometric accuracy and texture quality of the overall 3D model are evaluated by comparison with control points and field measurement data. Local adjustments and optimizations are performed when necessary, ultimately forming a high-quality, high-precision multi-view 3D model.

[0217] Through the above detailed steps, the multi-view 3D model finally generated is not only guaranteed in geometric accuracy, but also can truly reflect the actual situation of the survey area in terms of visual effect, meeting the needs of high-quality 3D modeling.

[0218] Furthermore, a high-precision digital terrain model is constructed using the ground points, and the Kriging method is used to achieve smoothness and detail fidelity of the terrain surface, including the following steps.

[0219] The classified ground points are processed by outlier removal and density optimization.

[0220] The processed ground points are preliminarily interpolated using the triangular irregular network algorithm to construct the initial digital terrain model.

[0221] The spatial autocorrelation analysis of the initial terrain model was carried out to establish a variogram model describing the terrain characteristics.

[0222] Based on the established variogram model, the ordinary Kriging algorithm is applied for high-precision terrain interpolation.

[0223] Combined with the original high-density point cloud data, key terrain features are locally optimized to improve the expression accuracy of key terrain features and generate a digital terrain model.

[0224] Perform a comprehensive quality assessment and validation of the digital terrain model, making local adjustments where necessary to achieve the best balance of smoothness and detail fidelity.

[0225] Through the above adjustments, the smoothness and detail fidelity of the model can be optimized to better meet the actual situation and project requirements. Finally, after a comprehensive quality assessment and necessary adjustments, the generated digital terrain model will have high accuracy, high smoothness and detail fidelity, providing a reliable data foundation for subsequent applications and analysis.

[0226] Furthermore, based on the classified high-density three-dimensional point cloud data and the original image, a multi-view stereo matching technology is used to reconstruct the three-dimensional models of buildings and main landforms, including the following steps.

[0227] Using the known exterior orientation elements and control point information, the classified high-density 3D point cloud data and the original image are accurately registered to the same coordinate system. This step ensures spatial consistency between different data sources and avoids reconstruction errors caused by mismatched coordinate systems. Each image is projected onto the point cloud data using the image's interior and exterior orientation elements, and the data is accurately aligned using least squares fitting or other optimization methods. The accuracy of the registration directly affects the results of subsequent steps, so fine adjustments are usually required during the registration process to ensure high-precision overlap of the image and point cloud.

[0228] Based on the classified point cloud data, the region growing algorithm is applied to extract the outlines of buildings and major objects; and the image texture and color information are combined to optimize the boundary definition and generate a region of interest mask. The region growing algorithm is a commonly used image segmentation method used to extract the outlines of buildings and major objects from point cloud data. The algorithm starts from a seed point and gradually expands the region according to the similarity of the points (such as height, color, etc.) until the complete outline of the target object is covered. Based on the point cloud data, the extracted boundary definition can be optimized by combining the texture and color information of the image. This process involves comparing the point cloud data with the image texture to generate a more accurate region of interest (ROI) mask. This mask can effectively isolate the building and object areas, avoid interference with the background or irrelevant areas, and provide a clear target area for subsequent reconstruction.

[0229] For each target to be reconstructed, the best multi-view image combination is selected based on the spatial position and camera parameters, and the feature description algorithm is applied to extract and match the feature points between the multi-view images to establish a dense image correspondence. Selecting the best multi-view image combination is a key step to improve the quality of 3D reconstruction. First, according to the spatial position and camera parameters of the target to be reconstructed, the image combination with the best viewing angle and the highest coverage is selected. Then, the feature description algorithm (such as SIFT, SURF, etc.) is used to extract and match the feature points of these images to establish a dense image correspondence between the images. Feature point extraction involves identifying salient points in the image, and the feature description algorithm generates a descriptor for each point so that it can be matched between different images. The matching process is achieved by calculating the similarity of the feature point descriptors to ensure the establishment of a reliable image correspondence and provide accurate input for subsequent dense matching.

[0230] Based on the established feature point correspondence, considering multi-view geometric constraints and local smoothness constraints, a semi-global matching multi-view stereo algorithm is used for dense matching, and a depth map corresponding to each view is generated to describe the three-dimensional structure of the scene. Dense matching is the core step in reconstructing a three-dimensional model, which involves generating a depth map through the established feature point correspondence. The semi-global matching (SGM) algorithm is used for dense matching. This algorithm generates an accurate depth map through a global optimization method based on multi-view geometric constraints and local smoothness constraints. The depth map describes the scene depth information of each view and is the basic data for generating a three-dimensional model. The semi-global matching algorithm calculates the depth information of each pixel in the image pairs at different viewpoints by optimizing the cost function (usually the parallax cost function). The algorithm effectively balances the matching accuracy and computational complexity, and can handle complex parallax scenes and texture changes.

[0231] The multi-view depth map is projected into three-dimensional space to form a dense point cloud. The iterative closest point algorithm is used to register and fuse the dense point cloud with the original high-density three-dimensional point cloud, and the statistical filter and moving least squares method are applied to optimize the fused point cloud to improve the quality of the point cloud. After the depth map is projected into three-dimensional space to generate a dense point cloud, it needs to be registered and fused so that it can be merged with the original high-density point cloud data. The dense point cloud can be registered with the original point cloud using the iterative closest point (ICP) algorithm, and high-precision alignment can be achieved by minimizing the distance error between the point clouds. After registration, the fused point cloud is optimized using the statistical filter and the moving least squares method. The statistical filter is used to remove isolated noise points, and the moving least squares method is used to smooth the point cloud surface and eliminate local errors and noise that may exist in the point cloud. These steps can significantly improve the quality and accuracy of the point cloud, making it more suitable for subsequent modeling work.

[0232] Based on the optimized point cloud, the plane fitting and contour extraction algorithms are used to reconstruct the 3D models of buildings and major objects. For regular buildings, a model-based reconstruction method is used; for complex objects, a Poisson surface reconstruction algorithm is used to generate a fine triangular mesh model. Reconstructing the 3D model is the last step based on the optimized point cloud data. For regular buildings (such as buildings, bridges, etc.), a model-based reconstruction method, such as a plane fitting algorithm, can be used to generate a 3D model by extracting the plane structure of the building. This method accurately reconstructs the geometric shape of regular buildings by fitting lines and faces. For complex objects (such as trees, mountains, etc.), the Poisson surface reconstruction algorithm is an effective method. Poisson reconstruction can handle complex surface shapes and detailed features by generating a triangular mesh model to generate a high-quality 3D model. This method can retain the fine structure of complex objects and generate a realistic 3D model by modeling the global geometric features of the point cloud.

[0233] Through the above steps, the classified 3D point cloud data and the original image can be effectively combined to generate high-quality 3D models of buildings and major objects. This process covers the complete workflow from data registration, feature extraction, dense matching to model reconstruction, ensuring the accuracy and details of the final 3D model.

[0234] Furthermore, step 5 includes the following steps.

[0235] Preprocess the multi-view 3D model, including noise filtering and data simplification. After the 3D model is built, preprocessing is first required to improve the data quality. This process includes noise filtering and data simplification. The purpose of noise filtering is to remove abnormal points in the point cloud data due to measurement errors or other factors. Commonly used methods include statistical outlier removal (SOR) and radius outlier removal (R) . These methods identify and remove noise points by statistically analyzing the local features of the point cloud, thereby improving the accuracy and credibility of the model. Data simplification involves reducing the computational complexity by reducing the density of the point cloud while retaining key geometric features as much as possible. Simplification methods include mesh resampling and point cloud downsampling. This not only helps to reduce processing time, but also improves the efficiency of subsequent analysis and processing. The preprocessed model is cleaner and more efficient, laying the foundation for subsequent feature extraction and parameter calculation.

[0236] The multi-view 3D model is segmented into different semantic regions using an algorithm based on semantic segmentation. The semantic segmentation algorithm is used to segment different regions in the 3D model into category regions with the same meaning. This process is usually implemented using deep learning techniques, such as convolutional neural networks (CNN) or graph convolutional networks (GCN). Semantic segmentation can not only distinguish different objects such as buildings, roads, and vegetation, but can also be refined to specific object types, such as distinguishing road types (such as main roads and branch roads). The purpose of semantic segmentation includes improving the model's ability to recognize details, making subsequent feature extraction more accurate. Through the segmentation model, different analysis methods can be applied in different semantic regions, so that each object can be processed specifically. For example, different algorithms can be used to extract and calculate their feature parameters for building areas and vegetation areas.

[0237] Deep learning algorithms are used to identify and extract building outlines, and automatically calculate the building's height, area, volume, and roof type parameters, while analyzing the building's orientation and density distribution. The application of deep learning algorithms in building outline extraction can significantly improve accuracy. By training neural networks (such as U-Net, Mask R-CNN) to identify building outlines, the boundaries and features of buildings can be automatically extracted. These algorithms are capable of handling complex building structures and accurately locating the building's outline in multi-view images. Once the building outline is extracted, the building's height, area, volume and other geometric parameters can be automatically calculated. The calculation of these parameters is usually based on the elevation information of the point cloud and the plane outline of the building. The roof type can be determined by analyzing the surface features of the building, such as identifying types such as flat roofs and pitched roofs. In addition, analyzing the building's orientation (through the building's azimuth) and density distribution (through the distribution of the number of buildings in a certain area) can provide important information about urban structure and land use. This information is of great reference value for urban planning, land management, and architectural design.

[0238] The principal component analysis method is applied to identify and extract the road network, automatically calculate the road width, road length and road slope, and analyze the connectivity and density of the road network. Principal component analysis (PCA) is used to extract the road network from 3D point cloud data. PCA can identify the main directions and structural features of the road and classify the roads. By performing principal component analysis on the point cloud data, the center line of the road can be extracted, and the width, length and slope of the road can be calculated based on this information. The calculation of road width is usually achieved by analyzing the point cloud data of the road boundary, while the road length can be obtained by calculating the total length of the road center line. The calculation of road slope is based on the elevation change of the road surface. Further analysis includes the connectivity and density of the road network, which can be achieved by constructing a road network graph (Graph) to analyze the connection relationship between roads and their distribution.

[0239] Use point cloud classification algorithms to distinguish different types of vegetation, calculate vegetation coverage, tree height and crown size, and perform vegetation density analysis. Point cloud classification algorithms are used to distinguish vegetation parts in point cloud data into different types (such as trees, shrubs, etc.). Common point cloud classification methods include feature-based classification and deep learning classification. Vegetation can be effectively classified by analyzing the height, density and spectral characteristics of point clouds. Calculating vegetation coverage refers to calculating the proportion of area occupied by vegetation in a certain area. Tree height and crown size can be obtained by analyzing the vertical and horizontal distribution of trees in point cloud data. Vegetation density analysis involves calculating the density of vegetation points in a certain area to understand the growth and distribution of vegetation.

[0240] Based on elevation and texture features, water system elements are identified and extracted, water body area, river length and flow direction are automatically calculated, and the catchment area is analyzed. The identification of water system elements depends on the analysis of elevation and texture features. Water bodies are usually shown as lower areas in elevation models, and their texture features may be different from the surrounding terrain. By analyzing these features, water body areas such as lakes and rivers can be automatically extracted. The area of ​​water bodies is calculated by measuring the two-dimensional projection of the water body area, while the length of the river can be obtained by extracting the center line of the river and calculating its total length. Flow direction analysis involves calculating the direction and pattern of water flow, usually based on elevation data for water flow simulation. The catchment area refers to the area where water flows converge within the basin, which is usually determined by analyzing the convergence path of the water flow.

[0241] Digital elevation models are used to analyze terrain features, including the calculation of slope, aspect, and terrain relief; at the same time, special landforms are identified and extracted, and terrain roughness and surface curvature are calculated. Digital elevation models (DEMs) are the basis of terrain analysis, through which various terrain features can be calculated. Slope represents the inclination of the surface, which is obtained by calculating the rate of change of elevation; aspect represents the direction of the surface slope, which is determined by analyzing the direction of elevation change. Terrain relief describes the degree of unevenness of the terrain, which is achieved by calculating the statistical characteristics of elevation changes. Special landforms (such as ridges and valleys) can be extracted by analyzing the terrain features in DEM data. The calculation of terrain roughness and surface curvature involves local fitting and analysis of elevation data to evaluate the smoothness and curvature of the surface. These analysis results help to understand the structure and change characteristics of the terrain.

[0242] Identify and extract artificial facilities including bridges, tunnels and transmission lines, calculate the geometric parameters of artificial facilities, and analyze their spatial distribution characteristics. The identification and extraction of artificial facilities involves locating structures such as bridges, tunnels and transmission lines from point cloud data and images. Feature detection and classification algorithms are usually used to identify the geometric characteristics of these facilities. The geometric parameters (such as length, width, height) of bridges, tunnels and transmission lines can be calculated by analyzing their point cloud data and image information. Spatial distribution feature analysis involves studying the distribution pattern and layout of artificial facilities in geographic space. For example, the distribution of bridges and tunnels can reveal the structure of the transportation network, while the layout of transmission lines can reflect the coverage of the power network. These analyses provide important data support for infrastructure management and urban planning.

[0243] The above steps involve the whole process from 3D model preprocessing to feature extraction and parameter calculation. By comprehensively applying noise filtering, semantic segmentation, deep learning, point cloud classification, geometric analysis and other technologies, it is possible to comprehensively and accurately extract the features of objects in multi-view 3D models, providing detailed data support for further application and analysis. The effective combination of these technologies makes 3D modeling and feature extraction in complex environments more efficient and reliable.

[0244] Furthermore, the building height H 建筑 Calculated by the following formula: 建筑 =Z roof -Z ground , where Z roof is the average elevation of the building roof, extracted from the point cloud data; Z ground is the average elevation of the building ground, extracted from the digital terrain model.

[0245] Building area A 建筑 It is calculated through binary mask, that is, the sum of the pixel areas of all buildings. The specific calculation formula is: A 建筑 =∑ (x＾,y＾)∈M Area(x^,y^), where M is the binary mask of the building, (x^,y^)∈M represents all pixels with a value of 1 in the mask; Area(x^,y^) is the actual ground area corresponding to the pixel (x^,y^).

[0246] The volume of the building V 建筑 It is obtained by integrating the height of the building in the mask area. The specific calculation formula is expressed as: V 建筑 =∑ (x＾,y＾)∈M H(x^,y^)·Area(x^,y^), where H(x^,y^) is the height of the buildings at each point in the mask area.

[0247] Roof type parameters T 屋顶 The deep learning classification algorithm is used to classify the extracted building roof features, which can be expressed as: T 屋顶 =f roof (I roof ), where I roof is the extracted building roof image or feature map; f roof is a deep learning based classifier for determining roof type.

[0248] Building orientation θ 建筑 It is calculated by extracting the main axis direction of the building outline, and the formula is: θ 建筑 =atan2(v y ,v x ), where vy and v x It is the direction vector component of the contour principal axis, which is calculated by principal component analysis method.

[0249] Building density ρ 建筑 It is calculated by the number of buildings per unit area. The specific formula is: 建筑 =A total / A region , where A total is the total area of ​​all buildings in the survey area; A region is the total area of ​​the survey area.

[0250] Road width W 道路 It is obtained by calculating the distance between road boundaries along the orthogonal direction of the road direction. The specific calculation formula is: W 道路 =1 / NB·∑ h=1 NB d h , where NB is the number of points sampled along the road direction; d h is the road width at the hth sampling point, calculated by the distance between road boundary points.

[0251] Road length L 道路 It is calculated by integrating along the center line of the road, and the formula is: L 道路 =∑ o=1 m √((x o -x o-1 ) 2 +(y o -y o-1 ) 2 ), where m is the number of sampling points on the road centerline; (x o ,y o ) is the coordinate of the oth sampling point; (x o-1 ,y o-1 ) are the coordinates of the o-1th sampling point.

[0252] Road slope S 道路 It is calculated by the ratio of the elevation difference between the starting point and the end point of the road to the horizontal distance. The formula is: S 道路 =(Z end -Z start ) / L 道路 ×100%, where Z star , Z end are the elevations of the start and end points of the road respectively.

[0253] Road network connectivity C 道路 It is represented by calculating the connected components of the road network: C 道路 =2E 道路 / (N 道路 (N 道路 -1)), where E 道路 is the number of edges in the road network; N 道路 is the number of intersections in the road network.

[0254] Road density D 道路 Calculated by the total length of roads within a unit area: D 道路 =L total / A region , where L total is the total length of all roads within the specified area.

[0255] The vegetation coverage rate is the ratio of the area covered by vegetation in the survey area to the total area of ​​the survey area.

[0256] Tree height H 树木 It is the difference between the maximum height of vegetation points and the minimum height of ground points in the point cloud data.

[0257] Tree crown size C size It is expressed by the horizontal projection area of ​​the tree crown: C size =(D max / 2)×(D min / 2), where D max and D min are the maximum and minimum diameters of the crown, obtained by analyzing the horizontal projection of the crown point cloud.

[0258] Vegetation density D veg It is the amount of vegetation per unit area or the proportion of vegetation coverage.

[0259] Water area A water The calculation formula is: water =∑ (p,q)∈R I(W(p,q))·A pixel , where R is the coordinate set of the survey area; I(W(p,q)) is an indicator function. When W(p,q) is "water body", I(W(p,q))=1, otherwise I(W(p,q))=0; A pixel is the actual geographic area corresponding to each pixel.

[0260] River length L river The calculation formula is: L river =∑ i=1 NR-1 √((x e+1 -x e ) 2 +(y e+1 -y e ) 2 ), where (x e,y e ) represents the coordinates of the e-th point on the river path; NR is the number of all points on the river path.

[0261] The formula for calculating the river flow direction is: D flow (α,β)=arctan((E(α2,β2)-E(α1,β1)) / √((α2-α1) 2 +(β2-β1) 2 )), where D flow (α, β) is the direction of the river, that is, the direction angle of the water flow from point (α1, β1) to point (α2, β2); E(α1, β1) and E(α2, β2) are the elevations of point (α1, β1) and point (α2, β2) respectively.

[0262] Furthermore, step 6 includes the following steps.

[0263] The data formats of multi-view 3D models, extracted features and calculated terrain parameters are unified and the coordinate system is converted to ensure that all data are in the same coordinate system and meet the data standards of the natural resource management system. The data from different sources are unified and the coordinate system is converted, including converting the 3D model, feature data and terrain parameter data into a unified data format, such as GeoTIFF, Shapefile, etc., to ensure that the data can be correctly identified and processed by the natural resource management system (NRMS). Coordinate system conversion is a key step, the purpose of which is to align all data into a common geographic coordinate system. Typically, 3D models and feature data use WGS84 or a local projected coordinate system (such as UTM). Coordinate system conversion is achieved by using conversion tools in a geographic information system (GIS), such as the coordinate conversion function in ArcGIS or QGIS. This step ensures that all data are spatially consistent, so that subsequent data fusion and analysis can be performed on an accurate geographic basis.

[0264] Import the multi-view 3D model into the natural resource management system and register and fuse it with the existing 2D map data to form a registered and fused model. Importing the processed 3D model into the natural resource management system is a key step in data integration. In NRMS, the 3D model needs to be registered and fused with the existing 2D map data. This is usually achieved through geo-registration technology, which aligns the 3D model to the existing 2D map to ensure the accuracy of its spatial position and size. The registration process involves adjusting the geographic coordinates of the 3D model to match the geographic features on the 2D map. This may involve point-to-point registration, image registration technology, and the use of ground control points. The fused model, called the registered fused model, will contain stereo information extracted from the 3D model and data from the 2D map. Such fusion can provide a more comprehensive view of geographic information.

[0265] The automatically extracted features are superimposed on the registration and fusion model in the form of layers, and are associated and attribute-matched with the corresponding elements in the natural resource management system to form a feature overlay model. After the registration and fusion of the 3D model is completed, the automatically extracted features (such as buildings, roads, vegetation, etc.) are superimposed on the model in the form of layers. This process is achieved by importing the feature data into the layer management system in NRMS. Each layer represents a different type of feature, and the layers can be color-coded and identified according to different attributes. After the layers are superimposed, they need to be associated and attribute-matched with the corresponding elements in the natural resource management system. This means comparing the extracted features with the existing feature data in NRMS to ensure that the new data is correctly associated with the elements in the system and update the relevant attribute information. The feature overlay model formed in this way can fully display all relevant features and provide basic data for subsequent analysis.

[0266] Integrate the calculated terrain parameters into the feature overlay model to form a comprehensive geographic information model. The calculated terrain parameters (such as slope, aspect, terrain relief, etc.) need to be integrated into the feature overlay model. This process adds terrain parameters as additional information to the existing model to form a comprehensive geographic information model. The comprehensive model not only contains the three-dimensional model and land features, but also includes detailed terrain analysis results. The technical effects of integrating terrain parameters include providing a more comprehensive view of geographic information, allowing users to better understand the complexity and changes of the terrain. This comprehensive geographic information model can support more in-depth spatial analysis, such as the impact of terrain on building layout, terrain considerations in road planning, etc.

[0267] Use the data processing function of the natural resource management system to perform quality checks and topological relationship verification on the integrated geographic information model to ensure the consistency and integrity of the spatial data. After completing the data integration, it is crucial to use the data processing function of NRMS to perform quality checks and topological relationship verification. This includes checking whether the spatial data in the model is consistent and ensuring that there is no data duplication, omission or error. Topological relationship verification is to check whether the connection and adjacency of spatial objects (such as roads and buildings) in the model conform to the actual situation of the geographic space. This can be achieved through the topology checking tool in NRMS to ensure that all spatial data is logically correct and avoid topological errors such as overlaps, gaps or discontinuities.

[0268] Combined with historical data from the natural resource management system, spatiotemporal analysis and change detection are performed on the integrated geographic information model. Spatiotemporal analysis and change detection using historical data in the NRMS can reveal changing trends and patterns in geographic information. This process involves comparing the current integrated geographic information model with the same areas in historical data, detecting areas of change and types of change, such as land use changes, building expansion, etc. Spatiotemporal analysis helps understand the evolution of geographic features and supports dynamic monitoring and planning decisions. Change detection can provide important information about environmental changes, urban expansion, the impact of natural disasters, etc., providing a scientific basis for decision makers.

[0269] By utilizing the visualization function of the natural resource management system, a variety of results including three-dimensional terrain maps, thematic layers and statistical reports can be generated. The visualization function provided by the natural resource management system can display the comprehensive geographic information model in various forms, including three-dimensional terrain maps, thematic layers and statistical reports. The three-dimensional terrain map provides an intuitive terrain view, allowing users to observe terrain features from multiple perspectives. Thematic layers can display detailed information on different land features, such as building distribution maps, road network maps, etc. Statistical reports provide quantitative data, such as regional area, number of buildings and other statistical information. These visualization results help to better understand and analyze geographic data and support application needs in different fields, such as urban planning, environmental protection, etc.

[0270] The integrated geographic information model and generated results are packaged to form a complete integrated space-ground-air mapping database, and ensure real-time synchronization and update mechanism with the natural resource management system. The integrated space-ground-air mapping database includes all information from 3D models, land features, terrain parameters to visualization results, and ensures real-time synchronization and update mechanism with the natural resource management system. The real-time synchronization mechanism ensures that the data in the database is always up-to-date and can reflect the latest geographic changes and data updates. This is crucial to ensure the accuracy and timeliness of information, allowing users to make decisions and analyses based on the latest data.

[0271] The various links of step 6 involve processes ranging from data format unification and coordinate system conversion to registration and fusion of three-dimensional models, feature extraction, parameter integration, quality inspection, spatiotemporal analysis, visualization and data packaging. These steps ensure the consistency, integrity and practicality of geographic data, and provide a solid foundation for the formation and application of comprehensive geographic information models. These technologies and steps combined can achieve comprehensive management and analysis of complex geographic information and support efficient natural resource management and decision-making.

[0272] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, a person skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for integrated space-ground-air mapping based on oblique photography three-dimensional mapping, characterized in that: The following steps are involved: Step 1: Develop a comprehensive oblique photogrammetry program to ensure complete coverage of complex terrain; Step 2: Deploy industrial RTK drones and consumer drones to work together to achieve a combination of large-scale coverage and local fine collection to obtain high-quality oblique photography image data; Step 3: Integrate RTK positioning data and ground control point data as control information to perform aerial triangulation encryption processing on the oblique photography images; Step 4: construct a high-quality, high-precision multi-view 3D model; Step 5: Automated ground feature extraction and recognition based on the multi-view 3D model, as well as automatic calculation of terrain element parameters; Step 6: Integrate the multi-view 3D model, land feature characteristics and terrain element parameters with the natural resource management system to generate an integrated air-ground-space mapping; Step 3 includes the following steps: Step 3.1, extract the precise RTK positioning information of each image from the flight data collected by the industrial RTK UAV, including the longitude, latitude, elevation and attitude angle data at the time of shooting; Step 3.2, integrate RTK positioning data and ground control point data into the same coordinate system to establish a unified control point database; Step 3.3, pre-process the acquired oblique photography images, including image evenness, noise removal and blur operations to improve image quality; at the same time, extract the EXIF ​​information of each image, match it with the RTK positioning data, and establish a preliminary correspondence between the image and the position; Step 3.4, using computer vision algorithms, automatically identify and match feature points between adjacent images, generate a large number of connection points and construct an image network for the entire survey area, specifically including the following steps: apply computer vision algorithms to each image to identify and extract feature points with significance and stability; for each extracted feature point, calculate its feature descriptor to describe the unique properties of the area around the feature point; use the nearest neighbor search algorithm to perform preliminary matching of feature points between adjacent images; apply the random sample consistency algorithm to perform geometric verification on the preliminary matched feature point pairs to effectively remove false matches and retain only matching point pairs that meet the expected geometric relationship; based on the verified matching points, construct an image connection network for the entire survey area, and calculate the relative orientation between each pair of adjacent images to provide initial values ​​for subsequent aerial triangulation; perform global optimization on the entire image network, including using the bundle adjustment algorithm to jointly optimize all connection points and camera parameters to minimize the reprojection error, so as to further improve the accuracy and consistency of the connection points; Step 3.5, using RTK positioning data, ground control point data and the image network of the entire survey area as input, use professional photogrammetry software to perform aerial triangulation calculations; Step 3.6, conduct a comprehensive error analysis on the aerial triangulation results, including control point residuals, tie point residuals and image network adjustment accuracy; Step 3.7, based on the error analysis results, remove gross error points, adjust weights, and perform multiple iterative calculations until the predetermined accuracy requirements are met; Step 3.8, based on the optimized aerial triangulation results, update the precise position and attitude information of each image to obtain optimized image data; Step 3.9, using the optimized image data and dense matching algorithm to generate high-density three-dimensional point cloud data; Step 4 includes the following steps: Perform comprehensive quality checks on high-density 3D point cloud data and optimized image data, including point cloud density, distribution uniformity, and image clarity; Filter noise points on high-density 3D point cloud data to remove abnormal points and low-quality points; Based on geometric features and spectral information, high-density 3D point cloud data are preliminarily classified into ground points, building points and vegetation points; Use ground points to build a high-precision digital terrain model, and use Kriging to achieve smoothness and detail fidelity of the terrain surface; Based on the classified high-density 3D point cloud data and the original image, the multi-view stereo matching technology is used to reconstruct the 3D models of buildings and major landforms, including the following steps: using the known exterior orientation elements and control point information, the classified high-density 3D point cloud data and the original image are accurately registered to the same coordinate system; based on the classified point cloud data, the regional growing algorithm is used to extract the outlines of buildings and major landforms; and the image texture and color information are combined to optimize the boundary definition and generate a mask of the region of interest; for each target to be reconstructed, the best multi-view image combination is selected based on the spatial position and camera parameters, and the feature description algorithm is used to extract and match the feature points between the multi-view images to establish a dense image correspondence relationship; based on the established feature point pair The corresponding relationship is considered, and the multi-view geometric constraints and local smoothness constraints are considered. A semi-global matching multi-view stereo algorithm is used for dense matching, and a depth map corresponding to each view is generated to describe the three-dimensional structure of the scene; the multi-view depth map is projected into the three-dimensional space to form a dense point cloud, and the iterative closest point algorithm is used to align and fuse the dense point cloud with the original high-density three-dimensional point cloud, and the fused point cloud is optimized by applying statistical filtering and moving least squares method to improve the quality of the point cloud; based on the optimized point cloud, the plane fitting and contour extraction algorithms are used to reconstruct the three-dimensional models of buildings and main objects. For regular buildings, a model-based reconstruction method is used; for complex objects, a Poisson surface reconstruction algorithm is used to generate a fine triangular mesh model; Project high-resolution original images onto the surface of the 3D model to generate realistic textures, and use multi-view texture fusion technology to process texture differences shot at different angles, eliminate seams and shadows, and improve texture quality; For trees and shrubs, point cloud clustering method is used for 3D reconstruction; for bridges and tower structures, professional knowledge base and parametric modeling technology are combined to improve modeling accuracy and realism; Perform topological inspection and repair on the overall 3D model, evaluate the geometric accuracy and texture quality of the overall 3D model by comparing it with control points and field measurement data, and make local adjustments and optimizations to ultimately form a high-quality, high-precision multi-view 3D model; Step 5 includes the following steps: Preprocess the multi-view 3D model, including noise filtering and data simplification; Use a semantic segmentation-based algorithm to segment the multi-view 3D model into different semantic regions; Use deep learning algorithms to identify and extract building outlines, and automatically calculate building height, area, volume, and roof type parameters, while analyzing building orientation and density distribution; Apply principal component analysis to identify and extract road networks, automatically calculate road width, road length and road slope, and analyze the connectivity and density of road networks; Use point cloud classification algorithms to distinguish different types of vegetation, calculate vegetation coverage, tree height and crown size, and perform vegetation density analysis; Based on elevation and texture features, it identifies and extracts water system elements, automatically calculates water area, river length and flow direction, and analyzes catchment area; Use digital elevation models to analyze terrain features, including calculations of slope, aspect, and terrain relief; identify and extract special landforms, and calculate terrain roughness and surface curvature; Identify and extract artificial facilities including bridges, tunnels and transmission lines, calculate the geometric parameters of artificial facilities, and analyze their spatial distribution characteristics; Step 6 includes the following steps: Unify the data format and coordinate system of multi-view 3D models, extracted features and calculated terrain parameters to ensure that all data are in the same coordinate system and meet the data standards of the natural resource management system; Import the multi-view 3D model into the natural resource management system and register and fuse it with the existing 2D map data to form a registered and fused model; The automatically extracted features of the objects are superimposed on the registration fusion model in the form of layers, and are associated and attribute matched with the corresponding elements in the natural resource management system to form a feature superposition model; Integrate the calculated terrain parameters into the feature overlay model to form a comprehensive geographic information model; Using the data processing function of the natural resource management system, the quality of the integrated geographic information model is checked and the topological relationship is verified to ensure the consistency and integrity of the spatial data; Combine historical data from natural resource management systems to conduct spatiotemporal analysis and change detection on integrated geographic information models; Utilize the visualization function of the natural resource management system to generate a variety of results including three-dimensional topographic maps, thematic layers and statistical reports; The comprehensive geographic information model and generated results are packaged to form a complete integrated air-space-ground mapping database, and ensure a real-time synchronization update mechanism with the natural resource management system.

2. The method for integrated space-ground-air mapping based on oblique photography three-dimensional mapping according to claim 1, characterized in that: Step 1 includes the following steps: Collect existing geographic information data of the survey area and conduct preliminary analysis to understand the terrain characteristics, distribution of land objects and their complexity; Detailed analysis of terrain features using GIS software to calculate and visualize slope, aspect, and elevation change factors, thereby identifying flat areas, steep areas, and complex terrain areas within the survey area; According to the results of terrain feature analysis, the survey area is divided into several blocks with similar features; For each block, flight routes with different altitudes and overlaps are designed according to its terrain features to ensure the integrity and accuracy of data collection; Determine the number and distribution of ground control points based on the survey area and terrain complexity to improve surveying and mapping accuracy; Determine the camera parameters and shooting parameters according to the characteristics of each block and mapping requirements to obtain high-quality oblique photography data; The flight route, control point layout and shooting parameters are evaluated through simulation software, and fine-tuning and optimization are carried out based on the evaluation results to ensure the scientific nature and operability of the final surveying and mapping plan.

3. The method for integrated space-ground-air mapping based on oblique photography three-dimensional mapping according to claim 1, characterized in that: Step 2 includes the following steps: Select industrial-grade RTK drones and consumer-grade drones based on the characteristics of the survey area and mission requirements, and ensure that the performance parameters of the selected equipment meet the mission requirements; Based on block division and flight route design, plan the operating range and flight routes of industrial RTK drones and consumer drones; Set up an RTK base station in the survey area to ensure good communication with the drone; at the same time, configure the mobile station of the industrial-grade RTK drone to obtain high-precision location information; According to the predetermined flight route and shooting parameters, the selected industrial-grade RTK UAV is used to collect oblique photography data; at the same time, the flight status, image acquisition and RTK signal quality are monitored in real time to ensure the integrity and accuracy of data collection; Based on preliminary results collected by industrial RTK drones and on-site observations, identify areas that require detailed collection; Use consumer drones to conduct supplementary collection of areas that require detailed collection, including close-up photography of complex terrain, multi-angle collection of special objects, or additional coverage of shadow areas; during the flight, conduct real-time quality checks on the acquired image data, including image clarity, overlap, and color uniformity. If problems are found, adjust flight parameters or re-collect in a timely manner; Perform cross-validation flights at the interface of data collected by industrial RTK drones and consumer drones to detect and eliminate data gaps or inconsistent areas.

4. The method for integrated space-ground-air mapping based on oblique photography three-dimensional mapping according to claim 1, characterized in that: Apply computer vision algorithms to each image to identify and extract significant and stable feature points, including the following steps: The gradient of each image is calculated to obtain the gradient amplitude and direction. The specific calculation formula is: M(x,y)=√{(I x (x,y) 2 +(I y (x,y) 2 },θ(x,y)=arctan(I y (x,y) / I x (x,y)), where I x (x,y) and I y (x,y) represents the partial derivatives of the image in the x and y directions respectively; θ(x,y) represents the gradient direction at the image position (x,y); M(x,y) represents the gradient amplitude at the image position (x,y); Images of different scales are generated by Gaussian blur technology. The specific formula is: L(x,y,σ)=G(x,y,σ)*I(x,y), where I(x,y) represents the pixel value at the image position (x,y); L(x,y,σ) represents the Gaussian blurred image at scale σ; G(x,y,σ) is the Gaussian kernel function: G(x,y,σ)=1 / (2πσ 2 )·exp(-(x 2 +y 2 ) / (2σ 2 )), where * represents the convolution operation; The differential Gaussian image is calculated between images of different scales. The specific formula is: D(x,y,σ)=L(x,y,kσ)-L(x,y,σ), where D(x,y,σ) represents the differential Gaussian image at scale σ; L(x,y,kσ) represents the Gaussian blurred image at scale kσ; k is a multiple of the scale and is used to generate adjacent scale images; L(x,y,σ) represents the Gaussian blurred image at different scales; In the differential Gaussian images of different scales, local extreme points, i.e., feature points, are detected; Optimize the sub-pixel accuracy of the detected feature points to improve the stability of the feature points. The formula is: x^=-H -1 ▽D, where H is the Hessian matrix, ▽D is the gradient of the differential Gaussian image; x^ represents the optimized feature point position.

5. The method for integrated space-ground-air mapping based on oblique photography three-dimensional mapping according to claim 1, characterized in that: For each extracted feature point, calculate its feature descriptor, including the following steps: For each feature point, a fixed-size neighborhood window is selected; Calculate the gradient magnitude and gradient direction of each pixel in the neighborhood window; The neighborhood is divided into sub-regions. In each sub-region, the gradient direction histogram is calculated and the gradient direction is quantized into n direction buckets. The value of each bucket is the accumulation of the gradient amplitude in that direction. The specific formula is: i,j =∑ k=1 n Bin i,j,k , where H i,j is the total gradient amplitude of the k-th direction bucket in the (i, j)-th sub-region; Bin i,j,k It is the cumulative value of the gradient magnitude of all pixels in the sub-region projected into the k-th direction bucket; Concatenate the histograms of all sub-regions into a feature descriptor; Normalize the feature descriptor to improve robustness to lighting changes: f i ′=f i / ‖f i ‖, where ‖f i ‖ is the feature descriptor f i The Euclidean norm of i ′ is the normalized feature descriptor vector.

6. The method for integrated space-ground-air mapping based on oblique photography three-dimensional mapping according to claim 1, characterized in that: The relative orientation between each pair of adjacent images is calculated by the following steps: From the verified feature point matching, collect all matching point pairs between adjacent images; By matching point pairs, estimate the fundamental matrix F AB , the specific formula is: F AB =argmin F ∑ g=1 N ρ(x A,g T Fx B,g ), where x A,g and x B,g is the g-th pair of matching points; ρ is the reprojection error; F AB =argmin F It means to find a basic matrix F so that the subsequent error function is minimized; ∑ g=1 N ρ(x A,g T Fx B,g ) is the cost function, which represents the sum of the reprojection errors for all matching point pairs; ρ(·) is the squared error metric; Using the known camera intrinsic parameter matrix K A and K B , the fundamental matrix F AB Converted to the essential matrix, the formula is: E AB :E AB =K B T F AB K A , where K B T represents the transpose of the camera intrinsic parameter matrix of image B; The essential matrix E AB Decomposed into the rotation matrix R AB and the translation vector t AB , the formula is: E AB =R AB [t AB ] × , where [t AB ] × is the translation vector t AB The antisymmetric matrix of ; Get the relative position, specifically expressed as: T AB =(R AB ,t AB ).

7. The method for integrated space-ground-air mapping based on oblique photography three-dimensional mapping according to claim 1, characterized in that: The RTK positioning data, ground control point data and the image network of the entire survey area are used as input, and professional photogrammetry software is used to perform aerial triangulation calculations, including the following steps: Import RTK positioning data, ground control point data and the image network of the entire survey area into professional photogrammetry software; Based on RTK positioning data and image EXIF ​​information, calculate the initial exterior orientation elements for each image, including position (X, Y, Z) and attitude angle (ω, φ, μ); Start the automatic aerial triangulation program, use the tie point information and initial exterior orientation elements in the image network to perform global adjustment operations, and minimize the reprojection error of the entire image network through iterative optimization; The software automatically identifies and precisely locates known ground control points on the corresponding images and matches them to the images; The software calculates the residuals of each connection point and control point, and automatically identifies and processes possible gross error points based on preset thresholds; Establish a dynamic weight allocation strategy based on the accuracy characteristics of different data sources; Perform the final adjustment calculation, integrate all RTK positioning data, ground control point data and constraints, and solve the optimal exterior orientation elements and ground point coordinates through an iterative algorithm; at the same time, generate a comprehensive accuracy assessment report including control point residuals, tie point residuals and image point residuals; Systematically analyze the adjustment results and accuracy reports, verify whether various indicators meet the technical specifications, and make optimization adjustments to ultimately output high-quality exterior orientation elements and ground point coordinate data.

8. The method for integrated space-ground-air mapping based on oblique photography three-dimensional mapping according to claim 1, characterized in that: Use ground points to build a high-precision digital terrain model, and use Kriging to achieve smoothness and detail fidelity of the terrain surface, including the following steps: Perform outlier removal and density optimization on classified ground points: The processed ground points are interpolated using a triangular irregular network algorithm to construct an initial digital terrain model; Conduct spatial autocorrelation analysis on the initial terrain model and establish a variogram model to describe the terrain characteristics; Based on the established variogram model, the ordinary Kriging algorithm is used for high-precision terrain interpolation; Combined with the original high-density point cloud data, key terrain features are locally optimized to improve the expression accuracy of key terrain features and generate a digital terrain model; Digital terrain models are thoroughly quality assessed and validated, and locally adjusted to achieve the best balance of smoothness and detail fidelity.

Images