A 3D modeling and mapping method for high-rise buildings based on UAV laser scanning

Abstract

This application relates to the fields of drone mapping and 3D building modeling. It discloses a method for 3D modeling and mapping of high-rise buildings based on drone laser scanning. The method comprises the following steps: deploying a lidar, millimeter-wave radar, a polarization camera, and an intelligent blind-spot supplementary drone for collaborative scanning; generating an optimal path through topological manifold learning; utilizing geometric algebra and Lie group optimization for high-precision registration of multi-source point clouds; fusing heterogeneous multi-sensor data using tensor decomposition; and dynamically adjusting the path through real-time blind-spot detection to ultimately generate a complete 3D building model. This method can significantly improve the modeling accuracy and data integrity of complex building structures (such as glass curtain walls), effectively avoiding scanning blind spots and single-point equipment failures, and providing an efficient and robust solution for 3D modeling of high-rise buildings.

Description

Technical Field

[0001] The present invention relates to the technical field of drone surveying and mapping and building three-dimensional modeling, and specifically to a three-dimensional modeling and surveying method for high-rise buildings based on drone laser scanning. Background Art

[0002] With the widespread adoption of drone technology in architectural surveying and mapping, LiDAR-based 3D modeling has become a crucial tool for high-rise building surveying. However, existing technologies still face significant challenges in practical application. Traditional drone scanning methods are prone to blind spots in complex building structures (such as glass curtain walls and dense support frames), resulting in missing modeling data. Furthermore, heterogeneous data collected by multiple sensors is difficult to effectively fuse due to coordinate system deviations and registration errors, compromising model accuracy.

[0003] Furthermore, existing systems lack fault tolerance for equipment failures or environmental disturbances, and single-point failures can easily lead to interrupted modeling tasks. This is especially true on reflective surfaces like glass curtain walls, where conventional sensors can misjudge open spaces due to reflected light interference, increasing flight risks.

[0004] These problems severely limit the reliability and practicality of drone mapping technology in high-rise building scenarios, and there is an urgent need for a solution that can balance data integrity, modeling accuracy and system robustness. Summary of the Invention

[0005] In response to the shortcomings of the existing technology, the present invention provides a three-dimensional modeling and mapping method for high-rise buildings based on drone laser scanning, which solves the technical problems of the existing technology in three-dimensional modeling of high-rise buildings caused by complex structures, such as scanning blind spots, multi-source data registration errors and insufficient system fault tolerance.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for three-dimensional modeling and mapping of high-rise buildings based on drone laser scanning, comprising the following steps:

[0007] Deploy multiple drones for collaborative scanning to collect LiDAR point cloud data, millimeter-wave radar penetration data, and polarized camera material data of buildings;

[0008] Generate optimal flight paths for drones covering building surfaces based on topological manifold learning;

[0009] Use geometric algebra and Lie group optimization to register multi-source point cloud data;

[0010] Fuse heterogeneous data from lidar, millimeter-wave radar, and polarization cameras through tensor decomposition;

[0011] Detect scanning blind spots in real time and dynamically adjust the drone path to generate a complete 3D building model.

[0012] Preferably, the drone comprises:

[0013] At least one drone equipped with lidar;

[0014] At least one drone equipped with millimeter-wave radar;

[0015] At least one drone equipped with a polarized light camera.

[0016] Preferably, the drone further comprises:

[0017] At least one intelligent blind spot supplementary drone is used to detect and fill scanning blind spots in real time.

[0018] Preferably, the step of generating the optimal flight path of the drone covering the building surface based on topological manifold learning includes:

[0019] Extract key structural nodes of building point clouds and construct topological skeleton graphs;

[0020] Generate geodesic distance optimal paths covering building surfaces through manifold space embedding.

[0021] Preferably, the key structural nodes are extracted by using a second-order derivative threshold of a point cloud density function.

[0022] Preferably, the step of registering multi-source point cloud data using geometric algebra and Lie group optimization includes:

[0023] Represent point cloud data as spinor form and optimize rotation and translation parameters through Lie group parameterization;

[0024] Minimizing multi-source point cloud registration error.

[0025] Preferably, the step of fusing heterogeneous data of the laser radar, millimeter wave radar and polarization camera by tensor decomposition includes:

[0026] Construct a four-dimensional tensor containing spatial coordinates, timestamp, sensor type, and feature confidence;

[0027] Core features are extracted through canonical multivariate decomposition and weights are dynamically assigned based on polarization data.

[0028] Preferably, the dynamic weight of the polarized light data is adjusted according to the polarization angle confidence of the glass curtain wall area.

[0029] Preferably, the steps of real-time detection of scanning blind spots and dynamic adjustment of the drone path include:

[0030] Real-time analysis of point cloud density distribution to identify low coverage areas;

[0031] Generate blind area supplementary paths based on manifold geodesic distance.

[0032] The present invention also provides a high-rise building three-dimensional modeling and mapping system based on drone laser scanning, comprising:

[0033] Data acquisition modules, including multiple drones equipped with lidar, millimeter-wave radar, and polarized light cameras;

[0034] Path planning module, used to generate UAV flight paths based on topological manifold learning;

[0035] Point cloud registration module, which aligns multi-source point cloud data through geometric algebra and Lie group optimization;

[0036] Data fusion module, which fuses heterogeneous sensor data through tensor decomposition;

[0037] Blind area processing module, real-time detection and dynamic filling of scanning blind areas;

[0038] Modeling output module generates a complete three-dimensional building model.

[0039] The present invention provides a method for three-dimensional modeling and mapping of high-rise buildings based on drone laser scanning. It has the following beneficial effects:

[0040] 1. This invention deploys multiple drones, including lidar, millimeter-wave radar, and polarized light cameras, working in tandem to comprehensively collect building surface point clouds, penetrating structures, and glass curtain wall material information. The lidar provides high-precision geometric data, the millimeter-wave radar penetrates obstructed areas to provide depth information, and the polarized light camera prevents misjudgment of glass curtain wall reflections. The complementary data from these three ensures comprehensive coverage of complex building structures.

[0041] 2. This invention generates an optimal drone flight path based on topological manifold learning, enabling coverage of building surfaces with minimal geodesic distance, avoiding the repeated scanning and blind spot issues inherent in traditional path planning. By integrating manifold space embedding technology with point cloud density and normal vector constraints, path planning efficiency is significantly improved, shortening overall scanning time.

[0042] 3. This invention utilizes geometric algebra and Lie group optimization to register multi-source point cloud data, efficiently aligning the spatial coordinate systems of different sensors and reducing registration errors. Through spinor representation and Lie group parameterized iterative optimization, it effectively addresses the cumulative error problem of traditional registration methods in complex structures and improves the spatial consistency of 3D models.

[0043] 4. This invention fuses heterogeneous data from multiple sources using four-dimensional tensor decomposition technology, dynamically assigning confidence weights to different sensors (for example, polarized light data has a higher weight in the glass curtain wall area), reducing noise interference. This technology combines physical properties with mathematical modeling to significantly improve the credibility of the fused data and the ability to restore model details.

[0044] 5. The intelligent blind spot supplementary drone of this invention detects low-coverage areas through real-time point cloud density analysis and dynamically generates supplementary paths based on manifold geodesic distances to ensure scanning integrity. Furthermore, the drone acts as a redundant backup device, automatically taking over tasks in the event of equipment failure, avoiding modeling interruptions caused by single points of failure and improving system fault tolerance and stability. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 Schematic diagram of the method flow of the present invention;

[0046] Figure 2 Schematic diagram of the system structure of the present invention.

[0047] Among them, 10, data acquisition module; 20, path planning module; 30, point cloud registration module; 40, data fusion module; 50, blind spot processing module; 60, modeling output module. DETAILED DESCRIPTION

[0048] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the present specification. 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.

[0049] Please see the attached Figure 1 The present invention provides a three-dimensional modeling and mapping method for high-rise buildings based on drone laser scanning, which aims to deploy multiple drones to work together, use different sensors to collect multi-source data of high-rise buildings, and generate an accurate three-dimensional model of the building through intelligent data fusion and path optimization technology.

[0050] like Figure 1 As shown, the high-rise building three-dimensional modeling and mapping method based on drone laser scanning may include the following steps:

[0051] S1. Deploy multiple drones to collaboratively scan and collect LiDAR point cloud data, millimeter-wave radar penetration data, and polarized camera material data of buildings;

[0052] S2, generating the optimal flight path of a drone covering the building surface based on topological manifold learning;

[0053] S3, using geometric algebra and Lie group optimization to register multi-source point cloud data;

[0054] S4, fuses heterogeneous data from lidar, millimeter-wave radar, and polarization camera through tensor decomposition;

[0055] S5. Real-time detection of scanning blind spots and dynamic adjustment of the drone path to generate a complete 3D building model.

[0056] The following is a detailed description of each step in the method of the present invention, which comprehensively explains the specific implementation principles, technical details and processes of each step.

[0057] Step S1 involves collaboratively scanning high-rise buildings with drones, collecting the building's LiDAR point cloud data, millimeter-wave radar penetration data, and polarized camera material data. To ensure comprehensive and accurate data collection, multiple drone systems work together to cover every part and feature of the building. The specific process is as follows:

[0058] First, multiple drones are deployed for collaborative scanning. Each drone is equipped with different sensors to collect different types of data, as follows:

[0059] At least one drone is equipped with a laser radar (LiDAR) to collect precise point cloud data on and around the building surface. LiDAR calculates the position and distance of points by emitting laser beams and measuring the time it takes for the laser to return, generating three-dimensional spatial coordinates.

[0060] At least one drone is equipped with a millimeter-wave radar, which is used to penetrate rain, fog, or other obstructions and provide depth information outside buildings or obscured areas. Millimeter-wave radar transmits high-frequency electromagnetic waves and receives reflected waves to obtain the location and shape data of the target object.

[0061] At least one drone is equipped with a polarization camera to collect information about the building's facade materials, particularly the characteristics of glass curtain walls. Polarization cameras detect the polarization state of light and analyze the polarization angle of light reflected from the surface, thereby identifying the characteristics of different materials.

[0062] Furthermore, to compensate for potential blind spots created by traditional scanning methods, at least one intelligent blind spot compensation drone is deployed. This drone monitors blind spots created during scanning in real time and automatically adjusts its path to fill these gaps. Equipped with a high-performance computing module, this intelligent blind spot compensation drone analyzes the density distribution of point cloud data in real time, identifies areas with low coverage, and dynamically generates a blind spot compensation path. By optimizing the path, the drone can cover unscanned areas, ensuring the integrity of the collected data.

[0063] All drone systems utilize a wireless communication network for real-time data synchronization and scheduling. During the scanning process, each drone dynamically adjusts its planned flight path to ensure coverage of all building surfaces, avoiding duplicate scans or missed areas. Drone path planning is based on the building's 3D model and LiDAR point cloud data, further optimizing flight paths. The wireless communication network ensures real-time sharing of flight data between drones, preventing data redundancy during the collaborative process.

[0064] Specifically, the lidar data collection process involves emitting pulsed laser light from a laser transmitter and calculating the time it takes for the light to reflect back to determine the location of a point. By continuously emitting laser light and measuring the time, a point cloud of the building can be mapped in three-dimensional space.

[0065] Millimeter-wave radar collects data by emitting high-frequency electromagnetic waves and measuring the time delay of the reflected waves. Millimeter-wave radar has strong penetrating capabilities, making it particularly suitable for scanning areas obstructed by the atmosphere or other obstructions. By measuring the time delay of the reflected waves, radar can calculate the distance and shape of the target object.

[0066] Data acquisition with polarization cameras is primarily based on polarization angle analysis. Polarization cameras determine the surface properties of different materials by detecting the polarization angle of light reflected from an object. For surfaces like glass curtain walls, polarization cameras can accurately determine the optical properties of the surface material by analyzing the polarization characteristics of reflected light, providing support for subsequent 3D modeling.

[0067] Furthermore, by analyzing the polarization angle of reflected light, polarized cameras can effectively distinguish between glass curtain walls and open spaces, avoiding misjudgments caused by high reflectivity. During drone navigation, if a glass curtain wall is not correctly identified, it may be misjudged as a traversable open space, leading to flight path planning errors and increasing the risk of a crash. By incorporating data from polarized cameras, the AI ​​system can provide more reliable material identification information, avoiding flight accidents caused by misjudgments. This also reduces reliance on deep learning models, mitigates the risk of AI decision-making becoming a black box, and improves the system's interpretability and safety.

[0068] In summary, the core of step S1 lies in the collaborative work of the multi-UAV system. During the flight, each UAV can automatically adjust its path according to the collected data, detect and supplement the scanning blind spots in real time.

[0069] In step S2, the optimal flight path for the drone covering the building surface is generated based on topological manifold learning. This step ensures that the drone completes high-precision scanning within the shortest path and optimizes data collection efficiency. The specific implementation method is as follows:

[0070] First, the initial 3D point cloud model of the building is constructed based on the LiDAR point cloud data. Assume that the point cloud dataset of the building surface is ,in Represents a three-dimensional coordinate point, is the total number of point cloud data points.

[0071] Then, the key structural nodes are extracted from the point cloud data. The point cloud data is screened using the point cloud gradient change rate calculation method, and points with significant local gradient changes are selected as key nodes. The local gradient function is set as:

[0072]

[0073] in, Representative Points The gradient change rate at , select The point is taken as the key structural node. is the gradient change threshold.

[0074] Next, build the topological skeleton graph ,in is the set of key structural nodes, For edges between adjacent nodes, the weight of the edge is defined using constraints based on geodesic distance. :

[0075]

[0076] in, Indicates a point and The Euclidean distance between are the normal vectors of the corresponding points, is the weight factor. This weight calculation method can take into account both geometric proximity and normal vector changes to ensure the rationality of the topological structure.

[0077] Based on the topological skeleton graph, the manifold learning method is used to build a UAV flight path optimization model. For the optimal path from the starting point to the target point, the path cost function is defined as:

[0078]

[0079] Use Dijkstra algorithm or A* algorithm to solve the minimum path cost The corresponding optimal path. Through this optimization process, it is ensured that the drone covers the building surface on the optimal path.

[0080] Finally, the optimized path is converted into the flight instructions of the UAV. Assume that the attitude parameters of the UAV are (corresponding to yaw angle, pitch angle, and roll angle respectively), the flight control input vector is ,in is the displacement speed of the UAV in three-dimensional space, is the angular velocity of rotation. The UAV control system adjusts the attitude through the following control law:

[0081]

[0082] in, is the position of the UAV in the world coordinate system, This control strategy ensures that the UAV flies smoothly along the optimized path and completes the task of scanning the building.

[0083] In summary, this step generates an optimal flight path using a topological manifold learning method, enabling the drone to efficiently cover the building surface while avoiding path redundancy and scanning blind spots. This optimized path can be directly converted into drone flight control commands, enabling precise trajectory planning.

[0084] For step S3, geometric algebra and Lie group optimization are used to align multi-source point cloud data to ensure that data from different sensors can be accurately aligned to form a unified 3D model. The specific implementation method is as follows:

[0085] First, a spatial representation model of multi-source point cloud data is established. Assume that the lidar point cloud dataset is , the millimeter wave radar point cloud dataset is , the polarization camera dataset is .in, are the number of points in their respective point cloud data.

[0086] Then, the point cloud data is modeled using the spinor representation method for rigid body transformation. Suppose the rotation matrix of a data set relative to the global coordinate system is , the translation vector is , then the transformation relationship of point cloud data in the unified coordinate system can be expressed as:

[0087]

[0088] in, is the transformed coordinate, belong Rotation group, belong Translation vector.

[0089] In order to optimize the point cloud registration results, Lie group Perform pose optimization. Assume the variables to be optimized Belongs to Lie algebra ,in:

[0090]

[0091] in, represents the rotation component, Represents the translation component.

[0092] Through index mapping Lie algebra Mapping to Lie groups , and get the transformation matrix:

[0093]

[0094] in, is the antisymmetric matrix representation of Lie algebra, and its specific form is:

[0095]

[0096]

[0097] The optimal registration parameters are solved by iterative optimization method so that the error function Minimum, where the error is defined as:

[0098]

[0099] Use Gauss-Newton method or Levenberg-Marquardt algorithm to minimize the solution and iteratively update the transformation matrix , so that the point cloud data of different sensors are aligned in the same coordinate system.

[0100] Finally, all the registered point cloud data is fused to form a complete 3D building model and output to the subsequent modeling module for further analysis. This step ensures accurate registration of multi-source data, improves the consistency of the point cloud data, and provides precise foundational data for subsequent 3D reconstruction.

[0101] In step S4, tensor decomposition technology is used to fuse the multi-source heterogeneous data from lidar, millimeter-wave radar, and polarization camera. This step aims to effectively fuse the data obtained by different sensors, improve the accuracy and reliability of the data, and provide accurate data support for subsequent 3D modeling. The specific implementation method is as follows:

[0102] First, construct a four-dimensional tensor containing lidar, millimeter-wave radar, and polarization camera data , which has the form:

[0103]

[0104] in, Represent different sensors, spatial coordinates, time and other features (such as confidence, etc.). Specifically, the dimension Corresponding sensor types (lidar, millimeter wave radar, polarization camera), Corresponding space coordinates, The corresponding timestamp, Corresponding confidence and other additional information.

[0105] Confidence is a key factor, especially in multi-source data fusion. Data acquired by different sensors in different environments has varying degrees of accuracy and reliability, so introducing a confidence level to reflect the trustworthiness of each data type is crucial. Specifically, confidence can be used to weight data from different sensors, ensuring that high-confidence data is given greater weight during fusion, thereby improving overall data quality.

[0106] Assume that each sensor The point cloud data generated by (such as lidar, millimeter wave radar, polarized light camera) is , each data point Corresponding to a confidence level , indicating the accuracy or reliability of the data point. The calculation of confidence is related to the measurement error of the sensor, environmental factors, the characteristics of the sensor itself, and the data quality. , the basic formula for confidence calculation is:

[0107]

[0108] in, For data points The measurement error (standard deviation) of the data. According to this formula, the smaller the measurement error, the greater the confidence level; conversely, the larger the measurement error, the smaller the confidence level.

[0109] In multi-source data fusion, each sensor data is weighted by calculating the weighted confidence of each sensor point cloud data. , weighted confidence It can be expressed as:

[0110]

[0111] When fusing multi-source data, it is assumed that the data of each sensor is ,in For the The weighted data fusion formula is as follows:

[0112]

[0113] in, For the The weight of each sensor is calculated as follows:

[0114]

[0115] This means that the weight of each sensor is calculated based on the sum of its confidence scores across all data points. They will be weighted according to their confidence and eventually form a comprehensive fused dataset.

[0116] Next, the four-dimensional tensor is decomposed using tensor decomposition technology. The common decomposition method is CP decomposition (Canonical Polyadic Decomposition). CP decomposition converts the original four-dimensional tensor into Decomposed into the product of multiple matrices, namely:

[0117]

[0118] in, is the rank of the decomposition, is a scalar factor, represents the outer product operation of tensors, , , , is the factor matrix of the tensor, corresponding to features such as sensor, spatial coordinates, timestamp, and confidence. Through this decomposition process, the main features in the data can be effectively extracted and redundant information can be reduced.

[0119] Furthermore, the decomposition results are optimized by introducing a regularization term to ensure the stability and accuracy of the decomposition results in practical applications. The form of the regularization term is:

[0120]

[0121] in, , , , is the regularization coefficient, Represents the Frobenius norm of the matrix. This regularization term can suppress overfitting and ensure that the fused data has good generalization ability.

[0122] The decomposed data is then used to perform multi-source data fusion. Data collected by different sensors can be weighted based on their physical properties and confidence levels to generate a comprehensive 3D model. Specifically, LiDAR point cloud data provides high accuracy for a building's appearance, while millimeter-wave radar data is more penetrating, and polarized camera data can provide material information. During data fusion, a weighted average method is used to adjust the weights of different sensors based on their accuracy and confidence levels to produce the resulting composite data:

[0123]

[0124] in, , , They are the point cloud data of laser radar, millimeter wave radar and polarization camera respectively. , , is the weight of each data, is the fused point cloud data.

[0125] Finally, the fused data is used to generate a high-precision 3D building model, providing reliable data input for subsequent modeling and analysis. This step enables the precise fusion of multi-source data, significantly improving the quality and reliability of point cloud data and providing a foundation for subsequent 3D modeling.

[0126] In step S5, a complete 3D building model is generated by real-time detection of scanning blind spots and dynamic adjustment of the drone's path. This step identifies low-coverage areas by analyzing the point cloud density distribution and generates a supplementary path based on manifold geodesic distance to ensure complete scanning coverage. The specific implementation is as follows:

[0127] First, based on the fused data processed in steps S1 to S4, a point cloud density distribution model is established. Suppose the point cloud data set is , define the point cloud density function as:

[0128]

[0129] in, is the Gaussian kernel function, is the bandwidth parameter, which is used to control the smoothness of density estimation. , will satisfy The area is determined as a scanning blind area. is the preset density threshold.

[0130] Furthermore, based on the topological manifold space generated in step S2, a blind spot supplementary path is constructed. Let the center point of the blind spot be , the current position of the drone is , generating the shortest complementary path through the geodesic distance optimization model:

[0131]

[0132] in, is the metric tensor of the manifold space, derived from the topological skeleton graph in step S2. The optimal path is solved by the A* algorithm , which enables the intelligent blind spot supplementary UAV to fly along the path and fill the blind spot.

[0133] At the same time, the intelligent blind spot supplementary drone serves as a redundant backup device. When other drones fail due to malfunction or communication interruption, it automatically takes over the scanning task of the failed drone. The redundant backup mechanism is achieved by real-time monitoring of the status of each drone. Let the drone state vector be ,in Indicates the The working status of each drone (0 means fault, 1 means normal). When the backup drone's mission takeover protocol is triggered:

[0134]

[0135] Then, the UAV flight control parameters are adjusted dynamically. Assume the target position is , the drone speed is , the control law is designed as:

[0136]

[0137] in, is the proportional gain, The PID controller is used to adjust the trajectory of the drone in real time to ensure that it moves along the Precision flying.

[0138] Finally, the scan data will be supplemented Combined with the fused data output from step S4, the registration is performed using the iterative closest point (ICP) algorithm:

[0139]

[0140] in, , are the rotation matrix and translation vector, The corresponding points in the global coordinate system are input into the 3D reconstruction module after registration to generate a complete 3D building model. .

[0141] In summary, this step achieves real-time detection and supplementation of scanning blind spots through density analysis, redundant backup, and dynamic path adjustment, while ensuring the robustness of the system in the event of partial equipment failure, and can effectively solve the scanning coverage problem in complex building structures.

[0142] In summary, this method achieves full scene coverage by deploying multiple drones equipped with lidar, millimeter-wave radar, and polarized cameras for collaborative scanning, combined with dynamic path optimization and redundant backup mechanisms for intelligent blind-spot supplementary drones. First, the optimal flight path is generated based on topological manifold learning. Multi-source point cloud registration is achieved through geometric algebra and Lie group optimization. Heterogeneous sensor data is fused using tensor decomposition. Finally, blind spots are detected in real time and the path is dynamically adjusted for supplementary scanning. This method can address modeling interruptions caused by glass curtain wall misjudgments, missed scans of obscured areas, and equipment failures, generating highly accurate and complete three-dimensional building models.

[0143] The three-dimensional modeling and mapping system for high-rise buildings based on drone laser scanning described below and the three-dimensional modeling and mapping method for high-rise buildings based on drone laser scanning described above can be referred to each other.

[0144] Please see the attached Figure 2 The present invention also provides a high-rise building three-dimensional modeling and mapping system based on drone laser scanning, comprising:

[0145] The data acquisition module 10 includes multiple drones equipped with laser radars, millimeter-wave radars, and polarized light cameras;

[0146] A path planning module 20 is used to generate a UAV flight path based on topological manifold learning;

[0147] Point cloud registration module 30, aligning multi-source point cloud data through geometric algebra and Lie group optimization;

[0148] The data fusion module 40 fuses heterogeneous sensor data through tensor decomposition;

[0149] Blind area processing module 50, real-time detection and dynamic filling of scanning blind areas;

[0150] The modeling output module 60 generates a complete three-dimensional building model.

[0151] The system of this embodiment can be used to execute the above method embodiments, and its principles and technical effects are similar, so they will not be repeated here.

[0152] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for three-dimensional modeling and mapping of high-rise buildings based on drone laser scanning, characterized in that: The following steps are involved: Deploy multiple drones for collaborative scanning to collect LiDAR point cloud data, millimeter-wave radar penetration data, and polarized camera material data of buildings; Generate optimal flight paths for drones covering building surfaces based on topological manifold learning; Use geometric algebra and Lie group optimization to register multi-source point cloud data; Fuse heterogeneous data from lidar, millimeter-wave radar, and polarization cameras through tensor decomposition; Real-time detection of scanning blind spots and dynamic adjustment of drone paths to generate complete 3D building models; The drone includes: At least one drone equipped with lidar; At least one drone equipped with millimeter-wave radar; At least one drone equipped with a polarized camera; At least one intelligent blind spot supplementary drone for real-time detection and filling of scanning blind spots; The steps of fusing heterogeneous data of the laser radar, millimeter-wave radar, and polarization camera through tensor decomposition include: Construct a four-dimensional tensor containing spatial coordinates, timestamp, sensor type, and feature confidence; Extract core features through canonical multivariate decomposition and dynamically assign weights based on polarization data; The dynamic weight of the polarized light data is adjusted according to the polarization angle confidence of the glass curtain wall area; The steps of real-time detection and scanning blind spots and dynamic adjustment of the drone path include: Real-time analysis of point cloud density distribution to identify low coverage areas; Generate blind area supplementary paths based on manifold geodesic distance.

2. The method for three-dimensional modeling and mapping of high-rise buildings based on drone laser scanning according to claim 1 is characterized in that: The step of generating the optimal flight path of the drone covering the building surface based on topological manifold learning includes: Extract key structural nodes of building point clouds and construct topological skeleton graphs; Generate geodesic distance optimal paths covering building surfaces through manifold space embedding.

3. The method for three-dimensional modeling and mapping of high-rise buildings based on drone laser scanning according to claim 2 is characterized in that: The key structural nodes are extracted by using the second-order derivative threshold of the point cloud density function.

4. The method for three-dimensional modeling and mapping of high-rise buildings based on drone laser scanning according to claim 1 is characterized in that: The step of registering multi-source point cloud data using geometric algebra and Lie group optimization includes: Represent point cloud data as spinor form and optimize rotation and translation parameters through Lie group parameterization; Minimizing multi-source point cloud registration error.

5. A high-rise building three-dimensional modeling and mapping system based on drone laser scanning, used to execute the method according to any one of claims 1 to 4, characterized in that: include: Data acquisition modules, including multiple drones equipped with lidar, millimeter-wave radar, and polarized light cameras; Path planning module, used to generate UAV flight paths based on topological manifold learning; Point cloud registration module, which aligns multi-source point cloud data through geometric algebra and Lie group optimization; Data fusion module, which fuses heterogeneous sensor data through tensor decomposition; Blind area processing module, real-time detection and dynamic filling of scanning blind areas; Modeling output module generates a complete three-dimensional building model.

Images