Disaster site three-dimensional surveying and mapping system based on unmanned aerial vehicle

Abstract

The invention relates to the technical field of disaster site surveying and mapping, and discloses a disaster site three-dimensional surveying and mapping system based on an unmanned aerial vehicle. According to the system, an unmanned aerial vehicle is controlled to carry a multispectral sensor to carry out omnibearing scanning on a disaster site, and earth surface deformation data and multiband reflectivity data are collected; carrying out time sequence alignment and noise filtering on the collected data, constructing an initial three-dimensional point cloud through dynamic voxelization processing, and fusing inertial measurement unit data to correct a spatial position; identifying earth surface fault zone edge and elevation mutation features from the point cloud, and dividing geological stability partitions by combining reflectivity data; a self-adaptive octree segmentation algorithm is adopted to carry out multi-level detail division on the stability partitions, the grid resolution is automatically adjusted according to the fault zone density, and a disaster site three-dimensional topological structure is generated; by comparing historical and current three-dimensional data, the earth surface displacement vector and the settlement gradient are calculated, a potential secondary disaster risk area is marked, and accurate assessment and risk early warning of a disaster site are achieved.

Description

Technical Field

[0001] This invention relates to the field of disaster site mapping technology, specifically a three-dimensional disaster site mapping system based on unmanned aerial vehicles (UAVs). Background Technology

[0002] Following natural disasters such as earthquakes, landslides, and debris flows, rapidly and accurately obtaining topographical information about the disaster site is crucial for emergency rescue, disaster assessment, and prevention of secondary disasters. Traditional disaster site mapping mainly relies on manual on-site surveys or aerial photogrammetry. Manual surveys are inefficient, risky, and unable to cover large or dangerous areas. Aerial photogrammetry can quickly acquire regional images, but data processing cycles are long, and its ability to identify fine features such as surface deformation and cracks is limited. Satellite remote sensing technology has the advantage of wide coverage, but its long revisit cycle and limited spatial resolution make it difficult to meet the high timeliness requirements of disaster emergency response.

[0003] In recent years, drone technology has been introduced into the field of disaster monitoring due to its flexibility and efficiency. Drones equipped with optical cameras can quickly acquire high-resolution images of disaster areas and generate digital surface models using photogrammetry principles. However, traditional drone mapping systems primarily focus on generating elevation models of the terrain, neglecting to explore deeper information such as surface deformation mechanisms and geological stability. The surface at disaster sites often exhibits complex deformations such as fractures, cracks, and subsidence; these characteristics are crucial for assessing the development of disasters and secondary risks. Existing systems typically use uniform resolution for 3D modeling, failing to adaptively adjust modeling accuracy according to the complexity of terrain features, resulting in either overall model redundancy or loss of detail in critical areas.

[0004] Multispectral sensors can acquire the reflectance characteristics of ground objects in different wavelength bands, playing a crucial role in identifying parameters related to geological stability, such as soil moisture and lithological changes. However, existing UAV mapping systems largely utilize multispectral data at the classification or identification level, failing to deeply couple it with high-precision 3D terrain models to support comprehensive zoning assessments of geological stability. Furthermore, disaster sites are dynamically changing, especially landslides and collapses, whose deformation is continuously evolving. Traditional mapping results are mostly static snapshots, lacking the ability to compare and analyze with historical data, making it impossible to effectively calculate displacement fields, predict evolution trends, and provide forward-looking information for disaster early warning. Summary of the Invention

[0005] The purpose of this invention is to provide a three-dimensional mapping system for disaster sites based on unmanned aerial vehicles (UAVs) to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides a three-dimensional mapping system for disaster sites based on unmanned aerial vehicles (UAVs), the system comprising: The disaster area scanning module is used to control the UAV equipped with a multispectral sensor to perform a comprehensive scan of the disaster site, collect surface deformation data and multi-band reflectivity data, and generate a raw terrain scan dataset. The real-time point cloud generation module is used to perform temporal alignment and noise filtering on the original terrain scan dataset, construct the initial 3D point cloud through dynamic voxelization, and fuse inertial measurement unit data to correct spatial position deviations. The terrain feature extraction module is used to identify the edge features of surface fault zones and abrupt elevation changes from the initial 3D point cloud, and to divide the geological stability zones by combining multi-band reflectance data. The dynamic modeling engine module uses an adaptive octree segmentation algorithm to perform multi-level detailed subdivision of geological stability zones, and automatically adjusts the local mesh resolution according to the fault zone density to generate a three-dimensional topological structure of the disaster site. The disaster evolution analysis module calculates the surface displacement vector field and settlement gradient field by comparing historical mapping data with the current three-dimensional topology, and marks potential secondary disaster risk areas.

[0007] Preferably, the disaster area scanning module includes: The flight path planning unit automatically generates spiral or grid scanning paths based on the type of disaster and adjusts the flight altitude and sensor tilt angle in real time. The multi-source data synchronization unit performs timestamp alignment and spatial registration of lidar point cloud data, visible light image data and infrared thermal imaging data; The scan quality control unit monitors the integrity of data acquisition in real time and triggers supplementary scan commands when there are obstructed areas.

[0008] Preferably, the real-time point cloud generation module includes: The motion distortion correction unit uses angular velocity data from the inertial measurement unit to compensate for point cloud distortion caused by changes in the attitude of the UAV. The multi-frame fusion unit uses an iterative nearest-point algorithm to convert continuous temporal scan frames into a dense point cloud in a unified coordinate system. The dynamic noise reduction unit automatically identifies and removes outlier noise points based on local surface fitting errors.

[0009] Preferably, the terrain feature extraction module includes: The fault zone identification unit detects areas of discontinuity on the earth's surface by calculating the extreme value of the principal curvature, and confirms the geological fault boundary by combining the reflectance abrupt change threshold. The elevation clustering unit uses the density peak clustering algorithm to separate terrain blocks at different elevations and marks steep slopes and subsidence areas; The stability grading unit generates a geological stability level map based on the distribution density of fault zones and the coefficient of variation of elevation.

[0010] Preferably, the dynamic modeling engine module includes: The detailed level of the subdivision unit is automatically selected according to the point cloud density in the octree node, and the subdivision level is improved in the area around the fault zone. The topology reconstruction unit uses the Poisson surface reconstruction algorithm to generate a closed mesh model, preserving the geometric discontinuities at the fracture edges; The texture mapping unit projects multi-band reflectivity data onto the surface of the mesh model to generate a high-resolution texture map.

[0011] Preferably, the disaster evolution analysis module includes: The displacement field calculation unit obtains the displacement of corresponding points in the historical and current 3D models through a feature point matching algorithm; Gradient analysis unit calculates the spatial distribution of settlement rate and identifies areas of abnormally accelerated settlement. The risk prediction unit, combined with the soil mechanics parameter library, assesses the probability of slope slippage and the collapse risk index.

[0012] Preferably, the system further includes: The emergency communication relay module establishes a low-latency data link between the UAV and the ground command center, transmitting compressed 3D model key feature data in real time.

[0013] Preferably, the emergency communication relay module includes: Data segmentation units divide the 3D model into transmission blocks of different priorities according to the geological stability level; An adaptive coding unit dynamically adjusts the compression ratio of point cloud data to texture data based on the channel bandwidth. The link switching unit automatically switches to a backup frequency band to maintain data transmission when a single communication frequency band is blocked.

[0014] Preferably, the system further includes: The surveying and verification module generates accuracy correction parameters for the 3D model by analyzing the residuals between the measured coordinates of ground control points and the model coordinates.

[0015] Preferably, the mapping verification module includes: The control point matching unit uses the spatial coordinates of known ground markers to inversely calculate the model coordinate system transformation parameters. Error distribution unit calculates the root mean square error of different regions of the model and generates an accuracy heatmap; The parameter optimization unit automatically adjusts the iterative convergence threshold of the point cloud registration algorithm based on the accuracy heatmap.

[0016] Compared with the prior art, the beneficial effects of the present invention are: This invention utilizes a drone equipped with a multispectral sensor to scan and achieve efficient, simultaneous acquisition of topographic and spectral information of ground features at disaster sites. Traditional methods often require separate topographic surveys and ground feature attribute investigations, while this system can collect abundant data in a single flight, improving work efficiency and providing a unified data foundation for subsequent comprehensive analysis.

[0017] The real-time point cloud generation module ensures the accuracy and reliability of 3D point cloud data through temporal alignment, noise filtering, and inertial data correction. Disaster site environments are complex and data is easily affected by interference; this module's processing effectively eliminates various sources of error, generating high-quality point clouds and laying a solid foundation for refined feature recognition and modeling.

[0018] The terrain feature extraction module combines point cloud geometric features and multispectral data to achieve scientific zoning of geological stability. This method not only focuses on the macroscopic morphology of the terrain but also utilizes spectral information to reflect the internal properties of the rock and soil mass, making the stability zoning results more comprehensive and more consistent with geological laws.

[0019] The dynamic modeling engine module employs an adaptive octree algorithm, which dynamically adjusts the local model resolution based on features such as fault zone density. High resolution is used in complex regions to preserve details, while lower resolution is used in flat regions to reduce data redundancy. This non-uniform modeling strategy optimizes the model's data volume and processing efficiency while ensuring no key information is lost.

[0020] The disaster evolution analysis module compares historical and current data to quantitatively calculate surface displacement and subsidence gradient, enabling dynamic assessment of disaster development trends and prediction of secondary risks. This function elevates surveying and mapping from static description to dynamic analysis, providing crucial decision support information for disaster early warning and engineering management, significantly enhancing the system's practical value. This system provides a powerful technical tool for disaster emergency response and risk management. Attached Figure Description

[0021] Figure 1 A comparison chart showing the effects of processing terrain data at disaster sites; Figure 2 A flowchart illustrating the workflow of the disaster area scanning module; Figure 3 A flowchart illustrating the workflow of the terrain feature extraction module; Figure 4 This is a distribution map of slope slip probability and collapse risk index. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Please see Figure 1 This invention provides a UAV-based 3D disaster site mapping system. The system includes: a disaster area scanning module that controls a UAV equipped with a multispectral sensor to perform omnidirectional scanning of the disaster site, collecting surface deformation data and multi-band reflectivity data to generate an original terrain scan dataset; a real-time point cloud generation module that performs temporal alignment and noise filtering on the original terrain scan dataset, constructs an initial 3D point cloud through dynamic voxelization, and integrates inertial measurement unit data to correct spatial position deviations; a terrain feature extraction module that identifies surface fault zone edge features and elevation change features from the initial 3D point cloud, and divides geological stability zones based on multi-band reflectivity data; a dynamic modeling engine module that uses an adaptive octree segmentation algorithm to perform multi-level detail subdivision of the geological stability zones, automatically adjusts the local mesh resolution according to fault zone density, and generates a 3D topological structure of the disaster site; and a disaster evolution analysis module that compares historical mapping data with the current 3D topological structure to calculate the surface displacement vector field and subsidence gradient field, marking potential secondary disaster risk areas.

[0024] Example 1: See Figure 2The flight path planning unit of the disaster area scanning module automatically selects the scanning path mode based on the input disaster type parameters. For large-area collapses or debris flows, a spiral scanning path is used; for linear structures such as earthquake fault zones, a raster scanning path is used. The flight path planning unit integrates a digital elevation model data preloading function, avoiding known high obstacles during the planning phase. The real-time adjustment logic for flight altitude and sensor tilt angle relies on real-time terrain perception data. The UAV-mounted lidar performs a pre-scan of the terrain ahead to generate a local elevation map. The flight control system dynamically calculates the optimal flight altitude based on the local elevation map to maintain a constant relative distance to the ground surface. The sensor tilt angle control servo mechanism adjusts according to flight speed and distance to ensure that the image overlap rate and point cloud density meet preset specifications. The multi-source data synchronization unit is responsible for handling the fusion of three types of asynchronous data: lidar point cloud data, visible light image data, and infrared thermal imaging data. A high-precision atomic clock is deployed within the unit as a time reference, adding a microsecond-level timestamp to each data frame. The spatial coordinates of the lidar point cloud data are directly derived from the pose calculation of the laser ranging and inertial measurement unit. Visible light image data and infrared thermal imaging data are geometrically corrected through collinearity equations and projected onto the same spatial coordinate system. The spatial registration process uses a scale-invariant feature transformation algorithm to extract feature points from the visible light image and match them with the intensity information feature points in the lidar point cloud to calculate the accurate transformation matrix between the image data and the point cloud data. The infrared thermal imaging data is aligned with the visible light image through its optical system calibration parameters, achieving complete synchronization of the three types of data in the spatiotemporal dimensions.

[0025] The scan quality control unit continuously runs an integrity assessment algorithm during data acquisition. This algorithm is based on a real-time generated sparse point cloud density distribution map and compares it with a preset density threshold. Simultaneously, visible light image data is stitched in real-time to generate an orthophoto image, and its image overlap is analyzed. When a point cloud density below the threshold or an occluded area is detected, the scan quality control unit sends a supplementary scan command to the flight path planning unit. This command includes the boundary coordinates of the occluded area and a suggested supplementary scan path. Upon receiving the command, the flight path planning unit interrupts its current scan task and prioritizes supplementary scanning of the missing area. Once the data quality meets the standard, the original scan path is resumed. The motion distortion correction unit of the real-time point cloud generation module specifically addresses the point cloud distortion problem caused by the UAV platform's motion. The inertial measurement unit outputs the UAV's angular velocity and linear acceleration data at high frequency. The motion distortion correction unit uses this data to reconstruct the UAV's instantaneous attitude at each laser pulse emission moment. The coordinates of each point on the lidar are transformed based on its precise emission timestamp and the corresponding instantaneous attitude data to eliminate the point cloud stretching or compression distortion caused by UAV flight and attitude changes. This process requires a precise time synchronization system to ensure that the time alignment accuracy between the inertial measurement unit data and the lidar data is within milliseconds.

[0026] The multi-frame fusion unit processes consecutive temporal scan frames after motion distortion correction. The iterative nearest-neighbor algorithm is the core of multi-frame fusion. The algorithm selects one frame of point cloud as the reference frame and subsequent frames as the frames to be registered. By finding the nearest neighbor points, it calculates the rotation matrix and translation vector, gradually transforming the multi-frame point cloud into the coordinate system of the reference frame. The multi-frame fusion unit adopts a hierarchical registration strategy, first performing coarse registration to reduce the initial position error, and then performing fine registration to optimize the registration accuracy. The amount of dense point cloud data after fusion is significantly increased. The unit also manages the storage and indexing of point cloud data, providing a complete point cloud dataset with a unified coordinate system for subsequent processing steps. The dynamic noise reduction unit operates on the fused dense point cloud. Its algorithm is based on the principle of local surface fitting, fitting a local surface model for each point in the point cloud in its neighborhood. Usually, the least squares method is used to fit a local plane or quadratic surface. The distance from each point to its fitted surface is calculated as the fitting error. The magnitude of the fitting error reflects the geometric consistency between the point and its surrounding points. The dynamic noise reduction unit sets an adaptive error threshold, and points with fitting errors exceeding the threshold are identified as outliers and removed from the point cloud data. The size of the local neighborhood is dynamically adjusted according to the point cloud density, using a larger neighborhood in sparse point cloud areas and a smaller neighborhood in dense point cloud areas to ensure noise reduction while preserving realistic terrain features. The collaborative work of the disaster area scanning module and the real-time point cloud generation module constitutes the basic link for 3D mapping data acquisition and preprocessing. The intelligent path generation of the flight path planning unit, the high-precision registration of the multi-source data synchronization unit, the real-time monitoring and feedback of the scan quality control unit, the motion error elimination of the motion distortion correction unit, the data integration of the multi-frame fusion unit, and the data cleaning of the dynamic noise reduction unit are all interconnected, forming a complete data pipeline.

[0027] Example 2: See Figure 3 The fault zone identification unit of the terrain feature extraction module analyzes surface discontinuities based on the geometric attributes of the initial 3D point cloud. The unit constructs a k-dimensional tree spatial index for the point cloud data to support efficient neighborhood queries. For each point, it calculates the eigenvalues ​​of the covariance matrix of its local neighborhood points, deriving the principal curvature value from these eigenvalues. Principal curvature extrema identify areas of dramatic surface curvature changes, typically corresponding to potential fault zone edges. The fault zone identification unit sets a curvature threshold to initially screen points with principal curvature exceeding the threshold as candidate fault points. Multi-band reflectance data provides spectral information for verification. The reflectance in the visible and near-infrared bands often exhibits abrupt changes on features on both sides of the fault zone. The fault zone identification unit overlays the spatial locations of candidate fault points with the multi-band reflectance images. Only points that simultaneously meet the conditions of geometric curvature extrema and abrupt spectral reflectance changes are ultimately confirmed as geological fault boundary points. These boundary points are connected to form a continuous fault zone vector line.

[0028] The elevation clustering unit processes the elevation values ​​of the initial 3D point cloud. The unit employs a density peak clustering algorithm for unsupervised classification of the elevation data. The algorithm first calculates the local density of the elevation value of each point and the elevation values ​​of points within a certain range around it, while simultaneously calculating the distance from each point to its higher-density points. The elevation clustering unit identifies points with high local density and large distances from higher-density points as cluster centers. These cluster centers naturally correspond to the core regions of terrain units at different elevations. All points are divided into different terrain blocks based on their distance from the cluster centers, such as flat areas, gentle slope areas, steep slope areas, and subsidence areas. The elevation clustering unit calculates elevation statistics for each block, including mean elevation, elevation standard deviation, and block boundary slope. Based on these features, significant steep slopes and subsidence areas are marked, providing topographic morphological evidence for stability analysis.

[0029] The stability grading unit integrates the outputs of the fault zone identification unit and the elevation clustering unit to quantitatively assess geological stability. The unit divides the scanned area into a regular grid, calculates the total length of fault zones within each grid cell to obtain the fault zone distribution density, and simultaneously calculates the coefficient of variation of point cloud elevation values ​​within each grid cell to characterize the degree of surface undulation. After normalization, the fault zone distribution density and elevation coefficient of variation are input into a weighted scoring model. The stability grading unit assigns weights to the fault zone density and elevation coefficient of variation, and the weighted sum is used to obtain the stability score for each grid cell. The scoring results are divided into multiple consecutive levels, such as stable, relatively stable, relatively unstable, and unstable. The stability grading unit generates a geological stability level map by color-coding the level information of each grid cell. The map visually displays the geological stability status at different locations throughout the disaster area in raster image form.

[0030] The detail level partitioning unit of the dynamic modeling engine module is responsible for converting 3D point cloud data with accompanying stability grading information into an octree data structure that supports multi-resolution representation. The unit initializes a cube that can enclose the entire point cloud scene as the root node of the octree, and then recursively partitions the cube into eight child nodes. The subdivision decision of the detail level partitioning unit depends on the point cloud density within the node and the geological stability level of the area covered by the node. For node regions with high point cloud density and low geological stability, deeper subdivision operations are performed. For example, the mesh resolution is increased in areas surrounding fault zones to capture finer geometric features, while a coarser mesh resolution is maintained in areas with sparse point clouds and high stability. This adaptive subdivision strategy achieves optimized spatial allocation of computational resources, controlling the overall data volume without sacrificing details in important regions.

[0031] The topology reconstruction unit generates a triangular mesh model based on an octree structure. The unit employs the Poisson surface reconstruction algorithm to reconstruct a continuous, watertight surface model from point cloud data. The Poisson surface reconstruction algorithm converts the point cloud data into an indicator function; the zero isosurface of this function is the desired surface. The algorithm reconstructs this indicator function by solving the Poisson equation. During the processing, the topology reconstruction unit pays special attention to preserving the geometric discontinuities of fracture edges. The unit identifies characteristic edges by analyzing abrupt changes in point cloud normal vectors and constrains these edge lines from smoothing during meshing, thus ensuring that the final generated mesh model clearly displays discontinuous terrain features such as surface fractures and steep slopes. The reconstructed mesh model is a closed three-dimensional surface composed of vertices and triangular facets, representing the three-dimensional topological structure of the disaster site.

[0032] The texture mapping unit is responsible for attaching realistic surface texture information to the generated 3D topological model, using multi-band reflectance data as the texture source. The unit needs to solve the mapping problem between the 3D mesh surface and the 2D texture image. It establishes the UV mapping relationship by calculating the corresponding pixel coordinates of each mesh vertex in the original multispectral image. For each triangular facet, the texture mapping unit interpolates the texture coordinates corresponding to its vertices, thus accurately projecting the color information from the 2D image onto the 3D mesh surface. The texture mapping unit supports the generation of high-resolution texture maps, which can fuse reflectance information from multiple bands to generate natural true-color textures or enhanced false-color textures, significantly improving the visualization effect and feature identification capability of the 3D model. The joint work of the terrain feature extraction module and the dynamic modeling engine module completes the transformation from raw point clouds to a refined 3D model with semantic information and realistic texture. The fault zone identification unit reveals the weaknesses of the surface structure, the elevation clustering unit outlines the terrain pattern, the stability grading unit makes an initial risk assessment, the detail level division unit constructs a multi-scale geometric framework, the topology reconstruction unit generates a continuous surface, and the texture mapping unit gives the model a sense of realism. This series of processing steps makes the 3D model not only geometrically accurate, but also carries rich geological stability information.

[0033] Example 3: The displacement field calculation unit processes two 3D mesh models collected and constructed at different time points. The unit first performs initial registration on the two models, using common, unchanged stable region features for coarse alignment to reduce the impact of overall positional deviations on subsequent point matching. The feature point extraction process is based on the mesh's geometric properties. The displacement field calculation unit calculates the curvature of the normal vector at each vertex, selecting extreme points of average curvature or Gaussian curvature as candidate feature points. These points are located on terrain feature lines such as ridges, valleys, and slope breaks, exhibiting significant geometric characteristics and relative stability in the short term. For each extracted feature point, the displacement field calculation unit constructs a rotational projection statistical descriptor in its surrounding neighborhood. The descriptor characterizes the local 3D shape through the distribution of the normal vector within the neighborhood of the statistical point in multiple concentric spherical shells, exhibiting invariance to model rotation and translation transformations. The matching algorithm calculates the similarity between the feature point descriptors of the historical model and the current model, finding the most similar corresponding point in the current model for each feature point in the historical model. Successfully matched point pairs constitute the basic dataset for calculating surface displacement. The displacement field calculation unit calculates the displacement vector of each point based on the three-dimensional coordinates of the matching point pairs. The direction of the vector represents the direction of surface movement, and the magnitude of the vector represents the amount of displacement. The set of displacement vectors of all feature points forms a discrete displacement vector field that describes the surface deformation of the entire disaster area.

[0034] The gradient analysis unit performs spatial differentiation on the discrete displacement vector field generated by the displacement field calculation unit to analyze the spatial rate of change of deformation. The unit focuses on analyzing the settlement component, which is of significant indicative importance for disaster assessment. The gradient analysis unit treats the settlement displacement value of each grid vertex as a scalar field defined on the three-dimensional surface and calculates the gradient of this scalar field along the tangent plane of the surface. The gradient is a vector pointing in the direction of the fastest increase in settlement, and its magnitude represents the severity of the settlement change. The gradient analysis unit uses a finite difference method based on triangular meshes for calculation. For each vertex in the mesh, the gradient analysis unit finds all its neighboring vertices to construct a local neighborhood. The settlement gradient magnitude is calculated using the following mathematical model:

[0035] in: The magnitude of the settlement gradient at vertex v is a dimensionless scalar. This indicates the settlement displacement of the vertex, in meters. and Represents the Cartesian coordinates of the vertex on the horizontal projection plane, in meters. This represents the partial derivative of the settlement s along the X-axis of the plane coordinate system. This represents the partial derivative of the settlement s along the Y-axis of the plane coordinate system; both are dimensionless. The gradient analysis element calculates the settlement gradient magnitude at each grid vertex within the disaster area. A larger modulus value indicates a higher rate of subsidence change and more uneven surface deformation in the surrounding area. The gradient analysis unit sets a gradient threshold based on historical experience and geological conditions. Areas with values ​​exceeding the threshold are marked as abnormally accelerated settlement areas, which are important precursors to potential secondary disasters.

[0036] The risk prediction unit integrates displacement vector information from the displacement field calculation unit and gradient information from the gradient analysis unit, combined with an external soil mechanics parameter library, to conduct quantitative risk assessment. The risk prediction unit operates using a grid as the basic unit. The soil mechanics parameter library stores typical physical and mechanical parameters of soil and rock masses in the disaster area, including indicators such as internal friction angle, cohesion, unit weight, and pore water pressure coefficient. The risk prediction unit performs slope stability calculations on steep slope areas identified by the elevation clustering unit. The calculation adopts the simplified Bishop slice method, treating potential slip surfaces as circular arcs, vertically slices the sliding body, and considering the interaction forces between slices. The risk prediction unit calculates the safety factor for each potential slip surface based on the soil and rock parameters provided by the soil mechanics parameter library. The safety factor is defined as the ratio of the resisting moment to the sliding moment. The risk prediction unit further employs reliability analysis methods to calculate the slope slip probability, considering the spatial variability and measurement uncertainty of soil and rock parameters. A large number of parameter combinations are randomly generated using Monte Carlo simulation, and the corresponding safety factors are calculated. The frequency of safety factors less than 1 is used as an estimate of the slip probability. For anomalously accelerated settlement areas identified by the gradient analysis unit, the risk prediction unit focuses on analyzing the spatiotemporal evolution trend of settlement. The unit calls upon historical settlement data from multiple periods, fits the settlement amount of each grid vertex to a curve over time, and calculates the settlement acceleration. The risk prediction unit ultimately outputs two core risk indicators for each assessment unit: slope slip probability and collapse risk index. The slope slip probability is a value between 0 and 1, quantitatively representing the likelihood of sliding failure in that unit. The collapse risk index is a comprehensive indicator, obtained by weighted fusion of multiple factors such as displacement magnitude, settlement gradient modulus, settlement acceleration, slope gradient, and geotechnical conditions. A higher index value indicates a greater risk of collapse in that area. The disaster evolution analysis module captures the basic patterns of surface movement through the displacement field calculation unit, identifies spatial anomalies in deformation rates through the gradient analysis unit, and performs quantitative stability evaluation by integrating a geomechanical model through the risk prediction unit. The module's output, presented as a spatial distribution map, visually displays the location and hazard level of potential secondary disaster risk areas within the disaster site, providing quantitative spatial information support for disaster early warning and emergency decision-making. The entire analysis process establishes a complete technical chain from 3D model comparison to quantitative risk assessment, enabling the perception, quantification, and prediction of the dynamic evolution of disasters.

[0037] See Figure 4This diagram, presented in two sub-plots, systematically illustrates the geological risk situation in different areas of a disaster site from a quantitative perspective. The upper bar chart focuses on the probability of slope slippage, intuitively reflecting the extremely high likelihood of landslide failure in these areas; area E indicates relatively strong stability. The lower line chart displays the collapse risk index, with area D ranking first in risk, while areas A and B are also in the high-risk range; area E has extremely low risk. This dual-dimensional visualization not only reflects the quantitative results of surface deformation from the displacement field calculation unit but also integrates the soil mechanics model analysis from the risk prediction unit, clearly demonstrating the correlation logic from topographic features to risk levels. It provides precise spatial decision-making basis for disaster secondary risk early warning and emergency resource allocation, realizing the visualized implementation of the perception, quantification, and prediction of the dynamic evolution of disasters.

[0038] Example 4: The hardware foundation of the module is a multi-mode communication radio mounted on an UAV platform. The radio supports wireless communication protocols across multiple frequency bands, including VHF, L-band, and satellite communication links. The software system of the emergency communication relay module runs on the UAV's onboard computer, directly interacting with the dynamic modeling engine module and the disaster evolution analysis module to acquire the required 3D topology data and disaster risk analysis results. In the initial stage of establishing the communication link, the emergency communication relay module negotiates with the ground command center, exchanging their respective communication capability parameters, including supported frequency bands, maximum bandwidth, and encoding methods. Based on the channel quality indicators monitored in real time by the network status sensing unit, the emergency communication relay module selects the optimal communication channel from the available frequency bands. The channel selection decision considers multiple factors such as signal strength, signal-to-noise ratio, bit error rate, and link stability. During data transmission, the emergency communication relay module maintains a dynamically updated link status table, recording the current link performance indicators and historical transmission statistics, providing a basis for adaptive coding and link switching decisions.

[0039] The data chunking unit divides the 3D model into transmission blocks of different priorities according to geological stability levels. The input to the data chunking unit is the 3D topological structure model generated by the dynamic modeling engine module and the geological stability level map generated by the terrain feature extraction module. The data chunking unit divides the entire 3D scene into regular spatial grid blocks. The priority of each grid block is determined by the unit with the highest geological stability level within it; blocks containing unstable regions are given the highest transmission priority. The data chunking unit generates metadata descriptions for each block, including the block's spatial extent, number of vertices, number of faces, texture data size, geological stability level, and transmission priority label. The data chunking unit maintains a transmission queue management module, organizing the block data to be transmitted according to priority, with high-priority blocks placed at the front of the queue for priority transmission. The data chunking unit supports an incremental transmission mechanism. When the geological stability level map is updated or the 3D model is locally updated, the data chunking unit only needs to re-divide the changed areas to generate incremental data packets, reducing unnecessary data transmission. Referring to Table 1, the block division strategy of the data chunking unit is configured based on the following parameters.

[0040] Table 1: Data Transmission Priority Division Parameters

[0041] The adaptive coding unit dynamically adjusts the compression ratio of point cloud data to texture data based on channel bandwidth. It receives bandwidth measurement data from the communication link in real time, calculating round-trip latency and throughput by periodically sending probe packets. The adaptive coding unit designs independent compression pipelines for different data types. Point cloud data uses a progressive compression algorithm based on an octree structure, preserving different levels of detail and supporting layered transmission. Texture data compression uses a quality-scalable encoding format; the encoder dynamically adjusts quantization parameters based on the target bitrate, maintaining the most important visual information within limited bandwidth. Referring to the transmission priority table and real-time bandwidth information provided by the data chunking unit, the adaptive coding unit allocates an appropriate target bitrate to each data block, allocating more bandwidth resources to higher-priority blocks for higher-quality compression. The adaptive coding unit integrates multiple codec instances, including point cloud compression codecs, image compression codecs, and mesh compression codecs, selecting the optimal compression tool based on data type and channel conditions. The adaptive coding unit maintains a bitrate-distortion model, establishing the correspondence between data quality and file size under different compression parameters, providing a quantitative basis for bitrate allocation decisions.

[0042] When a single communication frequency band is blocked, the link switching unit automatically switches to a backup frequency band to maintain data transmission. The link switching unit continuously monitors the performance indicators of the currently used link, including received signal strength indication, signal-to-noise ratio, bit error rate, and link throughput. The link switching unit sets multiple threshold conditions to trigger switching. When the performance indicators of the primary link fall below the threshold for a certain period, the link switching unit initiates the switching process. The link switching unit first scans all available backup frequency bands, evaluates the connection quality and expected available bandwidth of each alternative link, and selects the optimal backup link as the switching target. The link switching unit employs a switching strategy combining hard and soft switching. For data transmission with low real-time requirements, a hard switching method is used, completing the transmission of the current data packet before switching, and establishing a new link after the connection is interrupted. For critical data with high real-time requirements, the link switching unit attempts to establish parallel transmission paths, gradually disconnecting the old link after the new link is successfully established, reducing data transmission interruption time. The link switching unit records an event log for each link switching, including information such as switching time, switching reason, original link quality, target link quality, and a comparison of data transmission integrity before and after switching. These logs are used to analyze link stability and optimize the switching strategy.

[0043] The three units of the emergency communication relay module work collaboratively to form a complete data transmission solution. The data segmentation unit identifies the differences in data importance from the application layer perspective, providing a foundation for differentiated transmission. The adaptive encoding unit optimizes representation efficiency from the data layer perspective, enabling limited bandwidth resources to carry more effective information. The link switching unit ensures the reliability of the transmission path from the physical link layer perspective, coping with the complex electromagnetic environment at disaster sites. The three units communicate and exchange data through an internal message bus, which transmits link status notifications, data priority information, compression parameter configurations, and switching control commands. The data transmitted by the emergency communication relay module to the ground command center includes complete 3D model structure information, texture mapping data, geological stability classification results, and disaster risk markers. After receiving the data, the ground command center can reconstruct the 3D scene of the disaster site and identify high-risk areas to support emergency decision-making. The design of the emergency communication relay module fully considers the practical challenges of potential damage to communication infrastructure at disaster sites. Utilizing the mobility and multi-mode communication capabilities of the UAV platform, it establishes a flexible and reliable data transmission channel, ensuring that key mapping and analysis results can be delivered to the command center in a timely manner. The module is implemented using software-defined radio technology, which enables support for different communication protocols through software configuration. It has good scalability and adaptability, and can cope with the development and changes in future communication technologies.

[0044] Example 5: The work of the mapping verification module begins with the establishment and measurement of ground control points. Before the UAV scans the disaster site, operators will establish a series of distinctive ground markers within the scanning area. These markers are typically made of high-contrast target materials, such as red and white checkered boards, to ensure clear identification in aerial imagery. The spatial coordinates of each ground control point need to be accurately determined using high-precision measuring instruments, such as a real-time dynamic differential GPS receiver or a total station. The measured coordinate data serves as the true reference benchmark. The distribution of ground control points within the scanning area follows certain principles, covering the edges and center of the entire area, and appropriately densifying the points in areas with significant terrain changes, such as slope tops, slope toes, and both sides of fault zones. This allows for comprehensive verification of the model's accuracy at different locations. After establishment, the spatial coordinates, point number, and location descriptions of the ground control points are entered into a control point database for management. The database records the latitude and longitude coordinates and elevation of each control point in the WGS84 coordinate system, as well as the corresponding Cartesian coordinates.

[0045] The control point matching unit uses the spatial coordinates of known ground markers to inversely calculate the model coordinate system transformation parameters. It reads the measured coordinates of ground control points stored in the control point database and simultaneously extracts the corresponding model coordinates from the 3D topology model generated by the dynamic modeling engine module. Model coordinate extraction is performed through a combination of human-computer interaction and automatic recognition. The operator clicks on the center position of the identified ground control point target on the 3D model surface, and the software automatically records the 3D model coordinates of that point. The control point matching unit combines the measured coordinates of a set of ground control points with the model coordinates to form observation pairs and uses the least squares method to solve for the transformation parameters between the two coordinate systems. The transformation model typically uses a seven-parameter similarity transformation model, including three translation parameters, three rotation parameters, and one scaling parameter. The control point matching unit constructs an error equation and iteratively calculates to minimize the sum of squared residuals on all control points. The control point matching unit performs gross error detection, using residual analysis to identify control points with mismatches or excessive measurement errors, discarding them and recalculating the transformation parameters until a stable solution is obtained. The seven parameters obtained from the solution are saved as model accuracy correction parameters, which are used to transform the entire 3D model into the real world coordinate system.

[0046] The error distribution unit calculates the root mean square error (RMSE) for different regions of the model and generates an accuracy heatmap. After the control point matching unit completes the coordinate transformation parameter calculation, the error distribution unit statistically analyzes the residuals at each ground control point. The residual is the difference between the measured coordinates of the ground control point and the model coordinates transformed by the transformation parameters. The error distribution unit calculates the planar position residual and the elevation residual separately. The error distribution unit uses spatial interpolation methods, such as Kriging interpolation or inverse distance weighted interpolation, to generate a continuous error distribution surface for the entire model region based on the discretely distributed control point residual data. The interpolation process considers the influence of terrain undulations on the error distribution. In areas with large slope variations, the range parameters of the interpolation algorithm are appropriately adjusted to make the generated error surface more consistent with the actual spatial distribution. The error distribution unit maps the estimated error value of each grid vertex to a color code, generating an intuitive accuracy heatmap. The heatmap uses a gradient from green to red to represent the change in error from small to large, with green areas indicating high accuracy and red areas indicating low accuracy. The high-precision heatmap is overlaid with the 3D model, allowing users to quickly understand the reliability of different areas of the model by viewing the heatmap, providing an important reference for subsequent data use and analysis.

[0047] The parameter optimization unit automatically adjusts the iterative convergence threshold of the point cloud registration algorithm based on the accuracy heatmap. It analyzes the error distribution pattern reflected in the accuracy heatmap to identify areas with systematic biases. The parameter optimization unit interacts with the multi-frame fusion unit of the real-time point cloud generation module to adjust the registration parameters in the iterative nearest-point algorithm. When the accuracy heatmap shows local stretching or compression deformation in the model, the parameter optimization unit lowers the iterative convergence threshold of the point cloud registration in the corresponding area, forcing the registration algorithm to perform more refined optimization in these areas. The parameter optimization unit also considers the impact of the distribution density of ground control points on the parameter adjustment range, adopting a more aggressive parameter adjustment strategy in areas with dense control points and a more conservative strategy in areas with sparse control points. The parameter optimization unit sends the adjusted registration parameters to the real-time point cloud generation module, triggering the model's re-registration and reconstruction process, forming a closed-loop accuracy optimization cycle. The reconstructed 3D model after parameter optimization is again evaluated for accuracy by the mapping and verification module. If the accuracy indicators still do not meet the requirements, the parameter optimization unit will initiate a new round of parameter adjustments until the overall and local accuracy of the model reaches the preset standards.

[0048] To illustrate the practical application of the above process, consider a landslide disaster survey where 15 ground control points were established in the landslide area and surrounding stable regions. The control point matching unit used the measured coordinates and model coordinates of these 15 points to calculate coordinate system transformation parameters. The error distribution unit found that the planar residuals of the three control points at the rear edge of the landslide were significantly greater than those of the other points. The accuracy heatmap clearly showed that the rear edge of the landslide area was marked in red, indicating a large error in the model for that area. The parameter optimization unit analyzed that this was due to insufficient point cloud registration caused by landslide movement. Therefore, it adjusted the convergence threshold of the iterative nearest point algorithm in the multi-frame fusion unit of the real-time point cloud generation module for processing the landslide area from the default 0.01 meters to 0.005 meters and increased the upper limit of the number of iterations. The real-time point cloud generation module re-registers the point cloud and performs 3D modeling using optimized parameters. The newly generated model is then verified by the surveying and mapping verification module. The results show that the color of the accuracy heatmap of the landslide's rear edge area changes from red to yellow, and the planar position error decreases from an average of 0.15 meters to 0.08 meters, significantly improving the overall accuracy of the model. The surveying and mapping verification module establishes a coordinate relationship between the model and the real world through control point matching units, visualizes the spatial changes in model accuracy through error distribution units, and adjusts modeling parameters through parameter optimization units, forming a complete accuracy verification and improvement closed loop. This ensures that the final generated 3D model has reliable geometric accuracy, meeting the accuracy requirements for disaster emergency response and subsequent engineering analysis.

[0049] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0050] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A three-dimensional mapping system for disaster sites based on unmanned aerial vehicles (UAVs), characterized in that, The system includes: The disaster area scanning module is used to control the UAV equipped with a multispectral sensor to perform a comprehensive scan of the disaster site, collect surface deformation data and multi-band reflectivity data, and generate a raw terrain scan dataset. The real-time point cloud generation module is used to perform temporal alignment and noise filtering on the original terrain scan dataset, construct the initial 3D point cloud through dynamic voxelization, and fuse inertial measurement unit data to correct spatial position deviations. The terrain feature extraction module is used to identify the edge features of surface fault zones and abrupt elevation changes from the initial 3D point cloud, and to divide the geological stability zones by combining multi-band reflectance data. The dynamic modeling engine module uses an adaptive octree segmentation algorithm to perform multi-level detailed subdivision of geological stability zones, and automatically adjusts the local mesh resolution according to the fault zone density to generate a three-dimensional topological structure of the disaster site. The disaster evolution analysis module calculates the surface displacement vector field and settlement gradient field by comparing historical mapping data with the current three-dimensional topology, and marks potential secondary disaster risk areas.

2. The UAV-based three-dimensional disaster site mapping system according to claim 1, characterized in that, The disaster area scanning module includes: The flight path planning unit automatically generates spiral or grid scanning paths based on the type of disaster and adjusts the flight altitude and sensor tilt angle in real time. The multi-source data synchronization unit performs timestamp alignment and spatial registration of lidar point cloud data, visible light image data and infrared thermal imaging data; The scan quality control unit monitors the integrity of data acquisition in real time and triggers supplementary scan commands when there are obstructed areas.

3. The UAV-based three-dimensional disaster site mapping system according to claim 2, characterized in that, The real-time point cloud generation module includes: The motion distortion correction unit uses angular velocity data from the inertial measurement unit to compensate for point cloud distortion caused by changes in the attitude of the UAV. The multi-frame fusion unit uses an iterative nearest-point algorithm to convert continuous temporal scan frames into a dense point cloud in a unified coordinate system. The dynamic noise reduction unit automatically identifies and removes outlier noise points based on local surface fitting errors.

4. The UAV-based three-dimensional disaster site mapping system according to claim 3, characterized in that, The terrain feature extraction module includes: The fault zone identification unit detects areas of discontinuity on the earth's surface by calculating the extreme value of the principal curvature, and confirms the geological fault boundary by combining the reflectance abrupt change threshold. The elevation clustering unit uses the density peak clustering algorithm to separate terrain blocks at different elevations and marks steep slopes and subsidence areas; The stability grading unit generates a geological stability level map based on the distribution density of fault zones and the coefficient of variation of elevation.

5. The UAV-based three-dimensional disaster site mapping system according to claim 4, characterized in that, The dynamic modeling engine module includes: The detailed level of the subdivision unit is automatically selected according to the point cloud density in the octree node, and the subdivision level is improved in the area around the fault zone. The topology reconstruction unit uses the Poisson surface reconstruction algorithm to generate a closed mesh model, preserving the geometric discontinuities at the fracture edges; The texture mapping unit projects multi-band reflectivity data onto the surface of the mesh model to generate a high-resolution texture map.

6. The UAV-based three-dimensional disaster site mapping system according to claim 5, characterized in that, The disaster evolution analysis module includes: The displacement field calculation unit obtains the displacement of corresponding points in the historical and current 3D models through a feature point matching algorithm; Gradient analysis unit calculates the spatial distribution of settlement rate and identifies areas of abnormally accelerated settlement. The risk prediction unit, combined with the soil mechanics parameter library, assesses the probability of slope slippage and the collapse risk index.

7. The UAV-based three-dimensional disaster site mapping system according to claim 6, characterized in that, Also includes: The emergency communication relay module establishes a low-latency data link between the UAV and the ground command center, transmitting compressed 3D model key feature data in real time.

8. The UAV-based three-dimensional disaster site mapping system according to claim 7, characterized in that, The emergency communication relay module includes: Data segmentation units divide the 3D model into transmission blocks of different priorities according to the geological stability level; An adaptive coding unit dynamically adjusts the compression ratio of point cloud data to texture data based on the channel bandwidth. The link switching unit automatically switches to a backup frequency band to maintain data transmission when a single communication frequency band is blocked.

9. The UAV-based three-dimensional disaster site mapping system according to claim 8, characterized in that, Also includes: The surveying and verification module generates accuracy correction parameters for the 3D model by analyzing the residuals between the measured coordinates of ground control points and the model coordinates.

10. The UAV-based three-dimensional disaster site mapping system according to claim 9, characterized in that, The mapping verification module includes: The control point matching unit uses the spatial coordinates of known ground markers to inversely calculate the model coordinate system transformation parameters. Error distribution unit calculates the root mean square error of different regions of the model and generates an accuracy heatmap; The parameter optimization unit automatically adjusts the iterative convergence threshold of the point cloud registration algorithm based on the accuracy heatmap.

Images