Real scene three-dimensional construction method for confirming right of complex terrain area

Abstract

The invention relates to the field of natural resource right confirmation, and discloses a live-action three-dimensional construction method for right confirmation of a complex terrain area, and the method comprises the following steps: S1, carrying out the collaborative collection of multi-source data; s2, fusing the three-dimensional model and cadastral data; s3, automatically and intelligently extracting ownership boundaries; s4, dynamically binding and visualizing the ownership information; according to the method, the point cloud data and the image data are subjected to registration, de-noising and semantic segmentation processing, a real scene three-dimensional model with a layered cadastre label is constructed in combination with land block boundaries, ownership persons and land use type attributes in cadastre ownership investigation information, registration of the point cloud data and the image data is achieved through an improved ICP algorithm, and the registration accuracy of the point cloud data and the image data is improved. A weight coefficient, a regularization factor and a penalty matrix of point cloud curvature similarity are considered in a registration error function, the method is used for constraining registration distortion of a topographic relief area, and the improved registration algorithm can better adapt to characteristics of a complex topographic area and can effectively reduce registration errors especially in a place with large topographic relief.

Description

Technical Field

[0001] The present invention relates to the field of natural resource rights confirmation, and in particular to a real-scene three-dimensional construction method for rights confirmation in areas with complex terrain. Background Art

[0002] With the rapid development of social economy and the increasing scarcity of land resources, the importance of land rights confirmation has become increasingly prominent, especially in areas with complex terrain, such as mountainous areas, hilly areas, urban-rural fringe areas, etc. The clear definition of land ownership relations is of great significance to maintaining social stability, promoting economic development, and protecting farmers' rights and interests. In recent years, with the rapid development of technologies such as drones, oblique photography, and lidar, real-scene 3D modeling technology has gradually matured and has been widely used in surveying and mapping, planning, land and other fields. Real-scene 3D models can truly and intuitively reflect the spatial information of land objects and topography, providing new technical means for land rights confirmation work in areas with complex terrain.

[0003] According to Chinese Patent Publication No. CN118135137A, a method, system, and storage medium for constructing a real-world 3D scene for land title confirmation in complex terrain areas are disclosed. The method obtains terrain information from a pre-confirmed area by surveying it, then formulates a flight survey route based on the terrain information, and simultaneously sets corresponding control points within the pre-confirmed area as reference points for the pre-confirmed area's topography and other features. The pre-confirmed area is then framed in two dimensions based on the flight survey route. The real-world scene is then constructed using three-dimensional data, combining the topography and other features of the pre-confirmed area actually obtained from the control points, to achieve high-precision 3D construction of a real-world scene in the pre-confirmed area. That is, it effectively solves the shortcomings of the above-mentioned existing technologies in low efficiency and insufficient accuracy in collecting real-life three-dimensional data in mountainous areas with complex terrain. However, this method still has shortcomings. When the existing real-life three-dimensional model construction method processes areas with complex terrain, especially areas with large terrain undulations, the registration accuracy of point cloud data and image data is insufficient and distortion is easily generated. This is mainly because traditional registration methods such as the ICP algorithm do not fully consider the impact of terrain undulations on registration, resulting in large registration errors at places where the terrain suddenly changes. Summary of the Invention

[0004] In view of the shortcomings of the existing technology, the present invention provides a real-scene three-dimensional construction method for property rights confirmation in areas with complex terrain, which solves the above problems.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for constructing a real-scene three-dimensional structure for land title confirmation in a complex terrain area, comprising the following steps:

[0006] S1. Collaborative multi-source data acquisition: Using drone oblique photography, airborne LiDAR, and ground-based mobile surveying equipment, we acquire multi-perspective images, high-precision point cloud data, and cadastral ownership survey information for the target area.

[0007] S2. Fusion of 3D Models and Cadastral Data: This involves registering, denoising, and performing semantic segmentation on point cloud and image data. This process combines the land parcel boundaries, owners, and land use type attributes from the cadastral ownership survey to construct a realistic 3D model with layered cadastral labels.

[0008] S3. Automated Intelligent Extraction of Ownership Boundaries: Based on the terrain features and cadastral labels of the real-world 3D model, a deep learning algorithm is used to identify surface attachments and terrain abrupt changes, automatically generating 3D vector data of ownership boundaries that comply with land title confirmation regulations.

[0009] S4. Dynamic binding and visualization of ownership information: This system associates 3D vector data of ownership boundaries with cadastral attribute information to generate a 3D electronic ownership certificate containing spatial coordinates, owner, area, and use. This system supports interactive ownership information query and conflict detection within the 3D model.

[0010] S5. Lightweight model and multi-terminal adaptation: Simplify the mesh and compress the data of the real-life 3D model to generate a multi-resolution model that supports WebGL, mobile terminals, and GIS platforms, and output 2D / 3D property maps that comply with legal formats.

[0011] Preferably, in step S1, the drone uses a five-lens camera for oblique photography, and the flight altitude is dynamically adjusted according to the terrain undulations to ensure that the image overlap is ≥80%.

[0012] Preferably, in step S2, the registration of the point cloud data and the image data is achieved by an improved ICP algorithm, and the registration error function is defined as:

[0013]

[0014] Among them, P i For LiDAR point cloud, Q j is the oblique photography point cloud, T is the rigid body transformation matrix, w i is the weight coefficient based on the curvature similarity of the point cloud, λ is the regularization factor, and Λ is the penalty matrix, which is used to constrain the registration distortion in the terrain undulating area. The registered point cloud is used as the input data for terrain mutation line detection in step S3.

[0015] Preferably, the detection of the terrain mutation line in step S3 is based on the calculation of the surface curvature of the three-dimensional model, and its main curvatures k1 and k2 are solved by the following formula:

[0016]

[0017] Among them, L, M, N are the second basic form coefficients of the three-dimensional surface. When the surface curvature change rate When the threshold value τ is exceeded, it is determined to be a terrain mutation line. The threshold value τ is negatively correlated with the vegetation coverage density of the semantic segmentation in step S2.

[0018] Preferably, in step S4, the area calculation adopts a three-dimensional projection correction algorithm, and the actual area A of the ownership surface 3D Calculated by the following formula:

[0019]

[0020] Among them, (x i ,y i ) is the plane coordinate of the boundary point, θ i is the local slope angle, Δz i is the elevation difference between adjacent points, Δl i is the horizontal distance, the local slope angle θ i Derived from the curvature data detected by the terrain abrupt change line in step S3.

[0021] Preferably, in step S5, the mesh simplification adopts an edge collapse algorithm based on quadratic error metric, and the error matrix Q of the vertex ν is ν Defined as:

[0022]

[0023] Among them, K p is the plane equation coefficient of the triangle p, and the folding cost is given by Δ(ν)=ν T Q ν ν is minimized and the simplified model retains the integrity of the topological structure at the ownership boundary in step S3.

[0024] Preferably, the semantic segmentation processing in step S2 adopts the DeepLabv3+ network with multi-scale feature fusion, whose training data contains manually annotated complex terrain and object samples, the network output results are cross-validated with the land parcel boundaries in the cadastral ownership survey information, and the erroneous segmentation areas are iteratively corrected by manual annotation tools, and finally a three-dimensional semantic model with surface cover type and ownership pre-annotation layer is generated.

[0025] Preferably, in step S4, conflict detection is implemented through the following process:

[0026] S41. Construct a spatial topological relationship diagram based on three-dimensional ownership boundary vector data and automatically identify overlapping areas of adjacent plots;

[0027] S42: Calculate the three-dimensional surface integral of the overlapping area. If the overlapping area exceeds a preset threshold (≤0.1 square meters), mark it as an ownership conflict.

[0028] S43. The conflict area is highlighted in the three-dimensional model, and cadastral attribute information is associated with the conflict report, which includes the conflict location coordinates, area deviation, and a list of associated rights holders.

[0029] Preferably, in step S1, the ground mobile measurement equipment is equipped with a multi-sensor fusion module, specifically including: an RTK positioning unit: obtaining centimeter-level precision coordinates in real time, and performing spatiotemporal synchronization with the UAV LiDAR point cloud; a panoramic camera array: 360° surround shooting to supplement high-definition textures of vegetation-covered areas, steep slopes and building blind spots; an inertial navigation system: in areas where satellite signals are blocked, continuous positioning is achieved through gyroscope and accelerometer data to ensure spatial consistency between the point cloud and the image data.

[0030] Preferably, it also includes converting the model data from the local coordinate system to the 2000 geodetic coordinate system or the local independent coordinate system, and automatically marking the boundary point number, boundary line type, adjacent right holder information and area annotation.

[0031] Beneficial effects

[0032] The present invention provides a method for constructing a real-world 3D scene for land title confirmation in areas with complex terrain. Compared with existing technologies, it has the following advantages:

[0033] 1. In the present invention, a real-life 3D model with layered cadastral labels is constructed by registering, denoising, and semantically segmenting point cloud data with image data, and combining the land parcel boundaries, owners, and land use type attributes in the cadastral ownership survey information. The registration of point cloud data with image data is achieved through an improved ICP algorithm. Its registration error function takes into account the weight coefficient, regularization factor, and penalty matrix of the point cloud curvature similarity to constrain the registration distortion in areas with undulating terrain. This improved registration algorithm can better adapt to the characteristics of areas with complex terrain, especially in areas with large terrain undulations. It can effectively reduce registration errors and improve the accuracy and reliability of registration. The registered point cloud serves as input data for terrain mutation line detection, providing more accurate basic data for subsequent ownership boundary extraction, ensuring the accuracy of the entire title confirmation process.

[0034] 2. In the present invention, the detection of terrain mutation lines is based on the calculation of the surface curvature of the three-dimensional model. Its principal curvature is solved by a specific formula. When the rate of change of the surface curvature exceeds a threshold, it is determined to be a terrain mutation line. This threshold is negatively correlated with the vegetation cover density of semantic segmentation. This detection method based on the rate of change of curvature can accurately identify the location of terrain mutations, such as ridge lines and valley lines, providing an important basis for the demarcation of ownership boundaries. At the same time, associating the threshold with the vegetation cover density can better adapt to the detection of terrain mutation lines under different vegetation cover conditions, thereby improving the adaptability and accuracy of detection.

[0035] 3. In the present invention, the three-dimensional vector data of the ownership boundary is associated with the cadastral attribute information to generate a three-dimensional electronic ownership certificate containing spatial coordinates, ownership person, area and use, and supports interactive ownership information query and conflict detection within the three-dimensional model. This three-dimensional electronic ownership certificate displays ownership information in an intuitive three-dimensional form. Users can use interactive queries to directly view detailed information such as the ownership person, area, and use of each plot in the three-dimensional model. Compared with traditional two-dimensional drawings or tables, it is more intuitive and vivid, and easier for users to understand and use. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of a method for constructing a real-scene 3D structure for property rights confirmation in areas with complex terrain proposed by the present invention;

[0037] Figure 2 This is a process framework diagram of a real-scene three-dimensional construction method for property rights confirmation in areas with complex terrain proposed by the present invention. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] See also Figure 1-Figure 2 The present invention provides the following technical solutions, which specifically include the following embodiments:

[0040] Example 1:

[0041] A method for constructing a three-dimensional real scene for land rights confirmation in a complex terrain area includes the following steps:

[0042] S1. Collaborative multi-source data acquisition: Using drone oblique photography, airborne LiDAR, and ground-based mobile surveying equipment, multi-perspective imagery, high-precision point cloud data, and cadastral ownership survey information for the target area are acquired. In step S1, the drone oblique photography utilizes a five-lens camera, with its flight altitude dynamically adjusted based on terrain undulations to ensure image overlap of ≥80%. The ground-based mobile surveying equipment is equipped with a multi-sensor fusion module, specifically including: an RTK positioning unit that acquires centimeter-level precision coordinates in real time and synchronizes them with the drone's LiDAR point cloud in time and space; a panoramic camera array that captures 360° surround images to provide high-definition textures of vegetation-covered areas, steep slopes, and blind spots around buildings; and an inertial navigation system that uses gyroscope and accelerometer data for continuous positioning in areas blocked by satellite signals, ensuring spatial consistency between the point cloud and image data.

[0043] S2. Fusion of 3D Model and Cadastral Data: The point cloud data and image data are registered, denoised, and semantically segmented. The cadastral ownership survey information, including plot boundaries, owners, and land use type attributes, is combined to construct a real-world 3D model with layered cadastral labels. In step S2, the registration of the point cloud data and image data is achieved using an improved ICP algorithm, with the registration error function defined as:

[0044]

[0045] Among them, P i For LiDAR point cloud, Q j is the oblique photography point cloud, T is the rigid body transformation matrix, w i is a weight coefficient based on the curvature similarity of the point cloud, λ is a regularization factor, and Λ is a penalty matrix, which are used to constrain the registration distortion in the area of undulating terrain. The registered point cloud serves as the input data for terrain mutation line detection in step S3. The semantic segmentation processing in step S2 adopts the DeepLabv3+ network with multi-scale feature fusion. Its training data contains manually annotated complex terrain feature samples. The network output is cross-validated with the land parcel boundaries in the cadastral ownership survey information. The segmentation errors are iteratively corrected using manual annotation tools. Finally, a 3D semantic model with surface cover type and ownership pre-annotation layer is generated.

[0046] S3. Automated intelligent extraction of ownership boundaries: Based on the terrain features and cadastral labels of the real-world 3D model, a deep learning algorithm is used to identify surface attachments and terrain break lines, automatically generating 3D vector data of ownership boundaries that comply with land title confirmation rules. The detection of terrain break lines in step S3 is based on the calculation of the surface curvature of the 3D model, and its principal curvatures k1 and k2 are solved using the following formula:

[0047]

[0048] Among them, L, M, N are the second basic form coefficients of the three-dimensional surface. When the surface curvature change rate When it exceeds the threshold τ, it is determined to be a terrain mutation line. The threshold τ is negatively correlated with the vegetation coverage density of the semantic segmentation in step S2;

[0049] S4. Dynamic binding and visualization of ownership information: The three-dimensional vector data of the ownership boundary is associated with the cadastral attribute information to generate a three-dimensional electronic ownership certificate containing spatial coordinates, owner, area and purpose, supporting interactive ownership information query and conflict detection within the three-dimensional model. In step S4, the area calculation adopts the three-dimensional projection correction algorithm, and the actual area of the ownership surface A is obtained. 3D Calculated by the following formula:

[0050]

[0051] Among them, (x i ,y i ) is the plane coordinate of the boundary point, θ i is the local slope angle, Δz i is the elevation difference between adjacent points, Δl i is the horizontal distance, the local slope angle θ i Derived from the curvature data of the terrain abrupt line detection in step S3, the conflict detection in step S4 is implemented through the following process:

[0052] S41. Construct a spatial topological relationship diagram based on three-dimensional ownership boundary vector data and automatically identify overlapping areas of adjacent plots;

[0053] S42: Calculate the three-dimensional surface integral of the overlapping area. If the overlapping area exceeds a preset threshold (≤0.1 square meters), mark it as an ownership conflict.

[0054] S43, highlighting the conflicting area in the three-dimensional model, and generating a conflict report by associating it with the cadastral attribute information, the report including the conflicting location coordinates, area deviation, and a list of associated rights holders;

[0055] S5. Lightweight model and multi-terminal adaptation: Simplify the mesh and compress the data of the real-scene 3D model to generate a multi-resolution model that supports WebGL, mobile terminals and GIS platforms, and output 2D / 3D property maps that comply with the legal format. In step S5, the mesh simplification adopts the edge collapse algorithm based on the quadratic error metric, and the error matrix Q of the vertex ν is ν Defined as:

[0056]

[0057] Among them, K p is the plane equation coefficient of the triangle p, and the folding cost is given by Δ(ν)=ν T Q ν ν is minimized and the simplified model retains the integrity of the topological structure at the ownership boundary in step S3.

[0058] Preferably, it also includes converting the model data from the local coordinate system to the 2000 geodetic coordinate system or the local independent coordinate system, and automatically marking the boundary point number, boundary line type, adjacent right holder information and area annotation.

[0059] Example 2:

[0060] Based on Example 1, a mountainous and hilly area was selected as the test area, with an area of about 2.3 square kilometers, a terrain undulation of 150 meters, and a vegetation coverage rate of more than 70%. There are many terraces, woodlands and homestead ownership disputes in the area. The test goal is to verify the accuracy, efficiency and practicality of the method in complex terrain. Data collection optimization: The drone is equipped with a five-lens tilt camera (focal length 35mm) and RIEGL VUX-1LR LiDAR (point cloud density ≥ 50 points / ㎡), and the flight altitude is dynamically adjusted between 80-120 meters according to the terrain undulation. The heading overlap is 85%, the lateral overlap is 75%, and the ground equipment: Trimble R12RTK and a vehicle-mounted panoramic camera (6-lens array) were used to set up 12 collection paths along roads and valleys to supplement data in obscured areas. Cadastral survey: Combining historical ownership archives with field visits, 23 disputed land boundary points and 87 pieces of ownership information were recorded. Model construction and ownership extraction: Registration and segmentation: Using an improved ICP algorithm (λ = 0.3, Λ = diag(1,1,0.5)), LiDAR and oblique photography point clouds were fused, with a registration error of ≤ 0.05 meters. The DeepLabv3+ network achieved an IoU of 89.7% for object classification on the test set, and a 92.3% agreement between the ownership pre-annotation layer and the field verification. Boundary extraction: Based on a curvature threshold of τ = 0.15 (τ = 0.1 when vegetation density > 60%), the accuracy of the boundary extraction was 100%. 2) Automatically identify surface boundaries such as ridges and walls, extracting a total boundary length of 18.6 kilometers with a manual correction rate of only 4.1%; conflict detection and visualization: Projection correction of disputed plots is performed, and the maximum relative error is 0.48% compared with the actual area measured by traditional total stations; conflict analysis: 9 overlapping areas are detected, with a maximum overlapping area of 0.28 square meters, and conflict reports are automatically generated and located to the corresponding coordinates of the 3D model; lightweight output: The original model of 12.7GB is simplified to 1.3GB through the edge folding algorithm, and the web loading time is reduced from 58 seconds to 6 seconds. The geometric error at the ownership boundary is ≤0.1 meter, and a 2D cadastral map and 3D ownership model in the 2000 coordinate system are output, with boundary point numbers, adjacent owners and slope annotations marked.

[0061] Test results:

[0062] index The results of this method Comparison with traditional methods Plane accuracy (RMS) ≤0.05 m Total station: 0.03 meters Height accuracy (RMS) ≤0.08 m RTK: 0.05 meters Area calculation error ≤0.5％ Two-dimensional projection method: 1.2-3.8% Efficiency of extracting ownership boundaries 18.6 km / 4.2 hours Manual surveying: 1.5 km / day Conflict detection accuracy 100％(9 / 9) Manual investigation: 78% (7 / 9)

[0063] Conclusion: Through the coordinated collection of UAV LiDAR, oblique photography and ground mobile measurement, full coverage of blind spots such as vegetation-covered areas and steep slope blind spots is achieved. The point cloud registration error is ≤0.05 meters, and the three-dimensional model elevation accuracy reaches 0.08 meters (RMS). The three-dimensional property boundary extraction technology based on curvature analysis and deep learning solves the area distortion problem caused by two-dimensional projection. The relative error of the disputed land area calculation is ≤0.5%, which is more than 3 times the accuracy of traditional two-dimensional methods. The property boundary extraction efficiency reaches 18.6 kilometers / 4.2 hours, which is 8 times the efficiency of manual surveying. The overall property rights confirmation period is shortened by 53%, and the labor cost is reduced by more than 60%.

[0064] Example 3:

[0065] A central urban area was selected as the test site. This complex terrain features high-rise buildings, a dense street network, parks and green spaces, and underground facilities. The area, approximately 3.5 square kilometers, features a wide range of building heights and a vegetation coverage rate of approximately 40%. Numerous land ownership disputes exist within the area, primarily surrounding the renovation of older residential communities and the expansion of commercial land. The high-rise complex includes 12 super-high-rise buildings over 30 stories and six historic buildings. The ownership of three-dimensional space includes underground commercial facilities (B3 to B1 floors) and the use of skywalks. The focus of the dispute is the overlapping ownership of the renovation of older residential communities (involving five lanes) and the expansion of commercial land, as well as disputes arising from underground space development.

[0066] Collaborative collection of multi-source data:

[0067] The DJI M300 drone, equipped with a Zenmuse L1 LiDAR (240kHz scanning frequency) and a P1 full-frame camera, flew at night to avoid peak traffic flow, acquiring building facade point clouds (density ≥300 points / ㎡) and oblique imagery. For super-high-rise buildings, a layered, circular flight (collecting data every 50 meters) was used to ensure comprehensive facade coverage. A vehicle-mounted Mobile Measurement System (MMS) collected streetlight point clouds and 360° imagery along major arterial roads such as Nanjing East Road, supplementing areas obscured by the drone. A SLAM backpack scanner (GeoSLAM ZEB-Horizon) was used for underground facilities to collect point clouds from layers B3 to B1, with an accuracy of ±3cm.

[0068] Fusion of 3D models and cadastral data:

[0069] An improved ICP algorithm (λ = 0.2, Λ = diag(1, 0.8, 0.6)) was used to fuse aerial and underground point clouds. The overall error after registration was ≤ 0.06 meters. The DeepLabv3+ network added the "historical building facade" and "underground passage" categories. The training set contained 20,000 annotated samples, and the classification intersection over union reached 87.4%;

[0070] Intelligent extraction of ownership boundaries: The Mask R-CNN network is used to identify historical building walls and temporary fences, and curvature analysis (threshold τ = 0.18) is combined to extract irregular boundaries, automatically generating alternative boundary solutions for disputed areas (such as lane communities). Conflict detection and visualization: Detects spatial conflicts between the sunlight rights of a commercial building's sky corridor and adjacent residential buildings (corridor projection blocks residential windows for >2 hours / day). Lightweight and multi-terminal adaptation is used to associate the 3D vector data of ownership boundaries with cadastral attribute information, generating a 3D electronic ownership certificate containing spatial coordinates, owner, area, and use, supporting interactive ownership information query and conflict detection within the 3D model.

[0071] Test results:

[0072] index The results of this method Comparison with traditional methods Building facade modeling accuracy (RMS) ≤0.05 m Manual measurement: 0.10 m Underground space fusion error ≤0.08 m 2D drawing comparison: cannot be quantified Efficiency of extracting ownership boundaries 9.2 km / 3.5 hours Manual surveying: 0.8 km / day Stereo conflict detection rate 100% (6 / 6 locations) Manual investigation: 66.7% (4 / 6) Response speed to public inquiries ≤3 seconds (LOD1 model) Paper file access: ≥30 minutes

[0073] Conclusion: The use of drones equipped with lidar and cameras for nighttime data collection, combined with a vehicle-mounted mobile measurement system, has achieved the acquisition of high-density point clouds (≥300 points / ㎡) and oblique images. The building facade modeling accuracy (RMS) has reached ≤0.05 meters, which is much higher than the 0.10 meters of traditional manual measurement. The underground space point cloud is collected with a SLAM backpack scanner and fused with the aerial point cloud through the improved ICP algorithm. The underground space fusion error is controlled at ≤0.08 meters, which solves the problem that traditional two-dimensional drawings are difficult to quantify and integrate underground space data. The improved ICP algorithm is used to achieve high-precision registration of aerial and underground point clouds (overall error ≤0.06 meters). The DeepLabv3+ network was used for semantic segmentation, and new categories such as "historical building facades" and "underground passages" were added, with a classification IoU of 87.4%. The Mask R-CNN network was used to identify the walls of historical buildings and temporary enclosures, and irregular boundaries were extracted using curvature analysis. Alternative boundary solutions for disputed areas were automatically generated, with an efficiency of 9.2 kilometers per 3.5 hours for extracting ownership boundaries, far exceeding the 0.8 kilometers per day of traditional manual surveying. Through 3D model analysis and visualization, automatic detection of three-dimensional spatial conflicts such as aerial corridors and sunlight rights was achieved, with a conflict detection rate of 100% (6 / 6 locations), while traditional manual investigation achieved a rate of only 66.7% (4 / 6).

[0074] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements and improvements made within the spirit and principles of the application should be included in the scope of protection of the present application.

Claims

1. A method for constructing a real-scene 3D structure for land rights confirmation in areas with complex terrain, characterized by: The following steps are involved: S1. Collaborative multi-source data acquisition: Using drone oblique photography, airborne LiDAR, and ground-based mobile surveying equipment, we acquire multi-perspective images, high-precision point cloud data, and cadastral ownership survey information for the target area. S2. Fusion of 3D Models and Cadastral Data: This involves registering, denoising, and performing semantic segmentation on point cloud and image data. This process combines the land parcel boundaries, owners, and land use type attributes from the cadastral ownership survey to construct a realistic 3D model with layered cadastral labels. S3. Automated Intelligent Extraction of Ownership Boundaries: Based on the terrain features and cadastral labels of the real-world 3D model, a deep learning algorithm is used to identify surface attachments and terrain abrupt changes, automatically generating 3D vector data of ownership boundaries that comply with land title confirmation regulations. S4. Dynamic binding and visualization of ownership information: This system associates 3D vector data of ownership boundaries with cadastral attribute information to generate a 3D electronic ownership certificate containing spatial coordinates, owner, area, and use. This system supports interactive ownership information query and conflict detection within the 3D model. S5. Lightweight model and multi-terminal adaptation: Simplify the mesh and compress the data of the real-life 3D model to generate a multi-resolution model that supports WebGL, mobile terminals, and GIS platforms, and output 2D / 3D property maps that comply with legal formats.

2. The method for constructing a real-scene 3D structure for confirming property rights in a complex terrain area according to claim 1, characterized in that: In step S1, the drone uses a five-lens camera for oblique photography, and the flight altitude is dynamically adjusted according to the terrain to ensure that the image overlap is ≥80%.

3. The method for constructing a real-scene 3D structure for confirming property rights in a complex terrain area according to claim 1, characterized in that: In step S2, the registration of point cloud data and image data is achieved by an improved ICP algorithm, and the registration error function is defined as: Among them, P i For LiDAR point cloud, Q j is the oblique photography point cloud, T is the rigid body transformation matrix, w i is the weight coefficient based on the curvature similarity of the point cloud, λ is the regularization factor, and Λ is the penalty matrix, which is used to constrain the registration distortion in the terrain undulating area. The registered point cloud is used as the input data for terrain mutation line detection in step S3.

4. The method for constructing a real-scene 3D structure for confirming property rights in a complex terrain area according to claim 1, characterized in that: The detection of the terrain mutation line in step S3 is based on the calculation of the surface curvature of the three-dimensional model, and its main curvatures k1 and k2 are solved by the following formula: Among them, L, M, N are the second basic form coefficients of the three-dimensional surface. When the surface curvature change rate When the threshold value τ is exceeded, it is determined to be a terrain mutation line. The threshold value τ is negatively correlated with the vegetation coverage density of the semantic segmentation in step S2.

5. The method for constructing a real-scene 3D structure for confirming property rights in a complex terrain area according to claim 1, characterized in that: In step S4, the area calculation adopts the three-dimensional projection correction algorithm, and the actual area of the ownership surface A 3D Calculated by the following formula: Among them, (x i ,y i ) is the plane coordinate of the boundary point, θ i is the local slope angle, Δz i is the elevation difference between adjacent points, Δl i is the horizontal distance, the local slope angle θ i Derived from the curvature data detected by the terrain abrupt change line in step S3.

6. The method for constructing a real-scene 3D structure for confirming property rights in a complex terrain area according to claim 1, characterized in that: In step S5, the mesh simplification adopts the edge collapse algorithm based on the quadratic error metric, and the error matrix Q of the vertex ν is ν Defined as: Among them, K p is the plane equation coefficient of the triangle p, and the folding cost is given by Δ(ν)=ν T Q ν ν is minimized and the simplified model retains the integrity of the topological structure at the ownership boundary in step S3.

7. The method for constructing a real-scene 3D structure for confirming property rights in a complex terrain area according to claim 1, characterized in that: The semantic segmentation processing in step S2 adopts the DeepLabv3+ network with multi-scale feature fusion, and its training data contains manually annotated complex terrain and object samples. The network output results are cross-validated with the land parcel boundaries in the cadastral ownership survey information. The erroneous segmentation areas are iteratively corrected using manual annotation tools, and finally a three-dimensional semantic model with surface cover type and ownership pre-annotation layer is generated.

8. The method for constructing a real-scene 3D structure for confirming property rights in a complex terrain area according to claim 1, characterized in that: In step S4, conflict detection is achieved through the following process: S41. Construct a spatial topological relationship diagram based on three-dimensional ownership boundary vector data and automatically identify overlapping areas of adjacent plots; S42: Calculate the three-dimensional surface integral of the overlapping area. If the overlapping area exceeds a preset threshold (≤0.1 square meters), mark it as an ownership conflict. S43. The conflict area is highlighted in the three-dimensional model, and cadastral attribute information is associated with the conflict report, which includes the conflict location coordinates, area deviation, and a list of associated rights holders.

9. The method for constructing a real-scene 3D structure for confirming property rights in a complex terrain area according to claim 1, characterized in that: In step S1, the ground mobile measurement equipment is equipped with a multi-sensor fusion module, specifically including: an RTK positioning unit: acquiring centimeter-level precision coordinates in real time and performing spatiotemporal synchronization with the UAV LiDAR point cloud; a panoramic camera array: providing 360° surround shooting to supplement high-definition textures of vegetation-covered areas, steep slopes, and building blind spots; an inertial navigation system: achieving continuous positioning in areas blocked by satellite signals through gyroscope and accelerometer data to ensure spatial consistency between the point cloud and image data.

10. The method for constructing a real-scene 3D structure for confirming property rights in a complex terrain area according to claim 1, characterized in that: It also includes converting model data from the local coordinate system to the 2000 geodetic coordinate system or local independent coordinate system, and automatically marking boundary point numbers, boundary line types, adjacent rights holder information and area annotations.

Images