A method and system for generating a three-dimensional model of a water conservancy dam

Abstract

The application relates to the technical field of water conservancy dam detection and modeling, in particular to a method and system for generating a simulated three-dimensional model of a water conservancy dam, which comprises the following steps: a high-altitude acquisition platform dynamically scans a target dam along a planned path to obtain a multi-source visual data stream; a central processing unit extracts texture features, geometric contour lines, structural joint lines and material reflection characteristics; multi-dimensional similarity matching and two-stage correlation analysis are carried out based on a pre-constructed water conservancy dam component database to identify a model component set and a topological relationship thereof; standardized components are called and instantiated, initial assembly and positioning are carried out according to spatial constraint conditions to form an initial three-dimensional geometric framework; and optimization processing such as material adjustment, geometric smoothing and texture mapping is carried out to generate a structured and parameterized three-dimensional model; the application realizes automatic modeling from unstructured visual data to component-level models, and improves the timeliness, engineering semantic integrity and reconstruction accuracy of the three-dimensional model of the dam.

Description

Technical Field

[0001] This invention relates to the field of water conservancy dam detection and modeling technology, and in particular to a method and system for generating a three-dimensional model of a water conservancy dam simulation. Background Technology

[0002] In the field of water conservancy engineering, high-precision, highly realistic 3D models are crucial for dam design, construction monitoring, operation and maintenance, safety assessment, and digital management. Traditional 3D model construction methods mainly rely on manual on-site surveys, using equipment such as total stations to collect limited feature points, and then manually modeling in computer-aided design software in conjunction with design drawings. This method has significant limitations: First, it is inefficient and costly, especially for large-scale or remote water conservancy projects, where organizing large-scale on-site measurements is difficult and time-consuming; second, the model lacks timeliness, making it difficult to quickly respond to changes in the dam's shape and material due to natural aging, geological disasters, or operational wear and tear; third, the constructed models are mostly surface geometric shells, lacking structured and parametric information corresponding to the actual engineering components, limiting their value in in-depth engineering applications such as quantity calculation, structural analysis, and component management.

[0003] With the development of remote sensing and surveying technologies, large-scale and rapid 3D reconstruction using UAV oblique photogrammetry or airborne lidar has become a common method. These technologies can efficiently acquire massive point cloud and texture images of dams and surrounding areas and automatically generate realistic 3D models. However, the 3D models generated by such methods are essentially "surface models" or "real-world models" based on continuous triangular facets. Although visually realistic, the models themselves are holistic and unstructured. Specifically, the models do not deconstruct the dam into independent, engineering-semantic components such as piers, spillways, corridors, and wave walls. Therefore, the models cannot support independent editing of specific components, attribute queries, standard library-based replacements, or performance simulation analysis. Furthermore, when facing complex structures, occlusions, or scenarios requiring the integration of multi-period, multi-source data (such as historical design models and the latest monitoring data), existing methods struggle to achieve automatic and accurate stitching and fusion of different parts of the model under a unified, high-precision spatial coordinate system. Significant manual intervention is often still required for post-processing, making it difficult to guarantee the consistency within the model. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method and system for generating a three-dimensional model of a hydraulic engineering dam to solve the problems mentioned in the background art.

[0005] The main objective of this application is to provide a method for generating a 3D model of a hydraulic engineering dam, comprising the following steps: using at least one high-altitude acquisition platform deployed on a mobile vehicle, dynamically scanning and photographing the target hydraulic dam structure along a predetermined or adaptively planned set of spatial angles and relative height datasets to obtain a multi-source visual data stream containing different facades and details of the dam structure; wherein, the high-altitude acquisition platform is configured to receive and execute motion trajectory instructions from a central processing unit; The central processing unit receives and processes the multi-source visual data stream, and extracts the texture features, geometric contours, structural joints, and material reflection properties of the dam surface by extracting continuous video frames and analyzing frame images. Based on a pre-built database of hydraulic engineering dam components, the extracted texture features and geometric contours are matched and analyzed in multiple dimensions with the standardized dam component models in the database to identify the set of potential model components corresponding to the current dam structure and their topological relationships. Based on the results of the correlation analysis, the corresponding standardized dam component model is called from the database and instantiated. Based on the spatial constraints determined by the geometric contour lines, the initial assembly and positioning are carried out in the three-dimensional modeling environment to form the initial three-dimensional geometric framework of the dam structure. The initial three-dimensional geometric framework is optimized by: adjusting the material parameters of the model surface based on the extracted material reflection characteristics; smoothing the geometric contours and repairing gaps based on multi-angle visual data; and mapping high-resolution textures onto the corresponding model surface.

[0006] Furthermore, methods for multi-dimensional similarity matching and association analysis based on a pre-constructed database of hydraulic engineering dam components include: A first-level association is established by performing feature point matching and spatial analysis calculation between the data collected by the high-altitude acquisition platform at the first key visual marker point and the first group of components with known geometric attributes in the database. The central processing unit then determines at least one first spatial pose of the high-altitude acquisition platform relative to the dam structure. A second-level association is established by performing association analysis between data collected by the high-altitude acquisition platform at the second key visual marker point and a target component with unknown geometric attributes. This is combined with the attitude of the high-altitude acquisition platform determined in the first-level association, the known spatial positions of the first group of components, and other components with known geometric attributes in the database. The central processing unit then calculates the precise three-dimensional geometric dimensions and spatial positioning of the target component. The first key visual marker point and the second key visual marker point may be at the same or different spatial locations. The method for identifying key visual markers and determining platform pose includes: the central processing unit determining the theoretical position of at least one key visual marker in the real-world coordinate system based on pre-input dam design drawings or topographic map data; The central processing unit or the high-altitude acquisition platform itself uses a visual positioning algorithm to determine when the high-altitude acquisition platform arrives at or aligns with the effective data acquisition range of the key visual markers. The system controls the high-altitude acquisition platform to interact with the first set of field acquisition sensors located at known positions on or around the dam, or directly analyzes images containing the known first set of components through visual algorithms to obtain a series of distance or viewing angle measurements. Based on the set of measured values ​​and the known coordinates of the first set of field acquisition sensors or the known coordinates of the first set of components, spatial resection or multi-view geometric calculation is performed to determine at least one first spatial pose of the high-altitude acquisition platform at the key visual marker point.

[0007] Furthermore, the method also includes dynamic correction of database component parameters: when the target component already has an initial geometric parameter estimate stored in the database, the central processing unit compares the calculated three-dimensional geometric dimensions and spatial positioning with the initial geometric parameter estimate stored in the database. If the difference between the three-dimensional geometric dimensions and spatial positioning and the initial geometric parameter estimate exceeds a preset tolerance threshold, then the initial geometric parameter estimate in the database is updated using the three-dimensional geometric dimensions and spatial positioning.

[0008] Furthermore, the method also includes the step of modeling the dam's auxiliary structures using a continuous pose sequence: by performing continuous frame feature tracking and spatial intersection calculation between the data collected at multiple continuous pose points and the first group of components with known geometric properties in the database when the high-altitude acquisition platform moves along a predetermined trajectory traversing the target area, the central processing unit determines the multiple continuous spatial poses of the high-altitude acquisition platform within the target area; By performing stereoscopic visual analysis on the data collected from the high-altitude acquisition platform under multiple continuous spatial poses and on the dam ancillary structures within a target area to be determined, the central processing unit determines the three-dimensional model of the dam ancillary structures and their spatial position relative to the main dam body.

[0009] Furthermore, the spatial resolution method for multi-angle data correlation analysis in 3D model generation includes: First-level analysis: Using the high-altitude acquisition platform as a mobile known observation station, through multiple observations of the platform with known location points or known model components on the dam body, the precise trajectory and attitude sequence of the platform itself can be solved in reverse. Second-level analysis: Based on the known platform attitude, it is regarded as a sensor with a known spatial position. Using its observation data of the target component, combined with the observation constraints of other known position points on the same unknown structure, the precise geometry and position of the unknown structure in three-dimensional space are calculated through spatial forward intersection or multi-view triangulation algorithms. Third-level analysis: When dealing with multiple structural modules whose spatial relationships are unknown, the high-altitude acquisition platform is used to cross-observe them at multiple locations. The obtained observation network data is used for overall adjustment, and the relative positions and directions of all unknown modules are calculated to construct a self-consistent global three-dimensional model.

[0010] The present invention also provides a system for generating a three-dimensional model of a hydraulic engineering dam, including a central processing unit, at least one high-altitude acquisition platform, and a pre-built database of hydraulic engineering dam components; The high-altitude acquisition platform is configured to be mounted on a mobile vehicle for acquiring image or video data of the target dam from multiple angles and heights. The central processing unit is communicatively connected to the high-altitude data acquisition platform and the database, and is configured to: By analyzing the data collected by the high-altitude acquisition platform at the first key visual marker point and performing feature point matching and spatial analysis calculation with the first group of components with known geometric properties in the database, the spatial pose of the high-altitude acquisition platform at the first key visual marker point is determined. By analyzing the data collected by the high-altitude acquisition platform at the second key visual marker point and performing correlation analysis with the second group of components, which includes at least one target component with geometric attributes to be determined and other components with known geometric attributes in the database, the three-dimensional geometric model of the target component and its location are determined. The first group of components and the second group of components have an inclusion relationship, the same set, or a non-intersecting relationship in the component classification hierarchy of the database. The first key visual marker point and the second key visual marker point are the same or different spatial location points. The central processing unit is configured to determine the expected spatial location range of the first key visual marker point based on the dam design blueprint or preliminary survey data. By analyzing the real-time video stream or image sequence transmitted back by the high-altitude acquisition platform, a visual recognition algorithm is used to automatically detect when the high-altitude acquisition platform enters the effective data acquisition window of the expected spatial location range; triggering the high-altitude acquisition platform to perform collaborative measurement with the first set of field acquisition sensors deployed at known locations on the dam body, or triggering high-precision analysis of images containing specific known components to obtain a set of observation data for spatial calculation; Based on the observation data and the known precise coordinates of the first set of on-site acquisition sensors or specific known components, the precise spatial pose of the high-altitude acquisition platform at the first key visual marker point is calculated using photogrammetry or computer vision algorithms. The target component whose geometric attributes are to be determined has a current geometric parameter reference value in the database, and the central processing unit is further configured to: compare and analyze the three-dimensional geometric model and positioning data of the determined target component with the current geometric parameter reference value; if the determined data has a significant deviation from the current geometric parameter reference value, then automatically replace and update the current geometric parameter reference value in the database with the determined data, and record the metadata of this correction.

[0011] Furthermore, the central processing unit is also configured to: For each of the multiple continuous spatial poses recorded when the high-altitude acquisition platform moves along a scanning path, the specific pose of the high-altitude acquisition platform is determined by analyzing the data acquired in that pose and the first group of components with known geometric properties in the database. By combining the data collected by the high-altitude acquisition platform from a local structure of a dam under multiple continuous spatial poses, and combining the known spatial poses of the high-altitude acquisition platform, multi-view stereo reconstruction and data fusion are performed to determine a complete and consistent three-dimensional model of the local structure of the dam.

[0012] Furthermore, the central processing unit is also configured to: calculate the spatial attitude of the high-altitude acquisition platform at the observation point by analyzing the data collected by the high-altitude acquisition platform at at least one preset observation point and the first group of components with known geometric properties in the database; A specific physical feature in the dam area scene, or an independent dam structural unit that needs to be integrated with the main dam model, is precisely associated and bound with the calculated spatial attitude and the image content acquired under that attitude, thereby establishing an accurate placement position for the feature or structural unit in the digital 3D scene.

[0013] Furthermore, the first group of components with known geometric properties are defined in the database with spatial coordinates in a first coordinate system. The target component with the geometric properties to be determined originates from another independent measurement project, and its initial reference coordinates are defined in a second coordinate system. The central processing unit is further configured to, during correlation analysis, use the high-altitude acquisition platform as a common measurement bridge to solve and transform the positioning of the target component to the first coordinate system, thereby realizing the fusion of different coordinate systems and data.

[0014] Furthermore, the high-altitude data acquisition platform integrates a composite sensor unit configured to operate in a first data acquisition mode and a second data acquisition mode. The first data acquisition mode is a high-resolution optical camera mode, used to acquire texture and color information; the second data acquisition mode is a lidar scanning mode or a depth sensing mode, used to directly acquire three-dimensional point cloud data. Furthermore, the first group of components in the database that are associated with the high-altitude acquisition platform at the first level have high-resolution texture features pre-stored, which are suitable for optical feature matching; while the known components in the second group of components that are associated with the high-altitude acquisition platform at the second level have accurate 3D point cloud or mesh models pre-stored, which are suitable for 3D point cloud registration or depth data fusion.

[0015] Beneficial effects: (1) This invention uses a pre-constructed database of hydraulic engineering dam components to perform multi-dimensional similarity matching and association analysis between extracted texture features, geometric contours, and standardized dam component models in the database. This enables intelligent identification of potential model component sets and their topological relationships that match the current dam. The central processing unit instantiates parameterized 3D model components and automatically assembles and positions them based on the extracted contour spatial constraints. This overcomes the shortcomings of traditional manual modeling, such as low efficiency, reliance on experience, and unstructured models. It achieves rapid, accurate mapping and automated assembly of actual collected data into parameterized, editable component-level models with engineering semantics.

[0016] (2) The present invention adopts a hierarchical multi-angle data association analysis spatial analysis process: The first level of association uses an aerial acquisition platform to observe the reference component with known coordinates, and calculates the platform's precise six-degree-of-freedom spatial pose in real time through feature matching and spatial resection (multi-view geometric calculation); The second level of association uses the platform as a known sensor, combines the observation of the target component and other known components, and uses spatial forward intersection or multi-view triangulation algorithm to calculate the three-dimensional geometry and positioning of the target component; This method enables the system to accurately determine the size and position of any target component by transmitting and encrypting coordinates through a mobile platform in areas lacking global control points, relying only on a small number of known first-group components, which significantly reduces the dependence on ground control measurement and is suitable for remote and complex water conservancy engineering scenarios.

[0017] (3) After calculating the actual three-dimensional geometric dimensions and spatial positioning of the target component, this invention compares them with the initial geometric parameter estimates stored in the database. If the difference exceeds the preset tolerance threshold, the database is automatically updated. This mechanism enables the database to continuously optimize itself with new measured data, ensuring that the model always reflects the actual form of the dam (such as changes after deformation, aging, and repair). It solves the problem of poor timeliness and difficulty in responding to dynamic changes in the dam body in existing technologies. The database-driven closed-loop dynamic correction mechanism improves the timeliness of the model and the self-learning ability of the database. (4) The high-altitude acquisition platform of the present invention integrates high-resolution optical camera mode and lidar / depth sensing mode, respectively adapted to optical feature matching and three-dimensional point cloud registration. The central processing unit guides the platform to observe the target component from multiple angles at key visual marker points, effectively overcoming the single-view occlusion problem and obtaining more complete three-dimensional surface information. Multi-mode sensor fusion and multi-angle observation guidance improve the model integrity in complex occlusion scenarios. By fusing and reconstructing optical texture and lidar point cloud data, a three-dimensional model with both high-realism texture and accurate geometric structure is generated, improving the reconstruction quality of the model in complex local structures such as flood discharge channels, corridors, and gate piers; (5) This invention uses an aerial data acquisition platform as a common measurement bridge, which can solve and transform the positioning of target components (such as historical design models, supplier sub-components, and later added structures) originating from different independent measurement projects and defined in different coordinate systems into a unified global engineering coordinate system. This capability solves the technical problem of the difficulty in automatically and accurately splicing and fusing multi-period and multi-source data in traditional methods, laying the foundation for building a complete, consistent, and traceable digital model of the entire life cycle of the dam, and realizing the unification and accurate fusion of multi-source and multi-coordinate system data; (6) The final three-dimensional model generated by this invention is not a continuous triangular facet surface model, but a structured model instantiated and assembled from standardized components with engineering semantics (such as gate piers, overflow surfaces, corridors, and wave walls). Each component is associated with geometric parameters, material properties, and topological connection information, which can support in-depth engineering applications such as automatic calculation of engineering quantities, structural performance simulation analysis, independent editing and replacement of specific components, and standard-based compliance checks. This significantly improves the practical value of the model in the entire process of water conservancy engineering design, construction, and operation and maintenance. The generated model has structure, parameterization, and engineering semantics, supporting in-depth engineering applications. Attached Figure Description

[0018] Figure 1 This is an overall flowchart of the method for generating a three-dimensional model of a dam simulation in water conservancy projects according to the present invention; Figure 2 This is a flowchart of the spatial analysis method for multi-angle data correlation analysis in this invention. Detailed Implementation

[0019] 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.

[0020] This invention provides a method for generating a 3D model of a hydraulic engineering dam, comprising the following steps: Using at least one high-altitude acquisition platform deployed on a mobile vehicle, dynamic scanning and imaging are performed on the target hydraulic dam structure along a predetermined or adaptively planned set of spatial angles and relative height datasets to acquire a multi-source visual data stream containing different facades and details of the dam structure; wherein the high-altitude acquisition platform is configured to receive and execute motion trajectory instructions from a central processing unit; the central processing unit receives and processes the multi-source visual data stream, extracting texture features, geometric contours, structural joints, and material reflection properties of the dam surface through extraction of continuous video frames and analysis of frame images; based on a pre-built database of hydraulic engineering dam components, the extracted data is processed... The texture features and geometric contours of the model are compared with standardized dam component models in the database using multi-dimensional similarity matching and association analysis to identify potential model component sets and their topological relationships corresponding to the current dam structure. Based on the results of the association analysis, the corresponding standardized dam component models are retrieved from the database and instantiated. Then, based on the spatial constraints determined by the geometric contours, initial assembly and positioning are performed in the 3D modeling environment to form the initial 3D geometric framework of the dam structure. The initial 3D geometric framework is then optimized, including: adjusting the model surface material parameters based on extracted material reflection characteristics; smoothing the geometric contours and repairing gaps based on multi-angle visual data; and mapping high-resolution textures onto the corresponding model surface.

[0021] In the above embodiments, one or more mobile high-altitude acquisition platforms are deployed. These platforms are specifically implemented as unmanned aerial vehicle (UAV) systems or liftable robotic arm platforms equipped with high-resolution optical cameras, lidar sensors, and inertial measurement units (IMUs). The high-altitude acquisition platforms communicate with the central processing unit (CPU) via 4G / 5G or a data radio. The data frame format includes image blocks, timestamps, IMU data, and checksums. A custom binary protocol is used to ensure real-time performance and synchronization accuracy. The platform is configured to receive digital instructions from the CPU, which include detailed flight or movement path planning based on preliminary contour information of the target dam. This path planning is designed to move the platform along multiple different spatial axes, specifically including a sequence of horizontal circumferential angles around the dam and a sequence of height levels maintaining different vertical distances from the dam facade, thereby performing dynamic scanning imaging. The CPU receives and processes multi-source visual data streams, including high-resolution optical images, video frame sequences, and lidar point cloud data. By extracting continuous video frames and analyzing the frame images, the texture features, geometric contour lines, structural joint lines, and material reflection properties of the dam surface are extracted. The structural joint lines are used to help determine the topological connection relationships between components and serve as spatial constraints in subsequent component matching and assembly. The pre-construction method for the dam component database of the water conservancy project is as follows: First, collect the Building Information Modeling (BIM) model, as-built drawings, and historical laser scanning data of the target water conservancy project. Then, according to engineering semantics, the dam is decomposed into standardized components such as gate piers, spillway surfaces, guide walls, wave walls, galleries, spillway gates, and maintenance gate slots. For each type of component, extract its parametric geometric dimensions (length, width, height, radius of curvature, chamfer, etc.), typical texture features (such as concrete surface texture, joint patterns, water stain distribution), material properties (reflectivity, roughness, color), and topological connection point information (such as the coordinates of the mating surfaces of adjacent components and the location of bolt holes). Finally, store the above data in a structured form in the database, and assign a unique identifier and spatial coordinate reference system to each component. The database supports incremental updates and version management, and can be queried in real time by the central processing unit through a network interface. Multi-dimensional similarity matching algorithm: The central processing unit matches the extracted image features with component features in the database. Specifically, for texture features, scale-invariant feature transform (SIFT) or oriented fast and rotated binary feature point algorithm (ORB) descriptors are used to calculate Euclidean distance or Hamming distance as similarity metrics. For geometric contours, shape context descriptors are used for matching.

[0022] The adaptive planning spatial angle set and relative height dataset are dynamically generated by the central processing unit (CPU) based on real-time reconstruction quality. The specific method is as follows: The CPU continuously evaluates the quality indicators of the generated partial 3D models, including local point cloud density, texture clarity, number of feature points, and model coverage. When the point cloud density of a certain area is lower than a preset threshold (e.g., 100 points / square meter) or the texture clarity is insufficient (e.g., the average image gradient magnitude is less than 10), the CPU marks this area as an "undersampled area." Then, based on the information gain criterion, a path planning algorithm (such as Rapidly-exploring Random Tree, RRT) is used to generate new acquisition points: the spatial angle of the new acquisition points should be such that the angle between the camera optical axis and the normal of the undersampled area is between 30° and 60°, and the relative height should be such that the ground sampling distance (GSD) is better than 1 cm / pixel. The generated acquisition commands are sent to the high-altitude acquisition platform for execution via a wireless link. The shooting process generates multi-source visual data streams, which are received and processed by the central processing unit after transmission. The processing operations specifically include: applying a keyframe extraction algorithm based on scene content change detection to the continuous video stream (e.g., using optical flow to detect the rate of change between frames, and extracting keyframes when the rate of change exceeds a threshold); applying computer vision algorithms to the extracted keyframe images, specifically including edge detection algorithms to extract the geometric contour lines of the dam; applying Local Binary Patterns (LBP) or Gray-Level Co-occurrence Matrix (GLCM) texture analysis algorithms to identify the material type and weathering characteristics of the dam surface; and applying scale-invariant feature transformation or binary feature point algorithm oriented towards fast rotation to perform feature point detection and description algorithms to identify structural joints and feature corners. The central processing unit (CPU) accesses a pre-built database of hydraulic engineering dam components. This database stores various standardized dam component models in a structured format. Each model is associated with geometric parameters, typical texture features, material properties, and possible topological connection point information. The CPU matches the extracted texture features and geometric contours with this database. This matching process is achieved by calculating a multi-dimensional similarity score between the extracted features and the database model features. Based on the score threshold, a potential set of model components matching the current dam is identified, and the approximate spatial adjacency and connection relationships between these components are inferred. Then, based on the matching results, the CPU instantiates the corresponding parametric 3D model components from the database and imports them into a 3D modeling environment. In this environment, based on the spatial boundary constraints defined by the optimized geometric contours extracted from the image, these model components are initially assembled and precisely positioned using automatic alignment and snapping algorithms, thereby forming an initial 3D geometric framework representing the main structure of the dam. Subsequently, a series of optimization processes were performed on the initial framework: based on the extracted material reflection characteristics, the diffuse, specular, and normal map parameters of the 3D model surface shader were adjusted; using visual data acquired from multiple angles, dense point clouds generated by multi-view stereo vision algorithms were used to iteratively optimize the geometric surface of the initial assembled model and perform gap interpolation repair; and the high-resolution panoramic texture map generated through image stitching and correction was accurately wrapped onto the corresponding model mesh surface using UV mapping technology. Finally, the optimized 3D model data was imported into a 3D rendering engine, configured with a real-world lighting model and atmospheric scattering parameters, and ray tracing or rasterization rendering calculations were performed to generate a highly realistic 3D model of a hydraulic engineering dam simulation that can be used for multi-angle observation, structural stress analysis simulation, or virtual inspection drills.The method for extracting material reflection characteristics is as follows: To obtain the material reflection characteristics, the central processing unit controls a high-altitude acquisition platform to acquire multiple images of the same area under different lighting directions (e.g., by flying at different times or using the platform's own active light source). Then, a photometric stereo method is used to reconstruct the surface normal map and estimate the surface's albedo and roughness parameters. For areas with strong specular reflection (such as metal parts), a bidirectional reflection distribution function (BRDF) model is further used to fit and extract its specular component parameters. These parameters are subsequently used for material rendering of the 3D model. Geometric contour lines determine spatial constraints: Geometric contour lines extracted from the image (such as dam edge lines and pier edges) are fitted using the Random Sample Consensus (RANSAC) algorithm to obtain spatial straight lines or planes. These straight lines or planes serve as spatial constraints for component assembly. For example, the side contact surfaces of pier components are aligned with the extracted dam axis plane; the overflow surface contour lines are fitted with the extracted parabolic contour lines. The central processing unit utilizes these constraints in the 3D modeling environment, automatically snapping the instantiated component models to the correct positions using the Iterative Closest Point (ICP) algorithm. In some embodiments, a method for multi-dimensional similarity matching and association analysis based on a pre-constructed database of hydraulic engineering dam components includes: establishing a first-level association by performing feature point matching and spatial analysis calculations between data collected by the high-altitude acquisition platform at a first key visual marker point and a first group of components with known geometric attributes in the database, and determining at least one first spatial pose of the high-altitude acquisition platform relative to the dam structure by the central processing unit; establishing a second-level association by performing association analysis between data collected by the high-altitude acquisition platform at a second key visual marker point and a target component with unknown geometric attributes, and combining the pose of the high-altitude acquisition platform determined in the first-level association, the known spatial position of the first group of components, and other components with known geometric attributes in the database, and calculating the precise three-dimensional geometric dimensions and spatial positioning of the target component by the central processing unit; wherein the first group of components used for the first-level association and the second group of components used for the second-level association have an inclusion relationship, the same set, or a non-intersecting relationship in the component classification hierarchy of the database. The first key visual marker and the second key visual marker can be the same geographical location or different geographical locations, depending on the on-site observation conditions and data acquisition strategy. In the above embodiments, the multi-dimensional similarity matching and association analysis based on the pre-constructed database of hydraulic engineering dam components specifically includes two levels of spatial parsing association operations: The first level of association operation involves an aerial acquisition platform, a six-rotor UAV equipped with a 20-megapixel optical camera and a lidar, and a database containing reference components with known geometric properties: three permanent prisms (coordinates known) at the corner of the dam crest wave wall. The aerial acquisition platform flies along a preset path to the first key visual marker point (5 meters above the wave wall corner), hovers, and captures images. The central processing unit analyzes one or more frames of images captured at this point, and uses a feature point matching algorithm to associate the pixel coordinates of the reference component in the image with the 3D coordinates of the corresponding feature points on the 3D model of the component in the database, identifying the image points of the three prisms. Using the spatial resection principle in photogrammetry, combined with known camera intrinsic parameters, the first spatial position and attitude angle of the aerial acquisition platform in the global coordinate system when capturing this image are calculated, i.e., a six-degree-of-freedom pose estimation. The first group of components uses standard dam components. The second-level association operation involves the high-altitude acquisition platform continuing to fly to the second key visual marker point (on the other side of the same wave-breaking wall section) and capturing images of a target component with undetermined geometric attributes—a gate hoist base of a specific shape. The central processing unit analyzes the images captured by the high-altitude acquisition platform at the same or nearby positions where the pose has been calculated, and identifies the visual features of the target component. Then, it combines the high-altitude platform pose determined in the first-level association, the three-dimensional coordinates of one or more other fixed components with known positions and geometric dimensions (e.g., nearby gate piers) in the database, and the pixel relative relationships between the target component feature points measured from the current image and these known component feature points. Through spatial forward intersection or multi-view triangulation algorithms, the coordinates of the key feature points on the target component in three-dimensional space are calculated, thereby determining the overall geometric dimensions, shape, and precise spatial positioning of the target component in the global coordinate system. This is compared with the initial geometric parameter estimates stored in the database. If the values ​​exceed the tolerance threshold, the base model parameters in the database are automatically updated. In the database's component classification hierarchy, the relationship between the first set of components used for the first-level association (e.g., a set of known control points distributed along the dam axis) and the second set of components used for the second-level association can be as follows: the latter is completely contained within the former (subset relationship), for example, the second-level association only uses a portion of the control points from the first-level association; or both point to the exact same set of physical components in the database; or the components involved in both are completely separated in spatial location and do not overlap, for example, the first-level association uses known points on the left bank of the dam, and the second-level association uses known points on the right bank of the dam to solve for the intermediate target component.

[0023] In some embodiments, the method further includes the step of modeling dam ancillary structures using a continuous pose sequence: by performing continuous frame feature tracking and spatial intersection calculation between the data collected at multiple continuous pose points and the first group of components with known geometric properties in the database when the high-altitude acquisition platform moves along a predetermined trajectory traversing the target area, the central processing unit determines multiple continuous spatial poses of the high-altitude acquisition platform within the target area; By performing stereoscopic visual analysis on the data collected from the high-altitude acquisition platform under multiple continuous spatial poses and on the dam ancillary structures within a target area to be determined, the central processing unit determines the three-dimensional model of the dam ancillary structures and their spatial position relative to the main dam body.

[0024] In the above embodiments, the steps of performing multi-dimensional similarity matching and association analysis based on a pre-built database of hydraulic engineering dam components specifically include: the central processing unit plans a predetermined scanning trajectory covering a certain dam area (e.g., a spillway) and controls the high-altitude acquisition platform to move along this trajectory; during the movement, the platform continuously acquires images or point cloud data at a fixed frequency. The central processing unit processes the continuously acquired data sequence: in each frame of data, it identifies and tracks the same visual feature points on the first group of components with known geometric attributes in the database (e.g., prefabricated feature patterns on the side of the spillway gate pier); using visual odometry or simultaneous positioning and mapping technology, combined with the three-dimensional coordinates of these known feature points as constraints, and through a cluster adjustment algorithm, it continuously and accurately calculates each continuous spatial pose of the high-altitude acquisition platform during the movement, forming a high-precision platform motion trajectory. Then, for a specific dam ancillary structure to be determined within the scanning area, such as a traffic bridge on the spillway, the central processing unit selects images or point cloud slices containing the structure collected by the platform in multiple different continuous spatial poses near the structure; using the calculated precise platform pose as the extrinsic parameter of the camera or sensor, stereo vision matching or point cloud registration is performed on these multi-view data; through a multi-view stereo reconstruction algorithm, a dense 3D point cloud model of the traffic bridge is generated, and further converted into a mesh model through a surface reconstruction algorithm; at the same time, the position and orientation of the model in the global coordinate system are also automatically determined by the reconstruction process, thereby establishing a precise spatial positional relationship between the ancillary structure and the main dam model, which can be used for subsequent integrated integration.

[0025] In some embodiments, the method for identifying key visual markers and determining platform pose includes: the central processing unit determining the theoretical position of at least one key visual marker in the real-world coordinate system based on pre-input dam design drawings or topographic map data; the central processing unit or the high-altitude acquisition platform itself determining, through a visual positioning algorithm, when the high-altitude acquisition platform arrives at or aligns with the effective data acquisition range of the key visual marker; controlling the high-altitude acquisition platform to interact with a first set of field acquisition sensors arranged at known locations on or around the dam body, or directly analyzing images containing known first set of components through a visual algorithm to obtain a series of distance or viewing angle measurements; and performing spatial resection or multi-view geometric calculation based on the set of measurements and the known coordinates of the first set of field acquisition sensors or known first set of components to determine at least one first spatial pose of the high-altitude acquisition platform at the key visual marker. In the above embodiments, the preliminary preparation and triggering control process for ensuring the accuracy of the first-level association specifically involves the central processing unit defining and marking the theoretical three-dimensional coordinates of at least one of the first key visual markers in the real-world global engineering coordinate system on a digital map based on the input original dam design CAD drawings, as-built survey drawings, or high-precision preliminary topographic maps. This coordinate, along with its corresponding visual feature description, is then stored in a database. During data acquisition, the visual processor mounted on the high-altitude acquisition platform, or the central processing unit, receives the video stream transmitted back from the platform in real time and runs a visual positioning algorithm. This algorithm continuously matches the real-time images with the visual feature descriptions (such as SIFT or ORB feature descriptors) of the first key visual markers stored in the database and calculates the matching confidence level. When the confidence level exceeds a preset threshold and the pixel coordinates of the matched feature points are stable within a certain range in the center area of ​​the image, it is automatically determined that the high-altitude acquisition platform has reached or precisely aligned with the effective data acquisition range of the key visual marker, thereby triggering the subsequent precise measurement process.

[0026] To determine whether the high-altitude acquisition platform has reached the effective data acquisition range of key visual markers, this invention employs the following visual positioning algorithm: The high-altitude acquisition platform acquires images in real time and extracts ORB feature points, which are then brute-force matched with pre-stored key visual marker template images in the database (Hamming distance threshold set to 50). When the number of matched feature points is ≥20 and the average reprojection error is ≤2 pixels, the relative pose between the platform and the markers is calculated using the Perspective-n-Point (PnP) algorithm (using the efficient Perspective-n-Point algorithm, EPnP). If the calculated relative distance is less than a preset threshold (e.g., 1 meter) and the viewing angle deviation (i.e., the angle between the camera optical axis and the marker normal) is less than 15°, then the platform is considered to have entered the effective data acquisition window. This algorithm can run on the platform's onboard processor or central processing unit and output a trigger signal to initiate collaborative measurement. The step of determining at least one first spatial pose is further refined into an active collaborative measurement process: the central processing unit sends instructions to the high-altitude acquisition platform and a first set of field acquisition sensors (such as QR code targets or ultra-wideband base stations with wireless communication modules) that are pre-distributed and fixedly installed on and around the dam body with their positions precisely determined; the instructions trigger collaborative work between the two - the platform emits specific light signals or wireless ranging signals to the sensors, and the sensors return a response after receiving them; or, the platform directly images a specific pattern on the sensor; by calculating the signal flight time or analyzing the pixel size and position of the sensor in the image, a series of precise distance values ​​or orientation angle measurements from the platform to each sensor are obtained. Finally, the central processing unit collects these distance or angle measurements to form a set of measurement values, and combines them with the precise three-dimensional coordinates of the first set of field acquisition sensors or known first set of components stored in the database, using multi-view triangulation algorithms or spatial resection, to solve for the six-degree-of-freedom spatial attitude and position parameters of the high-altitude acquisition platform at the key visual marker point at the moment of shooting, i.e., at least one first spatial pose.

[0027] In some embodiments, the method further includes dynamic correction of database component parameters: when an initial geometric parameter estimate of the target component is already stored in the database, the central processing unit compares the calculated three-dimensional geometric dimensions and spatial positioning with the initial geometric parameter estimate stored in the database; if the difference between the three-dimensional geometric dimensions and spatial positioning and the initial geometric parameter estimate exceeds a preset tolerance threshold, the initial geometric parameter estimate in the database is updated with the three-dimensional geometric dimensions and spatial positioning to achieve dynamic correction of database component parameters. The initial geometric parameter estimate originates from design values, historical modeling values, or default parameters in the database; the preset tolerance threshold is set according to engineering accuracy requirements, for example, a dimensional deviation of no more than 5% and a positional deviation of no more than 0.5 meters. In the above embodiments, the dynamic verification and update mechanism for component parameters in the database is as follows: When the target component is identified and an association attempt is made, the central processing unit first queries the database of hydraulic engineering dam components. It finds that the component already has an initial geometric parameter estimate imported from design drawings or left over from historical modeling. This estimate may include approximate dimensions such as length, width, and height, and a rough spatial positioning matrix. After successfully calculating the precise three-dimensional geometric dimensions and spatial positioning of the target component through the second-level association operation, the central processing unit initiates a comparison procedure: it calculates the difference between the calculated precise dimensions (e.g., length values ​​accurate to the centimeter level) and the initial estimated dimensions in the database item by item; simultaneously, it calculates the transformation error between the calculated positioning matrix (including translation and rotation) and the initial positioning matrix. Next, the central processing unit compares these differences with a preset tolerance threshold, which is set according to engineering accuracy requirements, for example, a size deviation of no more than 5% and a position deviation of no more than 0.5 meters. If any difference exceeds its corresponding tolerance threshold, it is determined that there is a significant difference. At this point, the central processing unit automatically performs an update operation: it completely replaces the initial geometric parameter estimates stored in the database with the precise three-dimensional geometric dimensions and spatial positioning data calculated from the actual data. Furthermore, it records the timestamp of this update, the identifier of the data source used, and the numerical comparison before and after the update, thus forming a closed-loop learning and correction mechanism that allows the database to continuously optimize itself as more actual survey data is input.

[0028] In some embodiments, by performing feature matching and spatial calculation between data collected by an aerial acquisition platform deployed on a mobile vehicle at at least one key visual marker and a first group of components with known geometric properties in a pre-built database, a central processing unit determines at least one observation pose of the aerial acquisition platform relative to the dam. The central processing unit then spatially associates a physical entity object within the dam area and / or a dam structural part to be finely modeled with the visual data collected under the at least one observation pose, thereby binding the physical entity or structural part to a specific position in the scene coordinate system jointly determined by the observation pose and image content.

[0029] The central processing unit controls an aerial acquisition platform deployed on a mobile vehicle to fly or move to at least one pre-selected or real-time identified key visual marker within the dam area, such as a distinctive rock protrusion on the dam abutment or a permanent measurement marker. At this point, the platform acquires high-resolution images or laser point cloud data. Subsequently, the central processing unit performs the first stage of correlation calculation: it retrieves a first set of components with known geometric properties from a pre-built database (these components are within the field of view of the marker and their three-dimensional coordinates are known), and uses a feature matching algorithm to precisely match the pixel coordinates of these standard components in the image with their known three-dimensional model coordinates; using these two-dimensional to three-dimensional point pairs, combined with the camera's intrinsic parameters, and employing a direct linear transformation or perspective N-point algorithm, it calculates the observation pose of the aerial acquisition platform when capturing this frame of data. This pose includes the platform's three-dimensional position coordinates in the global coordinate system and rotation angles along three axes. Then, the central processing unit (CPU) performs the second stage of scene entity association: it analyzes data collected under the same observation pose to identify a specific physical entity within the dam area (such as a characteristic tree located below the dam, or a power distribution room) or a section of the dam structure that needs to be modeled in detail (such as a section of wave wall with a different material from the main dam). The CPU separates this entity or structural section from the background using image segmentation or point cloud clustering algorithms. Finally, using the calculated precise observation pose of the high-altitude platform as sensor extrinsic parameters, combined with the pixel boundaries of the entity in the image or the 3D point set in the point cloud, the CPU calculates the 3D bounding box or center point coordinates of the entity or structural section in the global scene coordinate system through coordinate transformation. Through this process, the physical entity or structural section to be refined is successfully bound to a definite 3D scene position derived from the precise observation pose and the original observation data. This provides an indispensable spatial positioning benchmark for subsequently inserting an independent 3D tree model, building model, or initiating a high-precision reconstruction process for the wave wall at that location.

[0030] In some embodiments, the spatial analysis method of multi-angle data correlation analysis in 3D model generation includes: First-level analysis: using the high-altitude acquisition platform as a mobile known observation station, through multiple observations of the platform with known location points or known model components on the dam body, the platform's precise trajectory and attitude sequence are solved; Second-level analysis: based on the known platform attitude, the platform is regarded as a spatially known sensor, and its observation data of the target component is used, combined with the observation constraints of other known location points on the same unknown structure, and the precise geometry and position of the unknown structure in 3D space are calculated through spatial forward intersection or multi-view triangulation algorithms; Third-level analysis: when dealing with multiple structural modules whose spatial relationships are unknown, the high-altitude acquisition platform is cross-observed at multiple locations, and the obtained observation network data is used for overall adjustment, while simultaneously solving the relative positions and directions of all unknown modules to construct a self-consistent global 3D model.

[0031] As a specific implementation of the aforementioned multi-dimensional similarity matching and association analysis steps, the central processing unit employs a hierarchical spatial analysis method, which includes the following three levels: The first level of analysis (resection layer) treats the high-altitude acquisition platform as a mobile, known observation station with known intrinsic parameters but unknown extrinsic parameters. The high-altitude acquisition platform synchronously or sequentially observes and images at least three (at least four in three-dimensional space) known location points on the dam body (i.e., reference component feature points with known coordinates in the database) along its flight path. Each observation yields a set of two-dimensional image coordinates. The central processing unit uses collinearity condition equations and these two-dimensional-three-dimensional point pairs to solve for the precise extrinsic parameters (position and attitude) of the platform at each observation and imaging moment through spatial resection. This essentially transfers the known ground control point coordinates to the mobile observation platform. The second level of analysis (forward resection layer) repositions the high-altitude acquisition platform, whose extrinsic parameters have become known after the first level of analysis, as a mobile sensor with a determined spatial position and attitude. When a known sensor observes any unknown spatial point (e.g., a feature point of a target component) and obtains a two-dimensional image coordinate, the equation of this observation ray in space is determined. A single ray cannot determine a three-dimensional point. However, if the same unknown point is observed again by the platform from another known pose, or by another fixed sensor with a known spatial location (or a virtual observation point on a known component), then two or more spatial rays are formed. By using spatial forward intersection or multi-view triangulation algorithms, the intersection point or the best approximation point in the least squares sense of these rays in space can be obtained, and the three-dimensional coordinates of the unknown point can be calculated. By solving multiple feature points on the target component, its overall geometry and location can be determined. The third level of analysis (overall adjustment layer) is used when dealing with a complex area composed of multiple structural modules whose initial spatial relationships are unknown (e.g., the sidewalls and bottom slabs on both sides of a spillway). The high-altitude acquisition platform cross-observes all modules from multiple different locations, forming a complex network of observation values ​​(containing a large number of two-dimensional image point coordinates). In this network, all platform pose parameters and the 3D coordinates of feature points of all unknown modules are parameters to be determined. Using the bundle adjustment algorithm in photogrammetry, global least-squares adjustment is performed on all observation equations. This allows for the simultaneous and optimal calculation of the poses of all unknown platforms and the relative positions and orientations of all unknown structural modules, constructing a highly self-consistent global 3D model. Functionally, this process is equivalent to using a mobile observation station to densify a sparse control network, or constructing a free network and performing adjustment, thus achieving multi-scale modeling from local precision to global consistency.

[0032] This invention also provides a system for generating a 3D model of a hydraulic engineering dam, comprising a central processing unit, at least one high-altitude acquisition platform, and a pre-built database of hydraulic engineering dam components. The high-altitude acquisition platform is configured to be mounted on a mobile vehicle for acquiring image or video data of the target dam from multiple angles and heights. The central processing unit is communicatively connected to the high-altitude acquisition platform and the database, and is configured to: determine the spatial pose of the high-altitude acquisition platform at the first key visual marker by analyzing the data acquired by the high-altitude acquisition platform at a first key visual marker and performing feature point matching and spatial analysis calculations with a first group of components in the database with known geometric attributes; and determine the 3D geometric model and location of the target component by analyzing the data acquired by the high-altitude acquisition platform at a second key visual marker and performing correlation analysis with a second group of components, the second group including at least one target component with undetermined geometric attributes and other components in the database with known geometric attributes. The first group of components and the second group of components have an inclusion relationship, are the same set, or are disjoint at the component classification level in the database, and the first key visual marker and the second key visual marker are the same or different spatial locations. It should be noted that the first key visual marker is used to determine the spatial pose of the high-altitude acquisition platform, while the second key visual marker is used for fine-grained calculation of the target component. In actual operation, these two markers can be at the same location (e.g., the high-altitude acquisition platform simultaneously performs pose self-calculation and target component observation at a hovering point), or they can be at different locations (e.g., the high-altitude acquisition platform first completes pose calibration at an open point, and then flies to another point closer to the target component for observation). The central processing unit automatically selects the most suitable strategy based on the real-time data quality. In the above embodiments, the system of the present invention includes: a ground-based central processing unit (a high-performance server equipped with a GPU accelerator card), one or more drones (as a high-altitude acquisition platform), and a pre-built database of hydraulic engineering dam components (deployed locally on the server or in cloud storage). The drone carries a composite sensor unit: a high-resolution optical camera (first acquisition mode) and a 3D lidar (second acquisition mode). The database stores optimized data for different association levels: high-resolution texture images of reference components used for the first level of association are pre-stored (for easy optical feature matching); known components used for the second level of association are pre-stored with accurate 3D point clouds or mesh models (for easy point cloud registration). The central processing unit communicates with the drone via 4G / 5G or a data radio, receives data in real time, and issues control commands; when the high-altitude acquisition platform flies to a selected first key visual marker point (such as the outlet structure of the dam's diversion bottom hole) and hovers, the central processing unit receives the image of that point transmitted back by the platform. It automatically retrieves a first group of components with known geometric properties from the database. These components are visible in the image and their 3D coordinates have been accurately determined, such as the concrete corner feature points of the diversion tunnel outlet. The central processing unit (CPU) uses feature detection and matching functions from computer vision libraries (such as OpenCV) to establish a correspondence between images and 3D models. It then calls photogrammetry algorithm libraries to perform multi-view geometric calculations, ultimately outputting the precise spatial pose (X, Y, Z, pitch, yaw, roll) of the high-altitude acquisition platform at its current hovering point. Next, based on this known pose of the high-altitude acquisition platform, the CPU analyzes the same or immediately following images to identify a second set of components. This set includes at least one target component with undetermined geometric properties (e.g., a damaged concrete panel above the exit structure) and other components from the database that are also visible from this viewpoint and have known geometric properties (e.g., an intact tunnel roof edge). Using the known platform pose as camera parameters, the CPU projects the feature points of the target component and known components in the images into 3D space. Through algorithms such as spatial intersection and scale recovery, it determines the 3D geometric model parameters of the target component (concrete panel), including its size, thickness, and indentation depth, and calculates its precise spatial positioning relative to known components, thus completing the modeling of this newly added or mutated component.

[0033] In some embodiments, the central processing unit is further configured to: for each of the multiple continuous spatial poses recorded by the high-altitude acquisition platform as it moves along a scanning path, determine the specific pose of the high-altitude acquisition platform by analyzing the data acquired in that pose and a first group of components with known geometric attributes in the database; and determine a complete and consistent three-dimensional model of the dam's local structure by integrating the data acquired by the high-altitude acquisition platform in multiple continuous spatial poses for a local structure of a dam to be modeled, and combining the known spatial poses of the high-altitude acquisition platform.

[0034] In the above embodiment, the central processing unit plans a scanning path covering the target dam area and controls the high-altitude acquisition platform to move at a constant speed along the path. During the movement, the platform continuously captures images at a fixed frequency and records its auxiliary sensor data. For each continuous spatial pose data packet recorded during the movement, which contains the raw GNSS / IMU data corresponding to the image and timestamp, the central processing unit performs real-time or post-processing. For each pose, it identifies and extracts feature points from the received corresponding image; simultaneously, it retrieves the first set of components with known geometric properties within the image's field of view from the database. By matching the image feature points with known points on these lamppost 3D models, and combining the initial pose provided by the platform's GNSS / IMU as the initial value for iteration, the optimized six-degree-of-freedom spatial pose of the high-altitude acquisition platform at that specific moment is accurately determined using visual inertial odometry or image matching optimization algorithms. After the platform completes coverage of the entire scanning area and obtains hundreds of such continuous accurate poses and corresponding images, the central processing unit initiates a comprehensive processing flow for a local structure of the dam that requires overall modeling. It selects all images containing the stilling basin and, using the precise camera pose of each image, performs a multi-view stereo matching algorithm. This algorithm searches for corresponding pixels across all image pairs and generates a large number of 3D points on the stilling basin surface using triangulation. Finally, the central processing unit applies Poisson surface reconstruction or rolling sphere algorithm to fuse and reconstruct a watertight, complete, and geometrically consistent 3D mesh model of the stilling basin, which naturally connects spatially with the other known parts of the dam.

[0035] In some embodiments, the central processing unit is configured to: determine the expected spatial location range of the first key visual marker based on the dam design blueprint or preliminary survey data; By analyzing the real-time video stream or image sequence transmitted back by the high-altitude acquisition platform, a visual recognition algorithm is used to automatically detect when the high-altitude acquisition platform enters the effective data acquisition window of the expected spatial location range; triggering the high-altitude acquisition platform to perform collaborative measurement with the first set of field acquisition sensors deployed at known locations on the dam body, or triggering high-precision analysis of images containing specific known components to obtain a set of observation data for spatial calculation; based on the observation data and the known precise coordinates of the first set of field acquisition sensors or the specific known components, photogrammetry or computer vision algorithms are used to calculate the precise spatial pose of the high-altitude acquisition platform at the first key visual marker point; In the above embodiments, before the task begins, the central processing unit loads the electronic version of the dam's design blueprint or the digital elevation model generated by previous laser scanning. Through human-computer interaction or automatic analysis, the expected spatial location range of the first key visual marker is defined in the central processing unit's memory. This range is represented by a three-dimensional spatial cube, whose center coordinates are theoretical coordinates, and whose side length is a tolerance range (e.g., ±2 meters) set according to the positioning accuracy requirements. During data acquisition, the central processing unit decodes the video stream transmitted back by the high-altitude acquisition platform via a wireless link in real time and calls a parallel visual recognition thread. This thread continuously runs a deep learning-based object detection model, which has been trained on the visual features of the first key visual marker in the database (e.g., a maintenance door of a specific shape). When the detection model continuously identifies the target marker with high confidence in consecutive video frames, and the pixel coordinates of the marker in the image are calculated, and its rough spatial position relative to the platform is estimated to fall within the aforementioned expected spatial cube through simple homography transformation, the central processing unit determines that the platform has entered the effective data acquisition window. Once the determination is successful, the central processing unit immediately sends a synchronization trigger command via downlink to the high-altitude acquisition platform and / or the first set of fixed field acquisition sensors (such as wireless beacons with active luminous markers). The command triggers at least one of the following coordinated measurements: the high-altitude acquisition platform emits a modulated laser pulse towards a specific sensor, which receives it and returns along the same path; the platform calculates the round-trip time to obtain the precise distance. Alternatively, the platform adjusts its gimbal to take a close-up image of the precision checkerboard calibration plate attached to the sensor. The central processing unit collects these distance values ​​or high-resolution calibration plate images, combines them with the known precise three-dimensional coordinates of these sensors in the database, and uses photogrammetric algorithms such as bundle adjustment or direct linear transformation to ultimately calculate the highly precise spatial position and attitude angle of the high-altitude acquisition platform at the trigger moment in an optimized manner.

[0036] In some embodiments, the target component whose geometric attributes are to be determined has a current geometric parameter reference value in the database, and the central processing unit is further configured to: compare and analyze the three-dimensional geometric model and positioning data of the determined target component with the current geometric parameter reference value; if the determined data has a significant deviation from the current geometric parameter reference value, then automatically replace and update the current geometric parameter reference value in the database with the determined data, and record the metadata of this correction.

[0037] In the above embodiment, when the central processing unit (CPU) is ready to process a target component whose geometric attributes need to be determined, it first queries the hydraulic engineering dam component database. The target component exists as a record in the database and is associated with a current geometric parameter reference value. This current geometric parameter reference value is derived from standard design drawings and is either a theoretical value or an estimate from the previous modeling. After the CPU successfully determines the actual three-dimensional geometric model and positioning data of the target component through image analysis and spatial calculation, a built-in comparison analysis module is activated. This module performs a structured comparison between the newly calculated actual data and the current geometric parameter reference value stored in the database. The comparison includes not only the absolute difference in basic dimensions but may also include shape similarity, position offset, and attitude rotation differences. All differences are summarized and compared with a preset, configurable engineering tolerance threshold. If the comparison results show a statistically significant deviation between the newly calculated data and the old reference value (e.g., the main dimension deviation does not exceed 5% and the position offset does not exceed 0.5 meters), the CPU automatically initiates an update transaction. The update transaction first locks the relevant records in the database to prevent concurrent conflicts, and then completely overwrites the old current geometric parameter reference values ​​with the newly calculated 3D geometric model and positioning data that represent the actual survey results.

[0038] In some embodiments, the central processing unit is further configured to: calculate the spatial attitude of the high-altitude acquisition platform at the observation point by analyzing the data collected by the high-altitude acquisition platform at at least one preset observation point and the first group of components with known geometric attributes in the database; and accurately associate and bind a specific physical feature in the dam area scene, or an independent dam structural unit that needs to be integrated with the main dam model, with the calculated spatial attitude and the image content collected under that attitude, thereby establishing an accurate placement position for the feature or structural unit in the digital 3D scene.

[0039] In the above embodiment, the central processing unit controls the high-altitude acquisition platform to fly to at least one preset observation point within the dam area. This observation point is selected at a location where both known standard components and the entity to be anchored can be seen simultaneously. The platform hovers and acquires panoramic images and laser point clouds of this point. The central processing unit then performs calculations: it retrieves the first set of components with known geometric attributes in this field of view from the database, and calculates the platform's current precise spatial attitude using the PnP algorithm through feature matching. Next, the central processing unit performs entity association: it applies a semantic segmentation algorithm to process the panoramic image to identify pixel regions of vegetation or building categories; or, it performs Euclidean clustering analysis on the laser point cloud to separate point cloud clusters independent of the ground. Assuming the identified target is a specific physical feature, the central processing unit extracts the precise contour pixel set of the feature in the image or the three-dimensional point clusters in the point cloud. Using the precise spatial attitude of the platform calculated in the previous step, combined with the camera model or LiDAR scanning model, the central processing unit transforms these two-dimensional pixels or three-dimensional point clusters from the sensor coordinate system to the global coordinate system, calculating the position, size, and orientation of the minimum enclosing cube of the observation station in three-dimensional space. Through this calculation, this independent physical structure is successfully bound to a unified global three-dimensional coordinate system defined by the platform's observation pose and the raw data. Similarly, if the target is an independent structural unit that needs to be finely modeled separately from the main dam model before integration, its precise placement base center point and rotation angle are determined in the same way, providing a unique and accurate positioning basis for the precise assembly of the main model and sub-models.

[0040] In some embodiments, the first group of components with known geometric properties are defined in the database with spatial coordinates in a first coordinate system, and the target component with the geometric properties to be determined originates from another independent measurement project, with its initial reference coordinates defined in a second coordinate system; the central processing unit is further configured to, during correlation analysis, use the high-altitude acquisition platform as a common measurement bridge to solve the positioning of the target component and transform it to the first coordinate system, thereby realizing the fusion of different coordinate systems and data.

[0041] In the above embodiments, in the hydraulic engineering dam component database, the spatial coordinates of the first group of components with known geometric attributes are defined in a first coordinate system. This coordinate system is usually the project's unified global engineering coordinate system (first coordinate system, with the origin located at the intersection of the dam axis and the dam crest). However, the target component with the geometric attributes to be determined may originate from another independent survey project, historical drawings, or sub-component models provided by different suppliers. The initial reference coordinates of the target component to be integrated (such as a gate hydraulic hoist model provided by a supplier) are defined in a local design coordinate system (second coordinate system). When the central processing unit needs to integrate the target component into the main model, the central processing unit uses the high-altitude acquisition platform as a common measurement bridge. The specific process is as follows: the high-altitude acquisition platform is controlled to observe the positions of the known components in the first coordinate system and the target components in the second coordinate system simultaneously. Through calculation, the platform's own pose can be determined in the first coordinate system. At the same time, the platform's observation data of the target component can be described in the second coordinate system or an intermediate sensor coordinate system. The coordinate fusion module of the central processing unit establishes a mathematical relationship between the first coordinate system, platform pose, target component observation, and the second coordinate system and target component model. It constructs a set of coordinate transformation equations. By solving these equations using the principle of coordinate transformation, the rotation matrix and translation vector required to transform the target component from its original second coordinate system to the first coordinate system can be obtained. Through this transformation, the positioning of the target component is successfully calculated and transformed to the first coordinate system consistent with the main model, thus seamlessly achieving the unification and accurate fusion of data from different sources and coordinate systems, laying the foundation for constructing a complete and consistent 3D model of the dam.

[0042] In some embodiments, the high-altitude acquisition platform integrates a composite sensor unit configured to operate in a first data acquisition mode and a second data acquisition mode; wherein, the first data acquisition mode is a high-resolution optical imaging mode for acquiring texture and color information; the second data acquisition mode is a lidar scanning mode or a depth sensing mode for directly acquiring three-dimensional point cloud data; and, the first group of components in the database that are first-level associated with the high-altitude acquisition platform pre-stores high-resolution texture features suitable for optical feature matching; while the known components in the second group of components that are second-level associated with the high-altitude acquisition platform pre-store accurate three-dimensional point cloud or mesh models suitable for three-dimensional point cloud registration or depth data fusion.

[0043] In some embodiments, the central processing unit is further configured to guide the high-altitude acquisition platform to perform multi-angle observations: the key visual markers correspond to the dam's gate pier corners, specific markings on the spillway sidewalls, corners of the dam crest railings, or pre-pasted visual marker targets; the central processing unit guides the movement of the high-altitude acquisition platform, including guiding it to the markers and observing the target component from multiple angles at the markers to obtain data covering more surfaces of the target component, overcoming the single-view occlusion problem; in this way, even for dam components that are partially occluded or located in complex backgrounds, the high-altitude acquisition platform can be guided to move to suitable, unoccluded key visual markers for observation, thereby successfully completing feature extraction and model association, and ultimately generating a complete 3D model of the dam.

[0044] Examples of key visual markers include the specific angular intersections of dam piers, which typically have clear shadow lines and are easily identifiable in images; high-contrast numbering or scale markings painted on the spillway sidewalls; the top corners of the dam crest wave walls or railing columns, which have stable geometric features; or artificial visual marker targets specifically designed and affixed to the dam surface to aid modeling, which have unique codes and known physical dimensions. The central processing unit's motion guidance strategy for the high-altitude acquisition platform includes not only guiding it to the airspace above a specific marker point, but also guiding the platform to perform a series of carefully designed sub-maneuvers around the target component associated with that marker point after arrival. For example, a hemispherical observation point array with a radius of 6 meters is generated centered on a key visual marker point (e.g., an ArUco marking painted on the spillway sidewall), with observation points spaced at 30° azimuth, 30°, 60°, and 90° elevation angles. The high-altitude acquisition platform is guided to sequentially fly to each observation point on the target component, hover, and capture images and laser point clouds. Each tiny area on the surface of each target component is covered by at least three different viewpoints. This multi-angle observation strategy aims to overcome the visual occlusion problem caused by a single viewpoint (such as the part behind the equipment room being obscured by a mountain). By complementing each other with data collected from different viewpoints, more complete and intact 3D surface information of the target component can be obtained. In this way, even for dam components that are partially obscured by other structures (such as trees or other buildings) from a conventional aerial perspective, or located in complex backgrounds (such as in shadows or near reflective water surfaces), the system can guide the high-altitude acquisition platform to a suitable, unobstructed sequence of "key visual markers" for multi-angle observation through intelligent path planning. After acquisition, the central processing unit uses a multi-view stereo matching algorithm (such as PMVS, Patch-based Multi-View Stereo) to generate a dense point cloud, and then obtains a complete, hole-free 3D mesh model through Poisson surface reconstruction.

[0045] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for generating a three-dimensional model of a hydraulic engineering dam, characterized in that, Includes the following steps: By deploying at least one high-altitude acquisition platform on a mobile vehicle, dynamic scanning images are taken of the target hydraulic dam structure along a predetermined or adaptively planned set of spatial angles and relative height datasets to obtain a multi-source visual data stream containing different facades and details of the dam structure; wherein, the high-altitude acquisition platform is configured to receive and execute motion trajectory instructions from a central processing unit. The central processing unit receives and processes the multi-source visual data stream, and extracts the texture features, geometric contours, structural joints, and material reflection properties of the dam surface by extracting continuous video frames and analyzing frame images. Based on a pre-built database of hydraulic engineering dam components, the extracted texture features and geometric contours are matched and analyzed in multiple dimensions with the standardized dam component models in the database to identify the set of potential model components corresponding to the current dam structure and their topological relationships. Based on the results of the correlation analysis, the corresponding standardized dam component model is called from the database and instantiated. Based on the spatial constraints determined by the geometric contour lines, the initial assembly and positioning are carried out in the three-dimensional modeling environment to form the initial three-dimensional geometric framework of the dam structure. The initial three-dimensional geometric framework is optimized by: adjusting the material parameters of the model surface based on the extracted material reflection characteristics; smoothing the geometric contours and repairing gaps based on multi-angle visual data; and mapping high-resolution textures onto the corresponding model surface.

2. The method for generating a three-dimensional model of a hydraulic engineering dam according to claim 1, characterized in that, Methods for multi-dimensional similarity matching and association analysis based on a pre-built database of hydraulic engineering dam components include: A first-level association is established by performing feature point matching and spatial analysis calculation between the data collected by the high-altitude acquisition platform at the first key visual marker point and the first group of components with known geometric attributes in the database. The central processing unit then determines at least one first spatial pose of the high-altitude acquisition platform relative to the dam structure. A second-level association is established by performing association analysis between data collected by the high-altitude acquisition platform at the second key visual marker point and a target component with unknown geometric attributes. This is combined with the attitude of the high-altitude acquisition platform determined in the first-level association, the known spatial positions of the first group of components, and other components with known geometric attributes in the database. The central processing unit then calculates the precise three-dimensional geometric dimensions and spatial positioning of the target component. The first key visual marker point and the second key visual marker point may be at the same or different spatial locations. The method for identifying key visual markers and determining platform pose includes: the central processing unit determining the theoretical position of at least one key visual marker in the real-world coordinate system based on pre-input dam design drawings or topographic map data; The central processing unit or the high-altitude acquisition platform itself uses a visual positioning algorithm to determine when the high-altitude acquisition platform arrives at or aligns with the effective data acquisition range of the key visual markers. The system controls the high-altitude acquisition platform to interact with the first set of field acquisition sensors located at known positions on or around the dam, or directly analyzes images containing the known first set of components through visual algorithms to obtain a series of distance or viewing angle measurements. Based on the set of measured values ​​and the known coordinates of the first set of field acquisition sensors or the known coordinates of the first set of components, spatial resection or multi-view geometric calculation is performed to determine at least one first spatial pose of the high-altitude acquisition platform at the key visual marker point.

3. The method for generating a three-dimensional model of a hydraulic engineering dam according to claim 1, characterized in that, The method also includes dynamic correction of database component parameters: when the target component already has an initial geometric parameter estimate in the database, the central processing unit compares the calculated three-dimensional geometric dimensions and spatial positioning with the initial geometric parameter estimate stored in the database. If the difference between the three-dimensional geometric dimensions and spatial positioning and the initial geometric parameter estimate exceeds a preset tolerance threshold, then the initial geometric parameter estimate in the database is updated using the three-dimensional geometric dimensions and spatial positioning.

4. The method for generating a three-dimensional model of a hydraulic engineering dam according to claim 1, characterized in that, The method further includes a step of modeling the dam's auxiliary structures using a continuous pose sequence: by performing continuous frame feature tracking and spatial intersection calculation between the data collected at multiple continuous pose points and the first group of components with known geometric properties in the database when the high-altitude acquisition platform moves along a predetermined trajectory through the target area, the central processing unit determines the multiple continuous spatial poses of the high-altitude acquisition platform within the target area. By performing stereoscopic visual analysis on the data collected from the high-altitude acquisition platform under multiple continuous spatial poses and on the dam ancillary structures within a target area to be determined, the central processing unit determines the three-dimensional model of the dam ancillary structures and their spatial position relative to the main dam body.

5. The method for generating a three-dimensional model of a hydraulic engineering dam according to claim 2, characterized in that, The spatial analysis method for multi-angle data correlation analysis in 3D model generation includes: First-level analysis: Using the high-altitude acquisition platform as a mobile known observation station, through multiple observations of the platform with known location points or known model components on the dam body, the precise trajectory and attitude sequence of the platform itself can be solved in reverse. Second-level analysis: Based on the known platform attitude, it is regarded as a sensor with a known spatial position. Using its observation data of the target component, combined with the observation constraints of other known position points on the same unknown structure, the precise geometry and position of the unknown structure in three-dimensional space are calculated through spatial forward intersection or multi-view triangulation algorithms. Third-level analysis: When dealing with multiple structural modules whose spatial relationships are unknown, the high-altitude acquisition platform is used to cross-observe them at multiple locations. The obtained observation network data is used for overall adjustment, and the relative positions and directions of all unknown modules are calculated to construct a self-consistent global three-dimensional model.

6. A system for generating a three-dimensional model of a hydraulic engineering dam, characterized in that, Includes a central processing unit, at least one high-altitude acquisition platform, and a pre-built database of hydraulic engineering dam components; The high-altitude acquisition platform is configured to be mounted on a mobile vehicle for acquiring image or video data of the target dam from multiple angles and heights. The central processing unit is communicatively connected to the high-altitude data acquisition platform and the database, and is configured to: By analyzing the data collected by the high-altitude acquisition platform at the first key visual marker point and performing feature point matching and spatial analysis calculation with the first group of components with known geometric properties in the database, the spatial pose of the high-altitude acquisition platform at the first key visual marker point is determined. By analyzing the data collected by the high-altitude acquisition platform at the second key visual marker point and performing correlation analysis with the second group of components, which includes at least one target component with geometric attributes to be determined and other components with known geometric attributes in the database, the three-dimensional geometric model of the target component and its location are determined. The first group of components and the second group of components have an inclusion relationship, the same set, or a non-intersecting relationship in the component classification hierarchy of the database. The first key visual marker point and the second key visual marker point are the same or different spatial location points. The central processing unit is configured to determine the expected spatial location range of the first key visual marker point based on the dam design blueprint or preliminary survey data. By analyzing the real-time video stream or image sequence transmitted back by the high-altitude acquisition platform, a visual recognition algorithm is used to automatically detect when the high-altitude acquisition platform enters the effective data acquisition window of the expected spatial location range; The system can trigger the high-altitude acquisition platform to perform collaborative measurements with the first set of field acquisition sensors deployed at known locations on the dam body, or trigger high-precision analysis of images containing specific known components to obtain a set of observation data for spatial calculation. Based on the observation data and the known precise coordinates of the first set of on-site acquisition sensors or specific known components, the precise spatial pose of the high-altitude acquisition platform at the first key visual marker point is calculated using photogrammetry or computer vision algorithms. The target component whose geometric attributes are to be determined has a current geometric parameter reference value in the database, and the central processing unit is further configured to: compare and analyze the three-dimensional geometric model and positioning data of the determined target component with the current geometric parameter reference value; if the determined data has a significant deviation from the current geometric parameter reference value, then automatically replace and update the current geometric parameter reference value in the database with the determined data, and record the metadata of this correction.

7. The system for generating a three-dimensional model of a hydraulic engineering dam according to claim 6, characterized in that, The central processing unit is also configured to: For each of the multiple continuous spatial poses recorded when the high-altitude acquisition platform moves along a scanning path, the specific pose of the high-altitude acquisition platform is determined by analyzing the data acquired in that pose and the first group of components with known geometric properties in the database. By combining the data collected by the high-altitude acquisition platform from a local structure of a dam to be modeled under multiple continuous spatial poses, and combining the known spatial poses of the high-altitude acquisition platform, multi-view stereo reconstruction and data fusion are performed to determine a complete and consistent three-dimensional model of the local structure of the dam.

8. The system for generating a three-dimensional model of a hydraulic engineering dam according to claim 6, characterized in that, The central processing unit is also configured to: calculate the spatial attitude of the high-altitude acquisition platform at the observation point by analyzing the data collected by the high-altitude acquisition platform at at least one preset observation point and the first group of components with known geometric properties in the database. A specific physical feature in the dam area scene, or an independent dam structural unit that needs to be integrated with the main dam model, is precisely associated and bound with the calculated spatial attitude and the image content acquired under that attitude, thereby establishing an accurate placement position for the feature or structural unit in the digital 3D scene.

9. The system for generating a three-dimensional model of a hydraulic engineering dam according to claim 6, characterized in that, The first group of components with known geometric properties are defined in the database with their spatial coordinates in a first coordinate system. The target component with the geometric properties to be determined originates from another independent measurement project, and its initial reference coordinates are defined in a second coordinate system. The central processing unit is further configured to, during correlation analysis, use the high-altitude acquisition platform as a common measurement bridge to solve the positioning of the target component and transform it to the first coordinate system, thereby realizing the fusion of different coordinate systems and data.

10. The system for generating a three-dimensional model of a hydraulic engineering dam according to claim 6, characterized in that, The high-altitude data acquisition platform integrates a composite sensor unit configured to operate in a first data acquisition mode and a second data acquisition mode. The first data acquisition mode is a high-resolution optical camera mode, used to acquire texture and color information; the second data acquisition mode is a lidar scanning mode or a depth sensing mode, used to directly acquire three-dimensional point cloud data. Furthermore, the first group of components in the database that are associated with the high-altitude acquisition platform at the first level have high-resolution texture features pre-stored, which are suitable for optical feature matching; while the known components in the second group of components that are associated with the high-altitude acquisition platform at the second level have accurate 3D point cloud or mesh models pre-stored, which are suitable for 3D point cloud registration or depth data fusion.

Images