Earthwork oblique photography collaborative modeling method for high, steep and narrow terrain

Abstract

The invention discloses a high steep narrow terrain earthwork oblique photography collaborative modeling method, and belongs to the technical field of engineering surveying and mapping and three-dimensional digital modeling. The method comprises the following steps: designing an imitated flight route based on a digital elevation model of a high steep narrow terrain, controlling an unmanned aerial vehicle carrying a laser radar and multiple lenses to synchronously collect data, arranging hierarchical image control points, de-noising a laser radar point cloud to generate a digital elevation surface, and processing an oblique photography image to generate a three-dimensional grid model. And fusing the two types of models to correct the elevation and re-project the texture, generating a collaborative live-action three-dimensional model, and finally calculating the earth-rock volume and planning the construction machinery path based on the model. The method can be used for earthwork surveying and mapping, metering and construction scheduling of high, steep and narrow terrains, and collaborative improvement of modeling geometric accuracy and texture authenticity is achieved.

Description

Technical Field

[0001] This invention relates to the field of engineering surveying and 3D digital modeling technology. More specifically, this invention relates to a collaborative modeling method for earthwork oblique photogrammetry in steep and narrow terrain. Background Technology

[0002] In the construction of pumped storage power stations, mines, mountain road slopes, and mine spoil heaps, there is often a need for surveying and modeling steep and narrow terrain. Such terrain is characterized by V-shaped gullies, extremely steep slopes (exceeding 60° in some areas), steep mountains, and narrow construction areas, posing a severe challenge to engineering surveying.

[0003] With the development of surveying and mapping technology, UAV LiDAR and oblique photogrammetry have been gradually applied to such scenarios. However, single technologies still have obvious limitations: UAVs often adopt a fixed altitude flight mode, which makes it difficult to adapt to the undulations of steep terrain, easily leading to image distortion and missing point clouds in steep cliff areas, resulting in insufficient data acquisition completeness; although LiDAR can efficiently acquire high-density terrain point clouds and penetrate vegetation canopy to directly detect the ground surface, single-view scanning will result in incomplete information on the sides of ground objects and blind spots; oblique photogrammetry can acquire rich surface texture information and build realistic 3D models, but in steep and heavily occluded areas, due to image matching errors, the generated geometric models will have accuracy problems such as distortion and stretching, and cannot meet the dual requirements of high-precision geometric measurement and complete texture expression.

[0004] Even when attempting to combine the two technologies, the challenge of collaborative processing of multi-source data remains: LiDAR and oblique photogrammetry equipment are mostly acquired independently, lacking a unified time synchronization mechanism, resulting in significant spatial offset errors in the data and a substantial reduction in subsequent fusion accuracy; point cloud denoising often employs fixed-parameter filtering algorithms, which are prone to misclassifying non-ground points in steep cliff areas as valid data, and may mistakenly delete ground points in gentle areas, exhibiting poor filtering adaptability; while LiDAR point clouds have high geometric accuracy, they lack texture information, and while oblique photogrammetry 3D mesh models have rich textures, their elevation accuracy is insufficient. The fusion of the two is often a simple superposition, failing to achieve a synergistic improvement in geometric accuracy and texture realism.

[0005] How to effectively integrate the high-precision geometric acquisition capabilities of lidar with the rich texture acquisition capabilities of oblique photogrammetry to generate accurate and realistic 3D models that can be directly applied to precise earthwork volume calculation and intelligent construction management is of great significance for improving the efficiency, safety and digitalization level of engineering construction in complex terrain. Summary of the Invention

[0006] This invention provides a collaborative modeling method for earthwork oblique photography in steep and narrow terrain, which can be used for earthwork engineering surveying, measurement and construction scheduling in steep and narrow terrain, and achieves a synergistic improvement in modeling geometric accuracy and texture realism.

[0007] To achieve these objectives and other advantages according to the present invention, a method for collaborative modeling of earthwork oblique photography in steep and narrow terrain is provided, comprising the following steps: S1, based on the digital elevation model of steep and narrow terrain, designs a terrain-following flight path, controls the UAV equipped with lidar and at least two tilting camera lenses to fly along the path, and ensures that lidar scanning and multi-angle tilting photography are executed synchronously through a time synchronization device; S2. Set up image control points in the survey area. Specifically, along the flight path, set up a first-level control point network with a density of 4-6 photo baselines in the flight direction and 2 flight paths in the lateral direction. For areas with abrupt changes in terrain or key construction areas, add control points at the four corners of the regional network as second-level densification points. S3. After performing coordinate transformation on the lidar point cloud data, an iterative local plane fitting filtering method is used to remove noise from the point cloud on steep cliffs. This includes constructing a neighborhood for the points in the point cloud, fitting a local plane based on the neighborhood points, calculating the distance from the point to the plane, and determining and removing noise points based on distance statistical features. S4, perform ground point classification and interpolation on the denoised point cloud to generate a digital elevation surface that meets the preset accuracy requirements; S5 performs aerial triangulation and dense matching on the oblique photogrammetric images to generate an oblique 3D mesh model. S6, the digital elevation surface is fused with the inclined 3D mesh model, and a geometric constraint texture mapping operation is performed. Based on the elevation value of the digital elevation surface, the vertex elevation of the inclined 3D mesh model is corrected so that the vertex elevation is aligned with the elevation value of the digital elevation surface at the same position on the plane. The texture is then reprojected according to the corrected mesh geometry to generate a collaborative real-scene 3D model of steep and narrow terrain. S7. Based on the collaborative real-scene 3D model, the earthwork volume is calculated by comparing the model elevation changes at different times, and the construction machinery path is planned by planning and displaying the trajectory in the model.

[0008] Preferably, step S1 includes: S1a: Based on the pre-acquired digital elevation model of steep and narrow terrain, a terrain-following flight path is designed, with the forward overlap range set to 75-85% and the lateral overlap range to 65-75%. The photographic baseline length B is calculated according to the formula B = GSD×H / f, and the shooting interval is set accordingly. Here, GSD is the preset ground sampling distance, f is the camera focal length, and H is the real-time flight altitude relative to the terrain below. S1b controls a drone equipped with a lidar and a multi-lens tilting camera to fly along a flight path, simultaneously collecting point cloud and image data, and calibrating and compensating for the spatial offset between the lidar and the camera, keeping the spatial offset within a preset range; synchronous acquisition is achieved through a time synchronization device, which outputs synchronization signals to the lidar and the camera and records the timestamps of their respective data acquisitions; point cloud data and image data with timestamp differences less than a preset value are associated and stored.

[0009] Preferably, the digital elevation model for steep and narrow terrain is obtained through the following steps: a. Use UAV lidar to conduct initial aerial surveys of the survey area and collect sparse point cloud data; b. After denoising and classifying the sparse point cloud, an initial digital elevation model is generated by interpolation; c. Add measured control points to the steep areas of the initial digital elevation model to correct the elevation values ​​of the initial digital elevation model, and obtain the final digital elevation model for route design.

[0010] Preferably, in step S3, the iterative local plane fitting filtering method includes: S3a: Construct an adaptive neighborhood for each point P in the point cloud, calculate the local planar roughness σ between point P and its K initial neighbors, and if σ > the preset steepness threshold σ t If it is determined to be a steep region, the first neighborhood search radius R is used. s Otherwise, use the second neighborhood search radius R. d , where R s <R d The value of K ranges from 15 to 20; among them, neighboring points are retrieved in the point cloud database through a spatial indexing algorithm. When the number of points retrieved within the initial retrieval radius is insufficient, the retrieval radius is gradually expanded until the number requirement is met. S3b: For the set of points in the adaptive neighborhood, fit the local plane using the principal component analysis algorithm and calculate the distance D from point P to the plane; S3c, calculate the mean μ and standard deviation δ of the distances from all points in the neighborhood to the plane; S3d, if D>μ+n×δ, then mark point P as a noise point, where the multiplier n used in steep regions is greater than the multiplier n used in gentle regions, and the value of n ranges from 2 to 3. S3e, iteratively execute steps S3a-S3d, and gradually shrink the neighborhood search radius or adjust the multiplier factor n until the number of marked noise points tends to stabilize.

[0011] Preferably, in step S6, the correction of the vertex elevations of the inclined 3D mesh model is achieved through the following steps: S6a, Project the triangular facet T of the inclined three-dimensional mesh model onto the digital elevation surface S to obtain the elevation point set {Z} of S within the coverage area of ​​the triangular facet T. S6b, for the vertex V of the triangular facet T, the constrained elevation z at its planar coordinates (x, y) is obtained by interpolation based on {Z}. dem ; S6c, calculate the original elevation z of vertex V. mesh With z dem Elevation residual Δz = z dem -z mesh If the absolute value of Δz is greater than the preset tolerance threshold τ, then a correction is performed, where τ is 0.05m. S6d, during correction, a smoothing weight factor λ is determined based on the geometric position of vertex V within its triangular facet T. The value of λ increases as vertex V approaches the center of triangular facet T and decreases as vertex V approaches the edge of triangular facet T. The new elevation z of vertex V is then adjusted. set Set to: z set = z mesh +λ×Δz.

[0012] Preferably, in step S4, meeting the preset accuracy requirement means that the elevation error of the digital elevation surface is ≤0.10m; The collaborative real-scene 3D model generated in step S6 has a planar accuracy error of ≤0.05m and an elevation accuracy error of ≤0.10m.

[0013] Preferably, the reprojection of texture in step S6 specifically includes: recalculating the normal vector of each triangle based on the corrected vertex coordinates of the triangles; and performing orthorectification and texture fusion on the original tilted image based on the new normal vectors and camera parameters.

[0014] Preferably, in step S7, the earthwork volume calculation based on the collaborative real-scene 3D model specifically includes: S7a, obtain collaborative real-scene 3D models of the same area before and after construction; S7b places the collaborative reality 3D models of the two phases in the same coordinate system and performs fine registration based on the stable and unchanged ground features in the collaborative reality 3D models. The planar error after registration is ≤0.03m. S7c, Construct the same regular triangular mesh on the surface of the collaborative reality 3D model. Set the side length of the triangular mesh unit to 5-10 times the ground sampling distance GSD. Using the triangular mesh as the basic unit, calculate the elevation difference ΔH of each unit on the collaborative reality 3D model in the two phases. S7d, based on the ΔH and area of ​​all grid cells, the volume of the part with ΔH > 0 is summarized as the fill volume, and the volume of the part with ΔH < 0 is summarized as the cut volume.

[0015] Preferably, in step S7, the construction machinery path planning specifically involves: importing the collaborative real-scene 3D model into the construction machinery control system; the system plans the transportation path in the collaborative real-scene 3D model according to preset safety rules, including a path slope of ≤15% and a distance of ≥5m from dangerous areas; and overlaying the real-time position and attitude information of the construction machinery onto the collaborative real-scene 3D model for construction monitoring and scheduling.

[0016] Preferably, the step of planning the transportation route also includes route smoothing: smoothing the initially planned route using spline curves, wherein the spline curves satisfy the curvature change rate ≤ 0.1 / m, and exporting the smoothed route coordinates in a format compatible with the construction machinery control system.

[0017] The present invention has at least the following beneficial effects: This invention combines the advantages of LiDAR and oblique photogrammetry. Through geometrically constrained texture mapping, the high-precision geometric information of LiDAR can effectively correct the geometric distortion of the oblique model, while the rich texture of the oblique model is perfectly preserved. Finally, a collaborative model with both high-precision geometric shape and realistic surface texture is generated, solving the problems of insufficient modeling accuracy or missing texture in traditional single technology.

[0018] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0019] Figure 1 This is a ground-following flight diagram of one technical solution of the present invention; Figure 2 Setting the takeoff altitude and flight scheme parameters for an unmanned aerial vehicle (UAV) according to a technical solution of the present invention; Figure 3 The present invention provides a technical solution for the variable altitude distribution of a UAV (a) and variable altitude flight path planning (b). Figure 4 This is a schematic diagram of the encrypted deployment of image control points according to one technical solution of the present invention; Figure 5 This is a laser point cloud model data diagram of one technical solution of the present invention; Figure 6 This is a schematic diagram of the encrypted deployment of image control points according to one technical solution of the present invention; Figure 7 This is a schematic diagram illustrating the change in excavation filling volume according to one technical solution of the present invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0021] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0022] It should be noted that, unless otherwise specified, the experimental methods described in the following embodiments are conventional methods, and the reagents and materials mentioned are commercially available. In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "setting" should be interpreted broadly. For example, they can refer to fixed connection or setting, detachable connection or setting, or integral connection or setting. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The terms "lateral," "longitudinal," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this invention and simplifying the description. They do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0023] Steep and narrow terrain refers to terrain areas where gullies are V-shaped, local slopes can reach over 60°, mountains are steep, and the horizontal width of the construction area is less than 50m. It is commonly found in pumped storage power station reservoir areas, mine valleys, and other similar locations. It is characterized by dense vegetation, poor visibility, and dramatic terrain undulations. This invention provides a collaborative modeling method for earthwork oblique photography in steep and narrow terrain, including the following steps: S1. Obtain the pre-surveyed digital elevation model (DEM) of the survey area, using ASTGTM2 elevation data. The plane coordinate system is the geographic coordinate system. Generate a back-projection coordinate file (TfW) using ArcGIS software to ensure that the model coordinates are consistent with the actual terrain. Design a terrain-following flight path based on the digital elevation model of steep and narrow terrain, such as... Figure 1 The shown ground-following flight diagram Figure 2 The drone takeoff altitude and flight plan parameter settings are shown, as well as Figure 3 The diagram shows the distribution of drone elevation points (a) and the elevation point flight path planning (b). In practice, use the intelligent aerial survey assistant software to import DEM data, select the elevation flight path function, choose 10 control points to fully cover the surrounding mountains of the survey area, set the area flight range, set the flight altitude to the highest point of the survey area +760m, and the takeoff altitude relative to the DEM model to 150m (to ensure flight safety). Set the terrain simulation detail to 10m to ensure the flight path conforms to the terrain undulations. Control the drone equipped with a LiDAR and at least two tilted camera lenses to fly along this flight path, such as one at +45° and one at -45° tilt. The system covers different angles and is equipped with a time synchronization device. The time synchronization device is connected to the lidar and camera lens respectively. Before flight, the synchronization device is confirmed to output a unified trigger signal through ground testing. The time synchronization device ensures that the lidar scanning and multi-angle oblique photography are executed synchronously, eliminating the time difference between lidar and camera lens acquisition, and ensuring that the point cloud and image data at the same spatial location are aligned in time and space. The UAV travels at a constant speed along the planned route, reads DEM elevation data in real time, and dynamically adjusts the UAV's flight altitude to keep the sensor at a fixed distance from the ground, thus solving the problems of uneven overlap and image distortion in steep terrain. S2. Image control points are established within the survey area. These points serve as the constraint benchmarks for aerial triangulation densification, and their density and location directly affect modeling accuracy. The image control points are divided into a first-level control point network with accurate geodetic coordinates and a second-level densification network. Based on the reservoir survey area's topographical data and flight requirements, the entire survey area is divided into three sub-areas: JM01, JM02, and JM03. Figure 4 The encryption combination shown is as follows: along the simulated flight path, a first-level control point network is set up with a density of 4-6 photographic baselines at the forward interval and 2 flight paths at the lateral interval. In one example, white cement markers (with stainless steel markers inlaid on the top) are used, with a total of 77 horizontal and vertical points set up, of which 72 are actually effective (5 are invalid due to obstruction). The HiCORS integrated service system (network RTK) is used to measure the plane coordinates, and the elevation is fitted by a geoid refinement model. For steep cliffs with a slope >60°, gully bends, dam shoulders, and other terrain abrupt changes or key construction areas, additional control points at the four corners of the regional network are set up as second-level encryption points. Each local regional network needs to cover terrain abrupt changes or key construction areas to ensure that it can be clearly identified by the oblique image. In one example, each regional network is a 20m×20m square, with a total of 12 encryption points set up. Portable red and white striped marker boards are used to facilitate image recognition. S3. Convert the polar coordinate (distance r, vertical angle α, horizontal angle θ) data collected by the lidar to geodetic coordinates (X, Y, Z) using the Cartesian coordinate transformation formula. The transformation relationship is as follows: After coordinate transformation of the lidar point cloud data, based on the different noise characteristics of point cloud in steep cliff areas and flat areas, a neighborhood dynamic construction and local plane fitting method is used for each point to more accurately distinguish effective points from noise points. For point cloud noise in steep cliff areas, an iterative local plane fitting filtering method is employed for denoising, avoiding the problems of traditional fixed-parameter filtering that mistakenly deletes effective points in steep cliffs and leaves residual noise in flat areas. This ensures the purity of the denoised point cloud and the authenticity of the terrain. The iterative local plane fitting filtering method includes constructing a neighborhood for each point in the point cloud, fitting a local plane based on the neighborhood points, calculating the distance from the point to the plane, and determining and removing noise points based on distance statistical characteristics. Specifically, firstly, a neighborhood is constructed for each point in the point cloud: based on the terrain characteristics of the point's location... (Steep or gentle) Determine the neighborhood range to ensure it accurately reflects the local terrain trend; secondly, fit a local plane: based on the neighborhood point set of each point, use a plane fitting algorithm (such as least squares) to construct a local plane that can represent the overall terrain trend within the neighborhood; then calculate the distance and determine noise: calculate the distance from the target point to its neighborhood fitting plane, and statistically analyze the distances from all points in the neighborhood to the plane (such as mean and dispersion), identifying points whose distances exceed a reasonable range of statistical characteristics as noise points; finally, perform iterative optimization: repeat the steps of constructing the neighborhood → fitting the plane → determining noise, gradually optimizing the neighborhood range and noise determination criteria until the number of noise points tends to stabilize, and finally remove the determined noise points to obtain the denoised point cloud data; S4. Ground point classification and interpolation are performed on the denoised point cloud. Specifically, ground point classification distinguishes ground points from non-ground points (such as vegetation and building remnants) based on point cloud features (elevation, density, reflection intensity, etc.). Only ground points are retained for subsequent terrain representation. Based on the classified ground points, interpolation is then performed using interpolation algorithms (moving surface interpolation, Kriging interpolation, etc.) to construct a continuous elevation surface, i.e., generating a Digital Elevation Surface (DES) that meets the preset accuracy requirements. Figure 5 The image shown is a laser point cloud model data diagram; S5 involves aerial triangulation and dense matching of oblique photogrammetric images to generate an oblique 3D mesh model. Specifically, the oblique photogrammetric images are first preprocessed (format conversion, distortion correction) to eliminate errors caused by lens distortion and equipment attitude deviations during image capture. Then, aerial triangulation is performed: the preprocessed oblique images and the control point data from S2 are imported into photogrammetric software (Pix4Dmapper) for aerial triangulation. By matching image feature points, the exterior orientation elements (position and attitude) of each image are solved, constructing a spatial triangulation network for the survey area, providing accuracy constraints for subsequent modeling. Finally, dense matching and mesh modeling are performed: based on the aerial triangulation results, dense matching is performed on the oblique images, setting the matching density to high to generate a high-density image matching point cloud. Using this point cloud as a basis, a Poisson reconstruction algorithm is used to construct a triangular mesh structure, and the texture information of the oblique images is mapped onto the mesh surface, ultimately generating an oblique 3D mesh model containing geometric shapes and surface textures. Figure 6 A schematic diagram of the tilted 3D mesh model is shown; S6. The digital elevation surface and the inclined 3D mesh model are fused and imported into ArcGIS Pro software. Spatial registration is performed based on three fixed landmarks (such as rock outcrops) at the boundary of the survey area. After registration, the planar error is ≤0.03m. Geometric constraint texture mapping is performed. Through elevation correction, the high-precision geometric constraint of the inclined model by the digital elevation surface is used to solve its geometric accuracy problem. Through texture reprojection, the texture advantage of the inclined model is preserved. Specifically, the elevation value of the digital elevation surface is used as a reference to extract the elevation value at the plane coordinates of each vertex of the inclined 3D mesh model. The vertex elevation of the inclined 3D mesh model is corrected, that is, the elevation value is compared with the original elevation value of the vertex of the mesh model. The vertex elevations with deviations exceeding the allowable range are corrected so that the vertex elevations are aligned with the elevation values ​​of the digital elevation surface at the same position in the plane. The texture is reprojected according to the corrected mesh geometry. That is, according to the corrected mesh geometry, the projection relationship of the inclined image texture on the mesh surface is recalculated. The texture image is attached to the mesh model according to the new geometric relationship to generate a collaborative real-scene 3D model of steep and narrow terrain. S7. Based on the collaborative real-scene 3D model, the earthwork volume is calculated by comparing the model elevation changes at different times. By comparing the elevation changes of the two models at the same plane position, the elevation difference of each spatial unit is calculated. Then, combined with the unit area, the fill volume (area with increased elevation) and cut volume (area with decreased elevation) of the entire region are obtained. In this model, based on the terrain features reflected by the model (such as slope and obstacle position), the trajectory is planned and displayed to plan the path of construction machinery.

[0024] The above technical solution combines the advantages of LiDAR and oblique photography. Through the core operation of geometric constraint texture mapping, the high-precision geometric information of LiDAR can effectively correct the geometric distortion of the oblique model. At the same time, the rich texture of the oblique model is perfectly preserved. The resulting collaborative real-scene 3D model has both high geometric accuracy and high visual realism, providing a unified and reliable 3D data foundation for accurate earthwork calculation and visualized construction management in steep and narrow terrain, and greatly improving work efficiency and reliability of results.

[0025] In another technical solution, step S1 includes: S1a, based on a pre-acquired digital elevation model of steep and narrow terrain, sets the relative altitude between the UAV and the ground, designs a terrain-following flight path, and ensures full coverage of key areas such as steep cliffs and gullies by aligning the flight path perpendicular to the main slope direction. The forward overlap (the overlap ratio of two adjacent oblique photographs within the same flight path) is set to 75-85%, and the lateral overlap (the overlap ratio of oblique photographs between two adjacent flight paths) is set to 65-75%. Specifically, for steep areas with a slope > 45°, the forward overlap is set to 80%-85%, and the lateral overlap to 70%-75%, to avoid missing image feature points due to terrain occlusion. For gentle areas with a slope ≤ 45°, the forward overlap is set to 75%-80%, and the lateral overlap to 65%-70%. The photographic baseline length B is calculated using the formula B = GSD × H / f, and the photographing interval is then set. Based on the UAV's flight speed v and the forward overlap, the photographing interval t = The calculation is performed using B×(1-heading overlap) / v to ensure that the overlap between adjacent images meets the standard. Here, GSD is the ground sampling distance preset according to the engineering accuracy requirements, f is the camera focal length, and H is the real-time flight altitude relative to the terrain below. S1b controls a drone equipped with a LiDAR and a multi-lens tilting camera to fly along a flight path, simultaneously collecting point cloud and image data, and calibrating and compensating for the spatial offset between the LiDAR and the camera. During data collection, the coordinates of the LiDAR point cloud and the camera image are automatically corrected, and the spatial offset is controlled within a preset range. Synchronous acquisition is achieved through a time synchronization device, which outputs a synchronization signal to the LiDAR and the camera and records the timestamp of their respective data collection. Point cloud data and image data with a timestamp difference less than a preset value are associated and stored.

[0026] In the above technical solution, the UAV adjusts its flight altitude in real time through the terrain elevation information provided by the digital elevation model, always maintaining a constant relative altitude with the ground. By using the range of overlap between heading and lateral directions, formula-quantified photographic baseline and shooting interval, and strict requirements for sensor spatial and temporal synchronization, the data acquisition process is refined and standardized, ensuring the quality and spatial consistency of image data under different terrain undulations, as well as the precise correspondence between laser point clouds and image pixels.

[0027] In another technical solution, a digital elevation model for steep and narrow terrain is obtained through the following steps: a. Use UAV lidar to conduct initial aerial survey of the survey area, set flight parameters according to the area of ​​the survey area, plan a grid-like flight path, and ensure point cloud coverage in steep cliff areas by taking sparse point cloud data that only reflects the key undulation features of the terrain. b. Statistical filtering is used to denoise the sparse point cloud. Based on the characteristics of point cloud reflection intensity and elevation distribution, ground points and non-ground points are distinguished. Ground points have stable and continuous reflection intensity, while non-ground points (such as low vegetation and rock fragments) have fluctuating and discrete reflection intensity. Ground points are automatically extracted using the Random Sample Consensus (RANSAC) algorithm, and then manually corrected to ensure the purity of ground points. Point cloud classification is then performed. An interpolation algorithm is used to calculate the elevation of unknown points using the elevation information of known points, transforming discrete points into a continuous digital elevation model. Specifically, a moving surface interpolation method is used to construct the initial digital elevation model. The interpolation process fits the local terrain using a quadratic surface equation. The formula for the elevation of the point to be interpolated is: Z = a0 + a1X + a2Y + a3X 2 +a4XY+a5Y 2 Where Z is the elevation of the point to be interpolated, X and Y are the plane coordinates of the point to be interpolated, and a0-a5 are the surface coefficients. An equation system is established using n ground points (n≥6) within the window of the point to be interpolated. The coefficients a0-a5 are solved by the least squares method and substituted into the coordinates to calculate Z. The size of the interpolation window is set (20m×20m for flat areas and 10m×10m for steep areas to adapt to the complexity of the terrain). For each point to be interpolated, the ground points within its window are retrieved. The local terrain is fitted by the quadratic surface equation to calculate the elevation of the point to be interpolated. Interpolation is then performed to generate the initial digital elevation model. c. In the steep areas (slope > 45°) of the initial digital elevation model, supplement the measured control points with a grid density of 50m × 50m. Use GNSSRTK (real-time dynamic positioning) technology to measure the elevation of the control points and correct the elevation values ​​of the initial digital elevation model. Specifically, for each measured control point, extract the elevation value Z1 at its corresponding plane coordinates (X, Y) in the initial digital elevation model and compare it with the measured elevation value Z2 to calculate the elevation deviation ΔZ = Z2 - Z1. Using the control point as the center, use the weighted average method to correct the initial model elevation within a 50m range around it: the corrected elevation Z = Z1 + ΔZ × w, where w is the weight, which is inversely proportional to the distance from the point to be corrected to the control point. The closer the distance, the larger w is, ranging from 0 to 1. After all control points are corrected, smooth the elevation transition between adjacent correction areas to obtain the final digital elevation model for flight route design.

[0028] In the above technical solution, the general terrain is quickly acquired by using UAV lidar, and then corrected by on-site measurements of key areas, thereby autonomously generating a reliable terrain model that meets the requirements of terrain-following flight planning, ensuring flight safety and the integrity of data collection.

[0029] In another technical solution, step S3, the iterative local plane fitting filtering method includes: S3a: Import the coordinate-transformed point cloud into the point cloud processing software. Use the KD-Tree spatial indexing algorithm to construct a point cloud database. Set an initial search radius and retrieve its K initial neighboring points. Construct an adaptive neighborhood for each point P in the point cloud. Calculate the local planar roughness σ between point P and its K initial neighboring points. Specifically, fit a temporary local plane using the least squares method, calculate the distance from each neighboring point to this plane, and then calculate the dispersion of these distances using the standard deviation formula to obtain the local planar roughness σ of point P. Steep regions are prone to occlusion, resulting in sparse and discrete point cloud distribution, while flat regions have a relatively uniform point cloud distribution. Determine the region type using the local planar roughness σ and dynamically adjust the neighborhood search radius. If σ > the preset steepness determination threshold σ... t If it is determined to be a steep region, the first neighborhood search radius R is used. s Otherwise, use the second neighborhood search radius R. d , where R s <R d The value of K ranges from 15 to 20. Neighboring points are retrieved from the point cloud database using a spatial indexing algorithm. When the number of points retrieved within the initial retrieval radius is insufficient, the retrieval radius is gradually expanded in steps of 0.1m until the requirement of K neighboring points is met. S3b, for the point set within the adaptive neighborhood, a local plane is fitted using principal component analysis (PCA), and the distance D from point P to this plane is calculated. Specifically, the covariance matrix of the point set is calculated, and three eigenvalues ​​(λ1≥λ2≥λ3) and their corresponding eigenvectors are obtained through eigenvalue decomposition. The eigenvector corresponding to the smallest eigenvalue λ3 is the normal vector of the local plane. Combining this with the coordinates of any point within the neighborhood, the local plane equation Ax + By + Cz + D = 0 is established. The three-dimensional coordinates (X, Y, Cz) of point P are then used to determine the local plane. p ,Y p Z p Substitute the values ​​into the local plane equation and calculate the distance D from point P to the plane using the point-to-plane distance formula; S3c: Traverse all points in the adaptive neighborhood, calculate the distance from each point to the S3b fitting plane, and obtain a distance set. Based on this set, calculate the mean μ of the distance from all points in the neighborhood to the plane, and calculate the standard deviation δ of the distance from all points in the neighborhood to the plane using the sample standard deviation formula. S3d, the distance from effective shape points in the neighborhood to the fitted plane should follow a normal distribution. The mean μ represents the central tendency of the distance, and the standard deviation δ represents the degree of dispersion. The multiplier n is set according to the region type. If D>μ+n×δ, it means that point P deviates far from the local terrain trend and is marked as a noise point. Otherwise, it is judged as an effective shape point. This can accurately separate discrete points (noise points) that deviate from the normal distribution. The multiplier n used in steep regions is greater than that used in flat regions. The value of n is 2-3. In steep regions, due to the complex terrain and high noise dispersion, n is 3. In flat regions, the noise distribution is relatively concentrated, so n is 2. S3e, iteratively execute steps S3a-S3d, record the number of marked noise points. In the second iteration, shrink the neighborhood search radius and adjust the n value of the smooth region from 2 to 2.5. In each subsequent iteration, fine-tune the parameters according to this rule, gradually shrink the neighborhood search radius or adjust the multiplier factor n until the number of marked noise points tends to stabilize.

[0030] In the above technical solution, the adaptability problem of traditional fixed parameter filtering is solved by adaptive neighborhood and iterative optimization: the adaptive neighborhood is dynamically adjusted based on the terrain roughness to adapt to the regional differences of high and steep terrain, avoiding the denoising bias caused by fixed neighborhood; principal component analysis fits the plane to ensure the accuracy of local terrain benchmark and improve the reliability of distance calculation; the differential factor and iterative optimization make the noise judgment gradually converge to the optimal, which can not only completely remove the hanging points and vegetation interference points of high and steep cliffs, but also completely retain the effective shape points.

[0031] In another technical solution, step S6, correcting the vertex elevation of the inclined three-dimensional mesh model, is achieved through the following steps: S6a, Spatial alignment preprocessing: For each triangular facet T of the inclined 3D mesh model, orthophoto projection is used to project the triangular facet T of the inclined 3D mesh model onto the digital elevation surface S, forming a projection area consistent with the plane range of T. Through spatial clipping, the elevation point set {Z} of S within the coverage area of ​​the triangular facet T is obtained. Each point contains plane coordinates (X, Y) and corresponding elevation value. The density of the point set needs to meet the interpolation requirements. Obviously abnormal elevation values ​​in {Z} are removed. S6b, for the vertex V of the triangular facet T, the constrained elevation z at its planar coordinates (x, y) is obtained by interpolation based on {Z}. dem Specifically, its planar coordinates (x, y) are extracted. The planar coordinates (x, y) of vertex V may not exactly fall on the elevation point of S. In the elevation point set {Z}, the four nearest elevation points around V are selected using the K-nearest neighbor search algorithm. The coordinates and elevation values ​​of the four points are recorded. The constrained elevation z corresponding to vertex V is calculated using bilinear interpolation. dem A local rectangular grid is constructed using four neighboring points as vertices. Linear interpolation is used to first calculate two intermediate elevation values ​​of V in the X direction of the rectangle. Then, interpolation is performed in the Y direction based on these intermediate values ​​to finally obtain the precise elevation z at the plane coordinates (x, y) of V. dem ; S6c, calculate the original elevation z of vertex V. mesh With z dem Elevation residual Δz = z dem -z mesh If the absolute value of Δz is greater than the preset tolerance threshold τ, it means that the original elevation of vertex V deviates too much from the datum (digital elevation surface S) and elevation correction needs to be performed. In this case, the correction is performed, where τ is 0.05m. In S6d, during the correction, the smoothing weight factor λ of vertex V within its triangular patch T is first determined. The value of λ is determined based on the relative position of vertex V within the triangular patch T. The closer vertex V is to the center region of the triangular patch T, the larger the value of λ is, so as to apply a stronger elevation correction. The closer vertex V is to any edge of the triangular patch T, the smaller the value of λ is, so as to maintain the original geometric smoothness of the mesh near the vertex, thereby achieving a smooth transition of the overall model correction.

[0032] In one specific implementation, the centroid coordinate method is used to determine λ, and the specific steps are as follows: Calculate the centroid coordinates: For a triangular facet T, assume its three vertices are A, B, and C. For the vertex V to be corrected, calculate its centroid coordinates (α, β, γ) in triangle ABC. The centroid coordinates satisfy α + β + γ = 1, where α, β, and γ represent the proximity of vertex V to edges BC, CA, and AB, respectively. When V is located on edge BC, α = 0; when V is closer to vertex A, the value of α increases. Determine the weighting factor: Define the weighting factor λ as the minimum value among the three centroid coordinate components, i.e., λ = min(α, β, γ). This definition satisfies the above design principle: when vertex V is located at the center of the triangular facet, α ≈ β ≈ γ ≈ 1 / 3, and λ takes a relatively large value (approximately 1 / 3, which can be adjusted to the desired range through subsequent normalization); when vertex V is infinitely close to a certain edge (e.g., edge BC), the corresponding centroid coordinate component (α) approaches 0, and λ also approaches 0. Perform elevation correction: After obtaining the weighting factor λ, adjust the new elevation z of vertex V. set Set to: z set = z mesh +λ×Δz. Where, z mesh Let z be the original elevation of the vertex, and Δz be the elevation residual calculated in step S6c. set Update to the tilted 3D mesh model, replacing the original elevation z. mesh Complete the elevation correction for a single vertex, traverse all vertices that need correction, and repeat the above steps until the elevation correction of all triangle facet vertices is completed.

[0033] In the above technical solution, the geometry of the tilted model is refined and progressively adjusted by using triangular patch projection, interpolation to obtain constrained elevations, calculation of residuals, and weight-based smoothing correction. The elevation correction amount of each vertex is smoothly weighted according to its position, which effectively avoids geometric cracks or distortions that may occur at the mesh edge due to direct elevation replacement, ensuring the overall smoothness and geometric consistency of the fused model. The corrected vertex elevation is accurately aligned with the digital elevation surface, and the geometric accuracy of the tilted model is significantly improved.

[0034] In another technical solution, in step S4, meeting the preset accuracy requirement means that the elevation error of the digital elevation surface is ≤0.10m, thus meeting the preset accuracy requirement. In step S6, the generated collaborative real-scene 3D model has a planar accuracy error ≤0.05m, meeting the accuracy requirements for planar positioning in construction machinery path planning, and an elevation accuracy error ≤0.10m, adapting to the elevation accuracy requirements for earthwork volume calculation in steep terrain (avoiding calculation errors exceeding the allowable range of the project). If the planar error is ≤0.05m and the elevation error is ≤0.10m, the model accuracy is deemed to meet the standard. Clearly defined accuracy provides a quantitative evaluation standard for the modeling results. The elevation and planar error thresholds adapt to the earthwork engineering needs of steep and narrow terrain, eliminating engineering risks caused by unclear accuracy and improving the reliability and engineering applicability of the results.

[0035] In another technical solution, the reprojection of texture in step S6 specifically includes: recalculating the normal vector of each triangle based on the corrected vertex coordinates of the triangles; extracting the coordinates of the three vertices (x1, y1, z1), (x2, y2, z2), and (x3, y3, z3) of the corrected triangles; calculating the two edge vectors (x2-x1, y2-y1, z2-z1) and (x3-x1, y3-y1, z3-z1); obtaining the normal vector through vector cross product; normalizing the normal vector to make its magnitude 1; establishing a texture projection relationship based on the new normal vector and camera parameters; performing orthorectification on the original tilted image to eliminate the stretching deformation caused by image tilt and terrain undulation; if the triangle is covered by multiple images, texture fusion is performed using distance weighting (the closer the image is to the center of the triangle, the greater the weight) to avoid stitching marks.

[0036] In the above technical solution, recalculating the normal vector can ensure that the texture projection direction fits the corrected model. Orthorectification restores the image imaging geometry through camera parameters and eliminates deformation. Texture fusion solves the connection problem of multiple image coverage through weight allocation, ensuring texture continuity and solving the problems of texture misalignment and stretching after vertex elevation correction.

[0037] In another technical solution, step S7, calculating the earthwork volume based on the collaborative real-scene 3D model, specifically includes: S7a, obtain collaborative real-scene 3D models of the same area before and after construction; S7b places the collaborative reality 3D models of the two phases in the same coordinate system and performs fine registration based on stable, unchanging ground features in the collaborative reality 3D models (prioritizing features that do not shift during construction, such as concrete benchmark piles, large rock outcrops, and permanent walls). The Iterative Closest Point (ICP) algorithm is used for fine registration, with 1000 iterations and a convergence threshold of 0.001m. By minimizing the spatial distance error between feature point pairs, the position of the model after construction is optimized. After registration, the planar distance (ΔX, ΔY) of all feature point pairs is calculated. The planar error after registration is ≤0.03m. S7c, Construct the same regular triangular mesh on the surface of the collaborative reality 3D model. Set the side length of the triangular mesh unit to 5-10 times the ground sampling distance GSD. Using the triangular mesh as the basic unit, calculate the elevation difference ΔH of each unit on the collaborative reality 3D model in the two phases. S7d: For each grid cell, extract its average elevation H1 before construction and average elevation H2 after construction (obtained by the arithmetic mean of the elevations of all vertices within the cell). Calculate the elevation difference using ΔH = H2 - H1. Based on the ΔH and area of ​​all grid cells, summarize the volume of the portion with ΔH > 0 as the fill volume and the volume of the portion with ΔH < 0 as the cut volume. Iterate through all grid cells, summing the total fill volume and the total cut volume to generate an earthwork calculation report. Mark key information such as the calculation range, grid parameters, and error range. Figure 7 The diagram shows the variation in excavation filling volume.

[0038] In the above technical solution, precise registration ensures that the two phases of the model space are consistent, avoiding measurement errors caused by coordinate system deviation. The regular triangular mesh divides the units with fixed side lengths, so that the calculation units of the two phases of the model are completely corresponding. The elevation difference of each unit can accurately reflect the local terrain changes. By classifying and summarizing positive and negative ΔH, it is possible to clearly distinguish between fill (increase in terrain elevation) and cut (decrease in terrain elevation).

[0039] In another technical solution, step S7, the construction machinery path planning, specifically involves: importing the collaborative real-scene 3D model into the construction machinery control system; removing redundant textures (such as vegetation details) unrelated to construction from the model; retaining the geometric information of terrain undulations and dangerous areas (steep cliffs, deep pits, etc.); and planning the transportation path in the collaborative real-scene 3D model according to preset safety rules, including a path slope ≤15% and a distance ≥5m from dangerous areas. The construction machinery is equipped with a GNSS positioning module and attitude sensors to collect real-time data on the machinery's plane coordinates, elevation, and fuselage pitch angle, which is then transmitted to the control system via a wireless link. The real-time position and attitude information of the construction machinery is overlaid and displayed on the collaborative real-scene 3D model for construction monitoring and scheduling. The model baseline ensures the path conforms to the actual terrain, safety rules mitigate the risk of machinery failure and collisions, and the real-time overlay display makes scheduling intuitive and efficient, facilitating timely adjustments to the work plan.

[0040] In another technical solution, the planning of the transportation route also includes route smoothing: from the planned preliminary transportation route, the coordinates of key nodes (including X, Y, and Z three-dimensional information) are extracted at 1-2m intervals to form a node sequence, ensuring coverage of the path's start point, end point, and turning points. The preliminary planned path is smoothed using spline curves, and the node sequence is fitted using a cubic spline curve algorithm. A piecewise polynomial curve is constructed using software to ensure the continuity of the first and second derivatives of adjacent curve segments at the nodes, avoiding abrupt changes in the broken line. During the fitting process, the curve curvature is calculated in real time. The spline curve satisfies a curvature change rate ≤ 0.1 / m. A verification point is taken every 0.5m along the smoothed curve, and the curvature difference between adjacent verification points is calculated and divided by the distance between the two points to obtain the curvature change rate. If there are segments with out-of-tolerance, the curve is refitted. After confirming that the requirements are met, the smoothed path coordinates are exported in a format compatible with the construction machinery control system. Cubic spline curves, by fitting the node sequence with a piecewise polynomial, can achieve curvature continuity while ensuring the path direction, avoiding abrupt changes in the initial path, and smoothing the curvature to avoid the risk of mechanical rollover in steep terrain, thus improving driving stability.

[0041] The number of devices and processing scale described herein are for the purpose of simplifying the description of the invention. Applications, modifications, and variations of the invention will be readily apparent to those skilled in the art.

[0042] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A collaborative modeling method for earthwork oblique photography in steep and narrow terrain, characterized in that, Includes the following steps: S1, based on the digital elevation model of steep and narrow terrain, designs a terrain-following flight path, controls the UAV equipped with lidar and at least two tilting camera lenses to fly along the path, and ensures that lidar scanning and multi-angle tilting photography are executed synchronously through a time synchronization device; S2. Set up image control points in the survey area. Specifically, along the flight path, set up a first-level control point network with a density of 4-6 photo baselines in the flight direction and 2 flight paths in the lateral direction. For areas with abrupt changes in terrain or key construction areas, add control points at the four corners of the regional network as second-level densification points. S3. After performing coordinate transformation on the lidar point cloud data, an iterative local plane fitting filtering method is used to remove noise from the point cloud on steep cliffs. This includes constructing a neighborhood for the points in the point cloud, fitting a local plane based on the neighborhood points, calculating the distance from the point to the plane, and determining and removing noise points based on distance statistical features. S4, perform ground point classification and interpolation on the denoised point cloud to generate a digital elevation surface that meets the preset accuracy requirements; S5 performs aerial triangulation and dense matching on the oblique photogrammetric images to generate an oblique 3D mesh model. S6, the digital elevation surface is fused with the inclined 3D mesh model, and a geometric constraint texture mapping operation is performed. Based on the elevation value of the digital elevation surface, the vertex elevation of the inclined 3D mesh model is corrected so that the vertex elevation is aligned with the elevation value of the digital elevation surface at the same position on the plane. The texture is then reprojected according to the corrected mesh geometry to generate a collaborative real-scene 3D model of steep and narrow terrain. S7. Based on the collaborative real-scene 3D model, the earthwork volume is calculated by comparing the model elevation changes at different times, and the construction machinery path is planned by planning and displaying the trajectory in the model.

2. The method for collaborative modeling of earthwork oblique photography in steep and narrow terrain according to claim 1, characterized in that, Step S1 includes: S1a: Based on the pre-acquired digital elevation model of steep and narrow terrain, a terrain-following flight path is designed, with the forward overlap range set to 75-85% and the lateral overlap range to 65-75%. The photographic baseline length B is calculated according to the formula B = GSD×H / f, and the shooting interval is set accordingly. Here, GSD is the preset ground sampling distance, f is the camera focal length, and H is the real-time flight altitude relative to the terrain below. S1b controls a drone equipped with a lidar and a multi-lens tilting camera to fly along a flight path, simultaneously collecting point cloud and image data, and calibrating and compensating for the spatial offset between the lidar and the camera, keeping the spatial offset within a preset range; synchronous acquisition is achieved through a time synchronization device, which outputs synchronization signals to the lidar and the camera and records the timestamps of their respective data acquisitions; point cloud data and image data with timestamp differences less than a preset value are associated and stored.

3. The method for collaborative modeling of earthwork oblique photography in steep and narrow terrain according to claim 1 or 2, characterized in that, A digital elevation model for steep and narrow terrain is obtained through the following steps: a. Use UAV lidar to conduct initial aerial surveys of the survey area and collect sparse point cloud data; b. After denoising and classifying the sparse point cloud, an initial digital elevation model is generated by interpolation; c. Add measured control points to the steep areas of the initial digital elevation model to correct the elevation values ​​of the initial digital elevation model, and obtain the final digital elevation model for route design.

4. The method for collaborative modeling of earthwork oblique photography in steep and narrow terrain according to claim 1, characterized in that, In step S3, the iterative local plane fitting filtering method includes: S3a: Construct an adaptive neighborhood for each point P in the point cloud, calculate the local planar roughness σ between point P and its K initial neighbors, and if σ > the preset steepness threshold σ t If it is determined to be a steep region, the first neighborhood search radius R is used. s Otherwise, use the second neighborhood search radius R. d , where R s <R d The value of K ranges from 15 to 20; among them, neighboring points are retrieved in the point cloud database through a spatial indexing algorithm. When the number of points retrieved within the initial retrieval radius is insufficient, the retrieval radius is gradually expanded until the number requirement is met. S3b: For the set of points in the adaptive neighborhood, fit the local plane using the principal component analysis algorithm and calculate the distance D from point P to the plane; S3c, calculate the mean μ and standard deviation δ of the distances from all points in the neighborhood to the plane; S3d, if D > μ+n×δ, then point P is marked as a noise point, where the multiplier n used in steep regions is greater than the multiplier n used in gentle regions, and the value of n ranges from 2 to 3. S3e, iteratively execute steps S3a-S3d, and gradually shrink the neighborhood search radius or adjust the multiplier factor n until the number of marked noise points tends to stabilize.

5. The method for collaborative modeling of earthwork oblique photography in steep and narrow terrain according to claim 1, characterized in that, In step S6, the correction of the vertex elevations of the inclined 3D mesh model is achieved through the following steps: S6a, Project the triangular facet T of the inclined three-dimensional mesh model onto the digital elevation surface S to obtain the elevation point set {Z} of S within the coverage area of ​​the triangular facet T. S6b, for the vertex V of the triangular facet T, the constrained elevation z at its planar coordinates (x, y) is obtained by interpolation based on {Z}. dem ; S6c, calculate the original elevation z of vertex V. mesh With z dem Elevation residual Δz = z dem -z mesh If the absolute value of Δz is greater than the preset tolerance threshold τ, then a correction is performed, where τ is 0.05m. S6d, during correction, a smoothing weight factor λ is determined based on the geometric position of vertex V within its triangular facet T. The value of λ increases as vertex V approaches the center of triangular facet T and decreases as vertex V approaches the edge of triangular facet T. The new elevation z of vertex V is then adjusted. set Set to: z set = z mesh +λ×Δz.

6. The method for collaborative modeling of earthwork oblique photography in steep and narrow terrain according to claim 1, characterized in that, In step S4, meeting the preset accuracy requirement means that the elevation error of the digital elevation surface is ≤0.10m; The collaborative real-scene 3D model generated in step S6 has a planar accuracy error of ≤0.05m and an elevation accuracy error of ≤0.10m.

7. The method for collaborative modeling of earthwork oblique photography in steep and narrow terrain according to claim 1, characterized in that, The reprojection of texture in step S6 specifically includes: recalculating the normal vector of each triangle based on the corrected vertex coordinates of the triangles; and performing orthorectification and texture fusion on the original tilted image based on the new normal vectors and camera parameters.

8. The method for collaborative modeling of earthwork oblique photography in steep and narrow terrain according to claim 1, characterized in that, Step S7, specifically the calculation of earthwork volume based on the collaborative reality 3D model, includes: S7a, obtain collaborative real-scene 3D models of the same area before and after construction; S7b places the collaborative reality 3D models of the two phases in the same coordinate system and performs fine registration based on the stable and unchanged ground features in the collaborative reality 3D models. The planar error after registration is ≤0.03m. S7c, Construct the same regular triangular mesh on the surface of the collaborative reality 3D model. Set the side length of the triangular mesh unit to 5-10 times the ground sampling distance GSD. Using the triangular mesh as the basic unit, calculate the elevation difference ΔH of each unit on the collaborative reality 3D model in the two phases. S7d, based on the ΔH and area of ​​all grid cells, the volume of the part with ΔH > 0 is summarized as the fill volume, and the volume of the part with ΔH < 0 is summarized as the cut volume.

9. The method for collaborative modeling of earthwork oblique photography in steep and narrow terrain according to claim 1, characterized in that, In step S7, the construction machinery path planning specifically involves: importing the collaborative real-scene 3D model into the construction machinery control system; the system plans the transportation path in the collaborative real-scene 3D model according to preset safety rules, including a path slope of ≤15% and a distance of ≥5m from dangerous areas; and overlaying the real-time position and attitude information of the construction machinery onto the collaborative real-scene 3D model for construction monitoring and scheduling.

10. The method for collaborative modeling of earthwork oblique photography in steep and narrow terrain according to claim 9, characterized in that, The steps of planning the transportation route also include route smoothing: the initially planned route is smoothed using spline curves, wherein the spline curves satisfy the curvature change rate ≤ 0.1 / m, and the smoothed route coordinates are exported in a format compatible with the construction machinery control system.

Images