Multi-data combined construction site three-dimensional model construction method and system

Abstract

The invention provides a multi-data combined construction site three-dimensional model construction method and system, and relates to the technical field of data processing, and the method comprises the steps: carrying out the calling of a preset resolution satellite image, and obtaining a full-region stereo image pair data set; after radiation correction is carried out, a region surface model and a region elevation model are generated; grid-level modeling defect detection is carried out, and distributed blind area blocks are positioned; performing obstacle avoidance path fitting to generate an obstacle avoidance optimized flight path; the unmanned aerial vehicle is driven to execute multi-angle surrounding scanning, and distributed blind area dense point clouds are directionally collected; constructing a distributed blind area model; and carrying out microscopic superposition, and outputting a three-dimensional model of the construction site. The technical problems that a construction site modeling method in the prior art is limited in the aspects of coverage, modeling efficiency and data real-time performance, depends on a single data source, cannot provide a comprehensive and accurate three-dimensional model of a construction site, and affects the construction progress and decision are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a construction site three-dimensional model construction method and system based on multi-data combination. BACKGROUND

[0002] With the continuous expansion of the scale of construction projects and the increasing complexity of the construction environment, the traditional construction site management method has been difficult to meet the requirements of modern engineering for efficiency and accuracy. Construction units not only need to master the overall spatial layout of the site, but also need to dynamically monitor the local structure, construction progress and environmental changes. Three-dimensional modeling technology can present the site information in an intuitive and interactive form, providing data support for construction organization, quality management and safety supervision. Therefore, how to efficiently and accurately obtain the three-dimensional spatial information of the construction site has become a key problem in the construction of engineering information.

[0003] At present, common construction site modeling methods include ground measurement, laser scanning and unmanned aerial vehicle aerial survey, although these methods perform well in local modeling, but there are still limitations in coverage range, modeling efficiency and data real-time, at the same time, a single data source often cannot balance macroscopic range and detail accuracy, resulting in the incompleteness of the model in application. SUMMARY

[0004] The present application provides a construction site three-dimensional model construction method and system based on multi-data combination, aiming at solving the technical problems that the existing construction site modeling method has limitations in coverage range, modeling efficiency and data real-time, and relies on a single data source, which cannot provide a comprehensive and accurate three-dimensional model of the construction site, affecting the construction progress and decision-making.

[0005] The first aspect of the present application provides a construction site three-dimensional model construction method based on multi-data combination, the method comprising: taking the minimum bounding rectangle boundary of the construction area as the geographic coordinate system parameter, performing preset resolution satellite image retrieval, and obtaining a full-area stereo image pair data set; after radiometric correction of the full-area stereo image pair data set, generating a regional surface model and a regional elevation model by fusing stereo image pairs; positioning the distributed blind area block by detecting grid-level modeling defects of the regional surface model and the regional elevation model; generating an obstacle-avoiding optimized flight trajectory based on the distributed blind area block; driving a UAV to perform multi-angle circumferential scanning on the distributed blind area block along the obstacle-avoiding optimized flight trajectory to directionally collect distributed blind area dense point clouds, wherein the UAV is equipped with a binocular stereo camera and an IMU; constructing a distributed blind area model based on the distributed blind area dense point clouds; microscopically superimposing the distributed blind area model on the regional surface model and the regional elevation model as a spatial reference framework, and outputting a construction site three-dimensional model.

[0006] In a second aspect, the application discloses a multi-data combined construction site three-dimensional model construction system, which is used for the multi-data combined construction site three-dimensional model construction method. The system comprises a satellite image calling module, which is used for calling satellite images of a preset resolution according to the minimum bounding rectangle boundary of a construction area as a geographic coordinate system parameter, and obtaining a full-area stereo image pair data set; a height model generation module, which is used for generating a regional surface model and a regional height model by fusing the stereo image pair after radiation correction of the full-area stereo image pair data set; a defect detection module, which is used for locating a distributed blind area block by performing grid-level modeling defect detection on the regional surface model and the regional height model; a path fitting module, which is used for fitting an obstacle-avoiding path based on the distributed blind area block, and generating an obstacle-avoiding optimized flight trajectory; a point cloud generation module, which is used for driving a UAV to perform multi-angle surrounding scanning on the distributed blind area block along the obstacle-avoiding optimized flight trajectory, and collecting a distributed blind area dense point cloud in a directional manner, wherein the UAV is provided with a binocular stereo camera and an IMU; a blind area model construction module, which is used for constructing a distributed blind area model based on the distributed blind area dense point cloud; and a micro superposition module, which is used for taking the regional surface model and the regional height model as a spatial reference framework, performing micro superposition of the distributed blind area model, and outputting a construction site three-dimensional model.

[0007] The one or more technical solutions provided in the application have at least the following beneficial effects: By taking the minimum outer package rectangular boundary of the construction area as the geographic coordinate system parameter, the advantages of high resolution of satellite images can be used to obtain detailed geographic information of the entire construction area; the radiation calibration preserves the radiation consistency of satellite image data, eliminates the image quality differences caused by different shooting conditions, which can obtain more accurate image data, and provides accurate terrain and surface data for the generation of surface and elevation models by stereo image pair fusion; the grid-level modeling defect detection can identify the data defect area in the model based on the rasterization processing of the regional surface and elevation model, and through fine rasterization analysis, the problems in modeling can be detected to realize accurate identification of blind areas; by analyzing the distributed blind area, the flight path can be fitted based on the position and shape of the blind area, and the generated obstacle avoidance optimized flight trajectory ensures that the unmanned aerial vehicle can bypass obstacles in the complex construction site environment when performing flight tasks, and adjust the flight path according to the terrain and structural characteristics to optimize the flight efficiency and data acquisition accuracy; through multi-angle ring scanning, the unmanned aerial vehicle can obtain detailed data of the blind area region from multiple perspectives, increase the comprehensiveness of data acquisition, and the unmanned aerial vehicle equipped with binocular stereo camera and IMU can accurately record the position and attitude in real time. The IMU provides high-precision positioning information for the SLAM system, and the binocular camera is used to obtain high-definition images, which can generate dense point cloud data by combining depth map information, providing rich three-dimensional information for subsequent modeling; using the dense point cloud data collected from the unmanned aerial vehicle, a high-precision distributed blind area model is constructed, which can accurately reflect the three-dimensional spatial form of the blind area; by taking the regional surface model and the regional elevation model as the basic space framework, the distributed blind area model is microscopically superimposed with the terrain data to ensure that the final three-dimensional model has consistent spatial reference and accuracy, and ensures the high-precision three-dimensional visualization of the construction site.

[0008] The above description is only a summary of the technical solutions of the present application. In order to more clearly understand the technical means of the present application, the following specific embodiments of the present application can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the following specific embodiments of the present application are described. BRIEF DESCRIPTION OF DRAWINGS

[0009] Figure 1 The multi-data combined construction site three-dimensional model construction method flowchart provided by the embodiments of the present application.

[0010] Figure 2 The multi-data combined construction site three-dimensional model construction system structure schematic diagram provided by the embodiments of the present application.

[0011] Explanation of reference numerals: satellite image retrieval module 10, elevation model generation module 20, defect detection module 30, path fitting module 40, point cloud generation module 50, blind area model construction module 60, micro superposition module 70. DETAILED DESCRIPTION

[0012] To further clarify the technical means and effects taken by the present application to achieve the predetermined inventive purpose and its effects, the specific embodiments, structures, features and effects according to the present application are described in detail below in conjunction with the drawings and preferred embodiments.

[0013] Embodiment one, as shown in the present application, a multi-data combined construction site three-dimensional model construction method is provided, the method comprises: Figure 1 According to the minimum bounding rectangle boundary of the construction area as the geographic coordinate system parameter, the preset resolution satellite image is called to obtain the full-area stereo image pair data set. The minimum bounding rectangle boundary of the construction area is determined, which is a rectangular boundary box that can completely contain all geographic information of the construction area. The boundary of the rectangle is defined by the geographic coordinates of the construction area. Specifically, by obtaining the latitude and longitude coordinate points of the construction area, the four boundaries of the minimum bounding rectangle of the area, such as east, west, south, and north, are calculated. These four boundaries are used as the extraction area of the subsequent image data.

[0014] Based on the minimum bounding rectangle boundary, the preset resolution satellite image is called. When calling, the resolution of the image needs to be set to ensure that the obtained image data can meet the needs of subsequent modeling and analysis. It can be obtained from a public satellite data provider. Through the rectangular area, an image with a suitable resolution is selected for download to form an image data set. The called image is generally a single image, but in order to perform subsequent stereo modeling, a stereo image pair with parallax needs to be obtained. By using different shooting angles or time periods of satellite images, a full-area stereo image pair data set is formed. The full-area stereo image pair data set can provide basic information for subsequent three-dimensional modeling and elevation extraction.

[0015] After the full-area stereo image pair data set is radiometrically corrected, the regional surface model and the regional elevation model are generated by fusing the stereo image pair.

[0016] The satellite image will be affected by factors such as atmosphere, solar angle, and terrain during the collection process, resulting in uneven radiation and affecting subsequent analysis and modeling. Therefore, the called image needs to be radiometrically corrected first to eliminate these effects. Radiometric correction is to make the pixel value of the image reflect the true ground radiation information. Radiometric correction includes atmospheric correction, sensor correction, and terrain correction.

[0017] The satellite image will be affected by factors such as atmosphere, solar angle, and terrain during the collection process, resulting in uneven radiation and affecting subsequent analysis and modeling. Therefore, the called image needs to be radiometrically corrected first to eliminate these effects. Radiometric correction is to make the pixel value of the image reflect the true ground radiation information. Radiometric correction includes atmospheric correction, sensor correction, and terrain correction.

[0018] The stereoscopic image after radiation correction is processed by stereovision to obtain a regional surface model. The three-dimensional shape of the ground object is estimated by calculating the parallax of the image pair, i.e., the displacement difference between pixels. Specifically, the parallax of each pair of corresponding pixels in the stereoscopic image pair is calculated, and the three-dimensional coordinates of each pixel are further obtained by triangulation to reconstruct the three-dimensional surface model of the region. The three-dimensional surface model reflects the ups and downs of the ground and represents the terrain features of the ground surface as a triangular mesh or a point cloud dataset. The regional elevation model (DEM, Digital Elevation Model) is calculated by fusing the stereoscopic image data. Each pixel corresponds to a ground height value. The regional elevation model provides spatial distribution information of the ground height, which is used for subsequent terrain analysis, path planning, etc.

[0019] By performing grid-level modeling defect detection on the regional surface model and the regional elevation model, the distributed blind area block is located.

[0020] In order to detect modeling defects in the regional surface model and the regional elevation model, the models are converted into a grid data structure. Rasterization divides the continuous space of the model into uniform small grids, and each grid cell stores a numerical value such as height, texture information, etc.

[0021] Defect detection is to identify inconsistencies or errors in the model, which are usually manifested as areas with unclear textures, geometric distortions, or inaccurate elevations. These defects are caused by image quality problems, calibration errors, or problems in the data fusion process. They include: texture defect detection, which identifies areas with unclear or distorted textures by evaluating the texture of the surface model. Gradient analysis, noise detection, etc. can be used to output a quantitative clarity array; elevation defect detection, which identifies areas with abnormal elevation changes by performing Gaussian mutation detection on the elevation model. Places with sudden changes in elevation can cause modeling errors or loss of terrain features.

[0022] By defect detection results, the regions with blind areas in the model are identified and located. These blind areas are caused by modeling defects, sensor errors, or data insufficiency, which may affect subsequent path planning or modeling accuracy. The detected low-precision grid area is combined with the elevation information to further refine the boundary and morphology of the blind area by methods such as watershed algorithm, and finally the distributed blind area block is located to provide data support for subsequent obstacle avoidance path planning.

[0023] Based on the distributed blind area block, an obstacle avoidance path is fitted to generate an obstacle avoidance optimized flight trajectory.

[0024] Drones need to avoid distributed blind spots and fly within feasible areas to prevent collisions with obstacles. The flight path must meet multiple constraints, such as safe flight altitude, flight speed, and energy consumption. First, for each blind spot, its center point is determined and used as a key node for path planning. Based on the blind spot center point, a planar flight trajectory is fitted. This trajectory needs to consider obstacles in the flight space and avoid blind spots as much as possible during planning. An area elevation model is used to identify elevation obstacles in the flight path, such as buildings and mountains, ensuring sufficient safe vertical distance. The initially fitted flight trajectory is optimized by performing elevation smoothing corrections to eliminate obstacle interference and ensure that the flight trajectory avoids collisions during actual execution. In path optimization, the drone's flight characteristics, such as maximum flight speed and rotation radius, are considered, and a path fitting algorithm, such as the A* algorithm, is used to fine-tune the obstacle avoidance path, ultimately generating an optimized obstacle avoidance flight trajectory.

[0025] The drone is driven to perform multi-angle surround scans of the distributed blind zone block along the obstacle avoidance optimized flight trajectory, and to collect dense point clouds of the distributed blind zone in a directional manner. The drone is equipped with a binocular stereo camera and an IMU.

[0026] Based on the generated obstacle avoidance optimized flight trajectory, the drone is driven to fly along this trajectory. During flight, the drone performs multi-angle surround scanning, meaning it scans along the trajectory from more than one perspective, ensuring coverage of all angles in blind spots and avoiding data acquisition dead zones. The drone captures a series of stereo images using its onboard binocular stereo camera. Utilizing the principle of stereo vision, the binocular camera calculates the depth information of the scene through image comparison. The IMU provides information about the drone's attitude, acceleration, angular velocity, etc. During flight, the data from the IMU and the binocular camera are tightly coupled, providing more accurate pose information. The IMU helps to offset potential image registration errors and ensures the accuracy and consistency of point cloud generation through data fusion.

[0027] The drone follows a planned path and performs multi-view spiral or circling flight. The image sequence of each viewpoint generates a depth map by matching the features of adjacent frames and combining the pre-integration results of the IMU raw data. Finally, through a point cloud reconstruction algorithm, a distributed blind zone dense point cloud is generated for the distributed blind zone area.

[0028] A distributed blind zone model is constructed based on the dense point cloud of the distributed blind zone.

[0029] Based on the geometric continuity analysis, the point cloud is segmented, and different boundaries, mutations and other features in the point cloud can be identified by normal vector calculation, and then different building component or structure area is segmented. These segmented areas are building component point clouds, which can provide necessary information for subsequent blind area modeling. Extract the local FPFH feature descriptor of each building component point cloud. This is a common point cloud feature extraction method that can find similar structures in point clouds. After extracting the FPFH features of multiple building components, match these features to find matching reusable components from the historical incremental model library. This step can speed up model construction and avoid starting from scratch. Using the extracted features and reusable components, use the Poisson reconstruction algorithm to generate a surface model from the point cloud data, further refine the blind area model, and finally align the reusable components with the point cloud reconstruction results and perform boundary fusion through rigid transformation to finally construct a complete distributed blind area model.

[0030] The regional surface model and the regional elevation model are used as a spatial reference framework for micro-stacking of the distributed blind area model, and a three-dimensional model of the construction site is output.

[0031] To ensure that the geographic coordinates of the regional surface model, the elevation model and the blind area model are consistent, first, the regional surface model and the regional elevation model are registered through ground control points. Control points are points with known locations obtained through GPS or other positioning technologies. Micro-stacking of the distributed blind area model with the regional surface model and the regional elevation model means embedding the distributed blind area model into the actual model framework of the terrain to ensure that the three-dimensional modeling of the entire construction site conforms to the geographical actual situation. In the merging process, the joint problem between different models is handled by performing boundary fusion to ensure that models from different sources can transition naturally at the joint, avoiding obvious joint marks. After micro-stacking and joint fusion, the final three-dimensional model of the construction site contains all key terrain, structure and blind area information and can be used for visualization, analysis, planning and decision support.

[0032] Further, by performing grid-level modeling defect detection on the regional surface model and the regional elevation model, the distributed blind area block is located, and the method comprises: After spatially aligning the regional surface model and the regional elevation model, the regional surface model and the regional elevation model are segmented in parallel based on a preset grid scale, to obtain a surface model grid array and an elevation model grid array; a quantified definition array is output by performing texture gradient evaluation on the surface model grid array; a quantified complexity array is output by performing Gaussian mutation detection on the elevation model grid array; a modeling defect probability array is obtained by spatially aligning the quantified definition array and the quantified complexity array for double-index fusion weighting; and the distributed blind area block is located by performing terrain continuity merging on the low-precision grid of the modeling defect probability array based on a watershed algorithm.

[0033] Spatial alignment refers to registering the regional surface model and the regional elevation model in a spatial coordinate system, so that they share the same coordinate system. Since the two models come from different sensors or data sets, it is necessary to ensure that their geometry and spatial reference are consistent, and avoid errors caused by coordinate deviation. The alignment method includes aligning key feature points or using known geographic coordinate systems for conversion. A registration algorithm can be used to achieve accurate alignment.

[0034] The aligned regional surface model and the regional elevation model are segmented in parallel according to a preset grid scale, which means that the entire region is divided into multiple small units according to a fixed grid size, and each grid unit represents a geographic area and contains a surface value and an elevation value. The selection of the grid scale needs to be adjusted according to the actual situation. If the grid is too large, details will be ignored; if it is too small, the amount of calculation will be too large. Generally, the grid size is balanced according to the specific requirements of the region, the data resolution, and the computing power. The processed regional surface and regional elevation model are converted into a surface model grid array and an elevation model grid array, respectively. These two arrays contain grid data of the regional surface and the regional elevation, respectively.

[0035] Texture gradient is an important feature of the surface model, which refers to the degree of gray level change in an image or surface within a certain area. In three-dimensional modeling, texture gradient can reflect whether the terrain surface is smooth and whether there are large-scale texture changes. The texture gradient evaluation process is based on the rasterized data of the surface model. Specifically, the gradient of the gray value or texture value in each grid cell is calculated, and the gradient operator such as the Sobel operator is used to calculate the texture change of the surrounding neighborhood for each grid cell. The texture change intensity of the region is obtained by gradient calculation.

[0036] Through the texture gradient evaluation, a quantized definition array is obtained, the value of each grid cell of the array represents the texture definition of the region, for example, the region with sharp texture change corresponds to larger gradient value, higher definition; the region with relatively gentle texture change, smaller gradient value, lower definition. The quantized definition array is a quantization processing of the texture gradient evaluation result, which converts the definition value of each grid cell into a scalar value between 0 and 1, where 1 represents the maximum definition and 0 represents the worst definition.

[0037] Gaussian mutation detection is a technique for detecting sharp changes or mutations in data, in the height model, mutation means that the change of terrain is very large, such as cliffs, sharp slope changes, etc. Gaussian mutation detection is based on Gaussian filter, Gaussian filter is a smoothing filter that can smooth out high-frequency noise and highlight low-frequency features. In height data, the grid data is smoothed by Gaussian filtering, and then the change rate of each grid cell is calculated, such as first derivative, to detect the mutation area. Mutation points are usually places with sharp height changes, such as cliff edges or places with large slopes. The goal of Gaussian mutation detection is to find these areas and mark them as high complexity areas. According to the results of Gaussian mutation detection, a quantized complexity array is generated, which reflects the complexity of the region height model. Areas with higher complexity are usually places with large terrain undulations, representing higher modeling difficulty.

[0038] Double-index fusion weighting is to combine the information in the quantized definition array and the quantized complexity array according to a certain proportion. In this process, definition and complexity have different effects on modeling defects, so different weights need to be given to these two indexes. For example, if the difference in surface definition is larger, it has a greater impact on modeling accuracy, and a higher weight can be given to definition. After weighting, a modeling defect probability array is formed, the value of each grid cell represents the probability of the region occurring modeling defects, this array can reflect which areas may have lower accuracy, modeling difficulty or blind area, which is convenient for subsequent blind area positioning.

[0039] The watershed algorithm is a classic image processing technique widely used in image segmentation. The basic idea is to treat the entire image as a terrain surface, with the image's gray values representing the terrain's elevation. The low points of the terrain represent basins or depressions. By simulating water flow from low points to high points, the watershed algorithm divides the image into different regions, with the boundaries of the segmentation being the watershed lines. In this step, the modeled defect probability array is treated as a terrain surface, with high-probability defect regions being considered as low-lying areas, i.e., blind zones. The watershed algorithm segments different blind zone blocks based on the terrain continuity of these low-lying areas. Through the watershed algorithm, the low-precision areas in the modeled defect probability array are first merged. The core of the merging process is to use the continuity of terrain elevation and defect probability to automatically merge grid cells that are connected and have similar defect probabilities, forming continuous blind zone regions. These continuous merged regions are called distributed blind zone blocks. In these regions, there are significant modeling accuracy issues that can lead to inaccurate subsequent modeling or data loss. These blind zone regions are the concentration of modeling defects in the entire construction site, and subsequent optimization and repair can be based on these regions.

[0040] Further, the terrain continuity merging of the low-precision grid of the modeled defect probability array based on the watershed algorithm locates the distributed blind zone blocks, including: Traverse the modeled defect probability array to screen distributed low-precision grids that meet a preset defect threshold. Use the terrain gradient field of the elevation model grid array as the input source, mark the distributed low-precision grids as initial depressions, perform terrain gradient-driven watershed inundation simulation, and generate continuous watershed partitions constrained by watershed boundaries. Use the terrain elevation gradient change threshold to traverse the continuous watershed partitions, perform non-continuous terrain segmentation exclusion, and locate the distributed blind zone blocks.

[0041] To effectively locate the blind zone, a defect threshold is preset. For example, if the modeled defect probability is higher than a certain threshold, such as 0.8, it is considered that the area belongs to a low-precision area and needs further processing. Traverse the modeled defect probability array to screen all grids higher than the defect threshold. These high-probability defect regions are distributed low-precision grids.

[0042] The terrain gradient field is calculated based on the elevation model grid array. It describes the rate of terrain change and reflects the slope and elevation change of different regions. A large terrain gradient indicates a sharp change in terrain, while a small gradient indicates a relatively flat terrain. The terrain gradient field can provide input for the watershed algorithm to help determine which areas belong to low-lying areas and which areas belong to high-lying areas.

[0043] The screened distributed low-precision grid is marked as an initial depression, that is, these areas are blind areas or modeling defect areas that need further attention. The watershed flooding simulation is the core process of the watershed algorithm, which simulates the flow from low-lying areas to highlands. Through this simulation, the watershed boundaries between different areas can be determined. In this process, the low-precision area with a high defect probability value (i.e., low-lying land) is flooded into a watershed area, and the high-precision area (i.e., highland) is divided into a watershed line. Through the watershed flooding simulation, a continuous watershed partition with a watershed boundary is finally formed, and each watershed represents a region with strong terrain continuity, and the areas within the watershed have relatively consistent elevations or defect probabilities.

[0044] To better screen the continuous watershed partition, a terrain elevation gradient change threshold is set, which is used to determine whether there is a significant discontinuous change in the terrain within the watershed. When the elevation gradient threshold is high, only in the watershed with gentle terrain changes will it continue to be retained. Each continuous watershed partition is traversed to check the terrain gradient change within it. If there is a significant elevation mutation or discontinuous terrain in a watershed, for example, a sudden height difference, it is considered that this area is not a continuous watershed and may be a local small-scale blind area. The watershed is screened by the elevation gradient change threshold to exclude areas with discontinuous terrain, that is, areas with large changes in elevation. After excluding non-continuous terrain, the remaining areas are the final distributed blind area blocks, which are the worst places for modeling accuracy.

[0045] Further, it also includes: Correlating construction progress data to obtain an updated construction zoning; analyzing the newly added structure area of the updated construction zoning to perform iterative incremental updating of the distributed blind area block.

[0046] The construction progress data includes construction phase, construction area, start and end time of construction task, equipment and personnel scheduling, etc. These data can be tracked and updated in real time through the construction management system. According to the latest construction progress data, the division of the construction area is updated, for example, if the construction of a certain area has been completed, the modeling accuracy of that area has been improved; on the contrary, if a new area has just started construction, its modeling accuracy is still low. Through this update, the latest updated construction zoning is obtained, providing accurate geographic information for subsequent blind area positioning and modeling.

[0047] After the construction progress data is updated, some new structural areas are introduced into the construction process, such as newly added building floors, independent buildings, underground facilities, etc. The positions, sizes and complexities of these new structural areas are analyzed, as well as their influence on the blind area of the construction site. For example, the excavation or construction of some areas causes changes in the original blind area or the appearance of new blind areas. Incremental updating refers to adjusting the existing blind area block model in combination with the newly added structural areas, for example, according to the construction state of the new areas, merging, correcting or fine-tuning the existing blind area block model to maintain the update of the blind area block of the entire site. As the construction progresses, new structural areas are continuously introduced, and at the end of each stage, the blind area block is iteratively updated to continuously optimize the distribution and accuracy of the blind area, ensuring that the construction data and modeling accuracy of each stage match.

[0048] Further, based on the distributed blind area block, an obstacle avoidance path is fitted to generate an obstacle avoidance optimized flight trajectory, comprising: The center point of the distributed blind area block is taken as a waypoint to fit a planar flight trajectory; the planar flight trajectory is projected to the regional elevation model to identify a plurality of elevation obstacles with a preset safe vertical distance; and the three-dimensional elevation smoothing correction of the planar flight trajectory is performed according to the plurality of elevation obstacles to output the obstacle avoidance optimized flight trajectory.

[0049] The center point of the distributed blind area block is the position of the geometric center of these blocks. The center point of the distributed blind area block is selected as the waypoint, i.e. the UAV passes through these waypoints during flight. In order to plan the flight path, all waypoints are fitted in the plane, i.e. in the horizontal plane, a flight trajectory is determined according to the positions of all waypoints, and an algorithm such as the shortest path algorithm or smooth path fitting is used to determine the path, thereby obtaining a planar flight trajectory. This planar flight trajectory is the ideal flight path of the UAV from one blind area to another blind area. However, this trajectory does not take into account actual spatial obstacles or elevation differences, so further optimization and adjustment are required.

[0050] The projected planar flight trajectory is projected into a regional elevation model to obtain a three-dimensional flight trajectory reflecting the actual terrain. The regional elevation model provides height data for various locations, and the projection can obtain the changes of the planar trajectory in the elevation dimension. The safety vertical distance refers to the minimum vertical distance that needs to be maintained between the UAV and the ground or other obstacles, such as buildings, trees, power lines, and other elevation obstacles that can affect the flight path of the UAV. Through the regional elevation model, obstacles above the flight path are identified, including high buildings or tower cranes, large topographic changes, temporary construction equipment or facilities, etc. According to the preset safety vertical distance, it is determined whether these obstacles will interfere with the flight of the UAV. If the obstacles are high, the flight path needs to be adjusted. According to the elevation differences of the terrain and the actual positions of the obstacles, multiple elevation obstacles are accurately positioned.

[0051] According to the multiple elevation obstacles, the planar flight trajectory is adjusted to ensure that the height of the flight trajectory in the three-dimensional space does not conflict with the obstacles. This step combines the elevation data of the region to make vertical corrections to the original planar flight path, so that the UAV can fly at a safe height and avoid collisions with elevation obstacles.

[0052] After the elevation correction of the flight trajectory, a discontinuous or non-smooth flight path is generated. In order to avoid sudden sharp turns or drastic changes in height during the flight of the UAV, the flight trajectory needs to be smoothed. Curve smoothing algorithms such as Bezier curves and spline curves are used to smooth the corrected flight path, making the flight trajectory more natural and avoiding unstable flight control due to sharp changes. The final obstacle avoidance optimized flight trajectory can ensure that the UAV can fly along a safe and smooth path.

[0053] Further, driving the UAV to perform multi-angle circumferential scanning on the distributed blind area block along the obstacle avoidance optimized flight trajectory, and directionally collecting the distributed blind area dense point cloud, including: Based on the tightly coupled SLAM system constructed by the binocular stereo camera and the IMU, multi-view spiral circumferential observation is performed along the flight trajectory to collect a stereo image sequence and IMU raw measurement data. After real-time adjacent frame matching of the stereo image sequence through point-line fusion feature extraction, combined with the optimized pose output by pre-integration of the IMU raw measurement data, depth map reconstruction processing is performed, and a time series depth map is output. The time series depth map is fused through TSDF to generate the distributed blind area dense point cloud.

[0054] The binocular camera equipped on the UAV can capture images with stereo information, and through feature extraction such as image contrast, texture, etc., the three-dimensional structure of the scene can be displayed. During flight, the binocular camera takes a series of stereo images along the flight path, which records spatial information from multiple perspectives. The IMU can provide real-time acceleration, angular velocity, and attitude information of the UAV, which is used for high-precision positioning and flight trajectory optimization. Through the raw measurement data of the IMU, the position and attitude of the UAV can be estimated more accurately, especially in cases where visual information is insufficient or discontinuous.

[0055] The tightly coupled SLAM system jointly processes binocular camera and IMU data, and simultaneously performs positioning and map construction during flight. This tightly coupled method can effectively reduce the error accumulation when using visual or inertial data alone. Through multi-view spiral observation, the UAV can observe the blind area from multiple angles and obtain more comprehensive spatial information. In order to maximize the coverage of data collection, the UAV performs multi-angle spiral scanning along the spiral path, which ensures that the data obtained from different angles has good overlapping areas, reduces blind areas, and improves the quality of subsequent point cloud reconstruction.

[0056] Using the image sequence captured by the stereo camera, the features of each frame of image are extracted through point-line fusion. Feature extraction includes extracting key points such as corner points, edges, and texture information in the image. Adjacent frame matching is performed between adjacent two frames of images to find their corresponding relationship, and through these matching points, the relative position and attitude of the UAV during image shooting can be calculated. The data provided by the IMU includes acceleration, angular velocity, and attitude information. Through pre-integration of IMU data, the incremental change of the UAV's position during flight can be calculated without visual data, and it can help to improve the errors that may exist in the image matching process.

[0057] The pre-integrated data of the IMU and the matching results of the adjacent frame images are combined to optimize the pose estimation. Through optimization algorithm, the pose estimation of each frame is optimized to reduce system error. Through the optimized pose estimation and image feature matching results, a depth map is reconstructed, which represents the distance from each pixel point to the camera. The depth map represents the three-dimensional coordinates of each pixel, which is realized through triangulation method or stereo matching algorithm. Through continuous images and pose estimation, a time-series depth map is generated, and the depth information corresponding to each frame of image is arranged in time sequence, providing basic data for subsequent point cloud generation.

[0058] TSDF (Truncated Signed Distance Function) is a voxel representation method used to represent the distance of object surface in three-dimensional space. TSDF calculates the distance of each voxel to the object surface and represents the position of the voxel inside or outside the object according to the sign of the distance. Positive value represents the outside of the object, negative value represents the inside of the object, and zero value represents the object surface. All time-series depth maps are input into TSDF for fusion to generate a global three-dimensional model. The depth map information at each time will affect the voxel value of the corresponding region, thereby realizing the gradual construction of three-dimensional space. Based on TSDF representation, depth data is converted into dense three-dimensional point cloud through voxel rendering or reconstruction method. These point clouds have very high spatial density and can accurately describe the surface and structure of distributed blind area. Through this process, the generated point cloud not only has fine geometric structure, but also reflects the actual situation of the construction site, especially in complex or occluded areas. Finally, a complete distributed blind area dense point cloud is generated, which contains detailed three-dimensional spatial information of all key blind areas and can be used for subsequent modeling, analysis and construction decision-making.

[0059] Further, based on the distributed blind area dense point cloud, a distributed blind area model is constructed, and the method comprises: Based on geometric continuity, the distributed blind area dense point cloud is segmented into multiple building component point clouds at the boundary of point cloud normal vector mutation; multiple groups of FPFH feature descriptors of the multiple building component point clouds are extracted; the multiple groups of FPFH feature descriptors are aggregated based on structure feature comparison to obtain N reference FPFH descriptors; the N reference FPFH descriptors are used to extract N reused components from a historical incremental model library; after stripping the component point cloud corresponding to the N reference FPFH descriptors from the multiple building component point clouds, Poisson reconstruction is performed to obtain an initial blind area model; the N reused components are rigidly transformed and aligned to the initial blind area model according to the multiple groups of FPFH feature descriptors for boundary fusion processing to generate the distributed blind area model.

[0060] For each point in the distributed blind area dense point cloud, its normal vector is calculated. The normal vector is a vector representing the orientation of the surface where the point is located. The normal vector can be estimated using the geometric relationship of neighboring points. For each point, the K-nearest neighbor algorithm can be used to calculate the covariance matrix of these points, and then the normal vector can be obtained through eigenvalue decomposition.

[0061] Geometric continuity refers to whether the normal vector of the point cloud surface changes smoothly in adjacent regions. Building components are composed of different geometric surfaces, such as walls, floors, and roofs, and there are obvious mutations in the normal vectors between them. When the normal vector changes, it indicates that the point may be the boundary or joint between the building surfaces. Therefore, detecting the mutation of the normal vector, i.e., the change in the angle between the normal vectors, can effectively segment different building components.

[0062] Based on the mutation of the normal vector, a threshold method is used, for example, setting the angle between the normal vectors greater than a certain threshold as the boundary, to segment the point cloud. The region with larger mutation is divided into different building component point clouds, and each part represents a building component or a part thereof.

[0063] FPFH is a feature descriptor based on local geometry, which is used for point cloud matching and recognition. It constructs a descriptor by analyzing the geometric relationship of each point, including the normal vector and distance. The FPFH feature calculates a feature vector by counting the normal vector distribution information of the neighborhood around each point. This feature vector can capture the shape features of the local region of the point cloud.

[0064] For each building component point cloud, the FPFH feature is extracted by selecting an appropriate neighborhood size, i.e., the number of neighboring points or the spatial range of each point. In the neighborhood of each point, the normal vectors and geometric relationships of all neighboring points are calculated, and based on this information, the FPFH descriptor is calculated, including: calculating the normal vector of each point, counting the normal vector information of the points in the neighborhood, such as the angle distribution of the normal vector, generating the FPFH feature descriptor as the local feature of the point. These feature descriptors can reflect the local geometric structure and shape of the building component.

[0065] By calculating the similarity of the FPFH feature descriptors between different building component point clouds, it is found out which descriptors have higher similarity. The similarity measurement methods include Euclidean distance, cosine similarity, etc. After comparison, the FPFH descriptors with similar structures are aggregated to form the reference FPFH descriptor representing the overall feature of the building component. This process converts local features to global descriptions, thereby modeling the building component as a whole. N reference FPFH descriptors represent the most significant features of the building component, where N is a positive integer.

[0066] The historical incremental model library contains building components or structural units that have been constructed and verified, each of which is encoded by different feature descriptors. Based on the extracted N benchmark FPFH descriptors, a feature matching algorithm is used to compare with the components in the historical incremental model library, such as using nearest neighbor search to match by calculating the distance between descriptors, finding the most similar historical component to the current building component point cloud, the smaller the distance between descriptors, the higher the similarity in geometric structure. Extract N reusable components from the historical incremental model library that are similar to the target building component, these reusable components provide verified standard components for subsequent modeling, thereby reducing the workload of re-modeling and improving modeling efficiency and accuracy.

[0067] Based on N benchmark FPFH descriptors, the corresponding component point cloud is stripped from multiple building component point clouds, this process is completed by pairing the target point cloud with the feature descriptors, and the most relevant point cloud data is selected. Poisson reconstruction is a point cloud surface reconstruction technique that constructs a continuous and smooth three-dimensional surface by using normal vector information and local geometric structure information of the point cloud. This algorithm is based on the solution of the Poisson equation in mathematics, aiming to generate a smooth surface model from given point cloud data. In blind area modeling, Poisson reconstruction can effectively fill the gaps in the point cloud and restore the complete surface of the building component, this step converts the stripped point cloud into an initial blind area model.

[0068] Rigid body transformation refers to the operation of rotating, translating, etc. to align the reusable component with the initial blind area model in space, this process requires accurate calculation of the transformation matrix between the reusable component and the blind area model, including rotation matrix and translation vector. In the alignment process, multiple sets of FPFH feature descriptors are used as matching basis to ensure geometric alignment of the reusable component and the blind area model.

[0069] After rigid body transformation, the reusable component is accurately aligned to the corresponding position of the initial blind area model. However, there may be joints or boundary mismatches between the initial model and the newly added component. Boundary fusion refers to smoothing the joints to make the transition between components more natural and eliminate obvious splicing marks. This is achieved by smoothing or refining the edges of the point cloud. After completing boundary fusion, all building components and structural units will seamlessly combine into a complete distributed blind area model, which reflects the distribution of blind areas on the construction site and provides detailed three-dimensional spatial information, which can be used for further analysis, simulation or construction planning.

[0070] Further, the regional surface model and the regional elevation model are used as a spatial reference framework for micro-superposition of the distributed blind area model, and a three-dimensional model of the construction site is output, including: After registering the regional surface model and the regional elevation model in the geographic coordinate system through ground control points, the grid block deletion of the distributed blind area block is performed to obtain a terrain reservation reference framework; after embedding the distributed blind area model in the terrain reservation reference framework, joint fusion is performed to obtain the construction site three-dimensional model.

[0071] Ground control points are specific points with known geographic coordinates used to provide accurate alignment and positioning for different spatial data sets. In geographic modeling, ground control points are used as reference points to convert regional surface models and regional elevation models from relative coordinate systems to real-world geographic coordinate systems. These ground control points can come from actual measurement data such as GPS measurements, total station measurements, or pre-collected geographic information system data. Registration is the process of converting the coordinate systems of regional surface models and regional elevation models to a unified geographic coordinate system, which is achieved through translation, rotation, and scaling adjustments, so that the two models are completely aligned in the same spatial coordinate system.

[0072] After registration, according to actual construction requirements, some irrelevant or inaccurate areas are removed, which can be grid blocks occupied by vegetation, buildings or other ground objects that do not meet the requirements. This process involves analysis of rasterized data, and by removing grid blocks that do not belong to the target area, the effective and required terrain area is retained, and finally the terrain reservation reference framework is output, which only retains the effective area and elevation data of the construction site.

[0073] Through precise spatial registration and geometric alignment, the distributed blind area model is embedded into the reserved terrain reservation reference framework, ensuring that the blind area model is consistent with the physical space of the terrain framework. This step ensures the seamless integration of blind areas and terrain in the construction site. Even after successful embedding of the blind area model and the terrain model, there may still be joints or mismatches between the models at the boundaries, which need to be jointed to smooth the transition between different areas and avoid visual cracks or splicing marks. Joint fusion uses methods such as point cloud smoothing, gridding fusion or texture mapping to ensure seamless connection at the model joint. After joint fusion, the final construction site three-dimensional model is output, which contains the complete terrain, blind area, building structure and other related spatial information of the construction site.

[0074] Further, histogram matching is used to unify the radiation of the drone close-range texture and satellite image texture of the construction site three-dimensional model.

[0075] Histogram matching is a technique in image processing that adjusts the hue, brightness, and contrast of one image to match the histogram features of another. In texture processing, this technique is primarily used to unify the textures of close-up images captured by a drone with those of satellite imagery. Since the two images originate from different sources and may differ in lighting, color temperature, etc., histogram matching is necessary for radiometric consistency adjustment. This includes performing histogram analysis on the close-up textures from both the drone and satellite images, and adjusting the histogram of the drone texture to match that of the satellite image texture. Figure 1 This process ensures that the two textures have similar color distributions. Histogram matching eliminates texture differences caused by variations in lighting or shooting angles, resulting in a more visually unified and realistic 3D model texture.

[0076] Example 2, based on the same inventive concept as the multi-data combination method for constructing a 3D model of a construction site in the aforementioned examples, such as... Figure 2 As shown in the embodiment of this application, a three-dimensional model construction system for construction sites combining multiple data is provided. The system includes: The satellite image retrieval module 10 is used to retrieve satellite images at a preset resolution based on the minimum outer rectangle boundary of the construction area as the geographic coordinate system parameter, thereby obtaining a full-area stereo image pair dataset. The elevation model generation module 20 is used to perform radiometric correction on the full-area stereo image pair dataset and then generate a regional surface model and a regional elevation model by fusing the stereo image pairs. The defect detection module 30 is used to perform raster-level modeling defect detection on the regional surface model and the regional elevation model to locate distributed blind spot blocks. The path fitting module 40 is used to perform obstacle avoidance based on the distributed blind spot blocks. The system includes: path fitting to generate an obstacle avoidance optimized flight trajectory; a point cloud generation module 50 to drive the UAV to perform multi-angle surround scanning of the distributed blind zone block along the obstacle avoidance optimized flight trajectory, and to directionally collect dense point clouds of the distributed blind zone; wherein the UAV is equipped with a binocular stereo camera and an IMU; a blind zone model construction module 60 to construct a distributed blind zone model based on the dense point cloud of the distributed blind zone; and a micro-overlay module 70 to use the regional surface model and regional elevation model as a spatial reference frame to perform micro-overlay of the distributed blind zone model and output a three-dimensional model of the construction site.

[0077] Furthermore, the defect detection module 30 is used to perform the following operation steps: After spatially aligning the regional surface model and the regional elevation model, the regional surface model and the regional elevation model are segmented in parallel based on a preset grid scale to obtain a surface model grid array and an elevation model grid array; a quantified definition array is output by performing texture gradient evaluation on the surface model grid array; a quantified complexity array is output by performing Gaussian mutation detection on the elevation model grid array; a double-index fusion weighting is performed on the quantified definition array and the quantified complexity array that are spatially aligned to obtain a modeling defect probability array; and a low-precision grid of the modeling defect probability array is merged based on a watershed algorithm to locate the distributed blind area block.

[0078] Further, the defect detection module 30 is configured to perform the following operation steps: The distributed low-precision grid satisfying the preset defect threshold is screened by traversing the modeling defect probability array; the distributed low-precision grid is taken as an initial depression mark with the elevation gradient field of the elevation model grid array as an input source, a terrain gradient driven watershed flooding simulation is performed to generate a continuous watershed partition constrained by a watershed boundary; and a non-continuous terrain segmentation exclusion is performed by traversing the continuous watershed partition with a terrain elevation gradient change threshold to locate the distributed blind area block.

[0079] Further, the defect detection module 30 is configured to perform the following operation steps: The updated construction division is obtained by associating the construction progress data; the added structure area of the updated construction division is analyzed to perform an iterative incremental update of the distributed blind area block.

[0080] Further, the path fitting module 40 is configured to perform the following operation steps: The center point of the distributed blind area block is taken as a waypoint to fit a flat flight trajectory; the flat flight trajectory is projected to the regional elevation model to identify a plurality of elevation obstacles with a preset safe vertical distance; and the flat flight trajectory is three-dimensionally and elevationally smoothed according to the plurality of elevation obstacles to output the obstacle-avoiding optimized flight trajectory.

[0081] Further, the point cloud generation module 50 is configured to perform the following operation steps: A tightly coupled SLAM system constructed based on the binocular stereo camera and the IMU is used to perform multi-view spiral wrap-around observation along a flight trajectory to collect a stereo image sequence and IMU raw measurement data; after real-time adjacent frame matching of the stereo image sequence is performed through point-line fusion feature extraction, depth map reconstruction processing is performed in combination with an optimized pose output by pre-integration of the IMU raw measurement data to output a time-series depth map; and the time-series depth map is fused through TSDF to generate the distributed blind area dense point cloud.

[0082] Further, the blind area model construction module 60 is configured to perform the following steps: Based on geometric continuity, the point cloud normal vector mutation boundary is used to divide the distributed blind area dense point cloud into a plurality of building component point clouds; a plurality of groups of FPFH feature descriptors of the plurality of building component point clouds are extracted; the plurality of groups of FPFH feature descriptors are aggregated based on structure feature comparison to obtain N reference FPFH descriptors; the N reused components are extracted from the historical incremental model library based on the N reference FPFH descriptors; after the N reference FPFH descriptor corresponding component point clouds are stripped from the plurality of building component point clouds, Poisson reconstruction is performed to obtain an initial blind area model; the N reused components are rigidly transformed and aligned to the initial blind area model according to the plurality of groups of FPFH feature descriptors to perform boundary fusion processing, thereby generating the distributed blind area model.

[0083] Further, the micro-superposition module 70 is configured to perform the following steps: After the geographic coordinate system of the regional surface model and the regional elevation model is registered by the ground control point, the grid block of the distributed blind area block is deleted to obtain a terrain reservation reference framework; after the terrain reservation reference framework is embedded with the distributed blind area model, joint fusion is performed to obtain the construction site three-dimensional model.

[0084] Further, histogram matching is used to perform radiation unification of the unmanned aerial vehicle close-range texture and the satellite image texture of the construction site three-dimensional model.

[0085] Through the foregoing detailed description of the construction site three-dimensional model construction method of the multi-data combination, those skilled in the art can clearly understand the multi-data combined construction site three-dimensional model construction system in the embodiment. Since the system corresponds to the method disclosed in the embodiment, the system is described relatively simply, and the relevant part is described in the method part.

[0086] The above is only a preferred embodiment of the present application, and does not limit the present application in any form. Although the present application has been disclosed as the preferred embodiment, it is not intended to limit the present application. Any person skilled in the art can make some changes or modifications to the above disclosed technical content to obtain equivalent embodiments with equivalent changes, without departing from the technical solution of the present application. Any modification, equivalent change and modification of the above embodiment based on the technical essence of the present application are still within the scope of the technical solution of the present application.

Claims

1. A method for constructing a three-dimensional model of a construction site by multi-data integration, characterized in that, The method comprises: According to the minimum outer rectangular boundary of the construction area, the geographical coordinate system parameters are taken as the geographical coordinate system parameters, the preset resolution satellite image is called, and a full-area stereo image pair data set is obtained; After radiation correction is performed on the full-area stereo image pair data set, a regional surface model and a regional elevation model are generated by fusing the stereo image pair; By performing grid-level modeling defect detection on the regional surface model and the regional elevation model, a distributed blind area block is located; Based on the distributed blind area block, an obstacle avoidance path is fitted, and an obstacle avoidance optimized flight trajectory is generated; A UAV is driven to perform multi-angle surrounding scanning on the distributed blind area block along the obstacle avoidance optimized flight trajectory, and a distributed blind area dense point cloud is collected in a directional manner, wherein the UAV is equipped with a binocular stereo camera and an IMU; Based on the distributed blind area dense point cloud, a distributed blind area model is constructed; The regional surface model and the regional elevation model are taken as a spatial reference framework, and the distributed blind area model is microscopically superimposed, and a construction site three-dimensional model is output.

2. The multi-data combined construction site three-dimensional model construction method of claim 1, wherein, By performing grid-level modeling defect detection on the regional surface model and the regional elevation model, a distributed blind area block is located, and the method comprises: After the regional surface model and the regional elevation model are spatially aligned, the regional surface model and the regional elevation model are segmented in parallel based on a preset grid scale, and a surface model grid array and an elevation model grid array are obtained; By performing texture gradient evaluation on the surface model grid array, a quantized definition array is output; By performing Gaussian mutation detection on the elevation model grid array, a quantized complexity array is output; The quantized definition array and the quantized complexity array are spatially aligned and double-index fusion weighted, and a modeling defect probability array is obtained; Based on a watershed algorithm, low-precision grids of the modeling defect probability array are merged based on terrain continuity, and the distributed blind area block is located.

3. The multi-data combined construction site 3D model construction method of claim 2, wherein, Based on a watershed algorithm, low-precision grids of the modeling defect probability array are merged based on terrain continuity, and the distributed blind area block is located, comprising: The modeling defect probability array is traversed, and distributed low-precision grids meeting a preset defect threshold are screened; Taking an elevation gradient field of the elevation model grid array as an input source and taking the distributed low-precision grids as initial depression marks, a terrain gradient-driven watershed submergence simulation is performed, and a continuous watershed partition constrained by a watershed boundary is generated; The continuous watershed partition is traversed by using a terrain elevation gradient change threshold, non-continuous terrain segmentation exclusion is performed, and the distributed blind area block is located.

4. The multi-data combined construction site 3D model construction method of claim 3, wherein, Further comprising: Associated construction progress data is obtained to update a construction division; An added structure area of the updated construction division is analyzed, and an iterative incremental update of the distributed blind area block is performed.

5. The multi-data combined construction site 3D model construction method of claim 1, wherein, Based on the distributed blind area block, an obstacle avoidance path is fitted, and an obstacle avoidance optimized flight trajectory is generated, comprising: Taking a center point of the distributed blind area block as a waypoint, a planar flight trajectory is fitted; The planar flight trajectory is projected to the regional elevation model, spatial obstacles of a preset safe vertical distance are identified, and a plurality of elevation obstacles are located; Performing three-dimensional elevation smoothing correction of the planar flight trajectory according to the plurality of elevation obstacles, and outputting the obstacle-avoiding optimized flight trajectory.

6. The multi-data combined construction site 3D model construction method of claim 1, wherein, Driving the unmanned aerial vehicle to perform multi-angle ring-around scanning on the distributed blind area block along the obstacle-avoiding optimized flight trajectory, and directionally collecting a distributed blind area dense point cloud, including: Based on the tightly coupled SLAM system constructed by the binocular stereo camera and the IMU, performing multi-view spiral ring-around observation along the flight trajectory, and collecting a stereo image sequence and IMU raw measurement data; After real-time adjacent frame matching of the stereo image sequence is performed through point-line fusion feature extraction, combining the optimized pose output by pre-integration of the IMU raw measurement data to perform depth map reconstruction processing, and outputting a time-series depth map; Fusing the time-series depth map through TSDF to generate the distributed blind area dense point cloud.

7. The multi-data combined construction site 3D model construction method of claim 1, wherein, Based on the distributed blind area dense point cloud, a distributed blind area model is constructed, and the method includes: Based on geometric continuity, the distributed blind area dense point cloud is segmented into a plurality of building component point clouds at a point cloud normal vector mutation boundary; A plurality of groups of FPFH feature descriptors of the plurality of building component point clouds are extracted; Based on structural feature comparison, the plurality of groups of FPFH feature descriptors are aggregated to obtain N reference FPFH descriptors; The N reused components are extracted from a historical incremental model library by comparison using the N reference FPFH descriptors; After the N reference FPFH descriptor corresponding component point clouds are stripped from the plurality of building component point clouds, Poisson reconstruction is performed to obtain an initial blind area model; According to the plurality of groups of FPFH feature descriptors, the N reused components are rigidly transformed and aligned to the initial blind area model for boundary fusion processing to generate the distributed blind area model.

8. The multi-data combined construction site 3D model construction method of claim 1, wherein, The regional surface model and the regional elevation model are taken as a spatial reference framework to perform micro superposition of the distributed blind area model, and a construction site three-dimensional model is output, including: After the geographic coordinate system of the regional surface model and the regional elevation model is registered through ground control points, grid block deletion of the distributed blind area block is performed to obtain a terrain reservation reference framework; After the distributed blind area model is embedded in the terrain reservation reference framework, joint fusion is performed to obtain the construction site three-dimensional model.

9. The multi-data combined construction site 3D model construction method of claim 1, wherein, Histogram matching is used to perform radiation unification of the unmanned aerial vehicle close-range texture and the satellite image texture of the construction site three-dimensional model.

10. A multi-data combined construction site three-dimensional model construction system characterized by, A system for implementing the multi-data combined construction site three-dimensional model construction method of any one of claims 1-9, the system comprising: A satellite image retrieval module configured to take the minimum bounding rectangle boundary of a construction area as a geographic coordinate system parameter to perform preset resolution satellite image retrieval, and obtain a full-area stereo image pair data set; An elevation model generation module configured to perform radiation correction on the full-area stereo image pair data set, and generate a regional surface model and a regional elevation model through fusion stereo image pair generation; A defect detection module configured to locate a distributed blind area block by performing grid-level modeling defect detection on the regional surface model and the regional elevation model; A path fitting module configured to perform obstacle-avoiding path fitting based on the distributed blind area block to generate an obstacle-avoiding optimized flight trajectory; and A flight control module configured to drive the unmanned aerial vehicle to perform multi-angle ring-around scanning on the distributed blind area block along the obstacle-avoiding optimized flight trajectory, and directionally collect a distributed blind area dense point cloud. The point cloud generation module is used to drive the UAV to perform multi-angle surround scanning of the distributed blind zone block along the obstacle avoidance optimized flight trajectory, and to collect dense point clouds of the distributed blind zone in a directional manner. The UAV is equipped with a binocular stereo camera and an IMU. The blind zone model construction module is used to construct a distributed blind zone model based on the distributed blind zone dense point cloud. The micro-overlay module is used to use the regional surface model and regional elevation model as a spatial reference frame to perform micro-overlay of the distributed blind zone model and output a three-dimensional model of the construction site.

Images