A river and lake shoreline unmanned aerial vehicle inspection three-dimensional visualization operation and maintenance management system

Abstract

A kind of river and lake bank unmanned aerial vehicle inspection three-dimensional visualization operation and maintenance management system, comprising: data preprocessing module, memory scheduling module and rendering pipeline intervention module;Data preprocessing module is used to receive the original data collected by unmanned aerial vehicle inspection equipment, generates multistage detail grid data file, solidifies as structured data block;Memory scheduling module is used to collect the space position of virtual camera in real time, determines index range, and the structured data block in index range is read to system video memory;Rendering pipeline intervention module includes fragment shading controller and vertex shading controller;Fragment shading controller is used to generate nonlinear mapping depth value and stretch mapping depth value to complete video memory overwrite;Vertex shading controller is used to output the vertex coordinates after deformation.The system provided by the present application adapts to the long and narrow topological structure of river system, and the data loading efficiency is high, which guarantees the picture stability of three-dimensional visualization.

Description

Technical Field

[0001] This invention relates to the field of drone inspection and image data processing technology, specifically to a three-dimensional visualization operation and maintenance management system for drone inspection of river and lake shorelines. Background Technology

[0002] With the development of UAV oblique photogrammetry technology, 3D visualization operation and maintenance management systems have been widely used in routine inspections of river and lake shorelines. When processing massive amounts of 3D mesh data collected by UAVs, in order to ensure the rendering frame rate of the front-end operation and maintenance terminal, the system usually needs to simplify and build the original model at multiple levels of detail in the background, and combine spatial scheduling algorithms to execute data loading and rendering pipeline output within the view area during the roaming process.

[0003] However, existing 3D visualization systems have significant limitations in their underlying data processing and rendering scheduling mechanisms when dealing with the unique, elongated geographical environment of river and lake shorelines. During the mesh model simplification stage, conventional 3D engines typically perform isotropic vertex fusion or edge collapsing operations based on a Cartesian coordinate system. Because this process does not consider the direction and distribution of river systems or the topological specificities of the land-water boundary, this indiscriminate simplification often leads to the over-collapse of key geometric features along the shoreline, causing the land-water boundary line in the inspection image to become blurred and distorted as the model level decreases.

[0004] Regarding viewport memory scheduling, existing systems primarily rely on collision detection between 3D bounding boxes (such as octrees or BVH) and the camera's view frustum to determine the loading status of data blocks. Since the trajectory of a UAV patrolling a river exhibits highly linear characteristics, maintaining high-frequency 3D bounding box intersection operations during high-speed roaming can cause unnecessary CPU performance degradation. Furthermore, existing scheduling algorithms lack a look-ahead prediction mechanism based on motion direction; when the viewpoint rapidly advances along the river, delays in data block loading can easily lead to memory overflow risks and severe screen stuttering.

[0005] Furthermore, in the graphics rendering stage, due to the flat terrain and extremely large visual span of the river surface, traditional rendering pipelines are often limited by insufficient depth buffer precision when viewed from a distance, resulting in severe depth conflicts in the water surface polygons. Simultaneously, existing mechanisms typically employ hard vertex coordinate replacement when triggering layer switching between multi-level detailed mesh files, lacking cross-layer geometric smoothing intervention. This coarse mesh replacement causes abrupt geometric jumps and lateral visual jitter in the image at the moment of model change, reducing the accuracy of inspectors' assessment of subtle potential hazards.

[0006] Therefore, it is necessary to design a technical solution for UAV inspection 3D visualization operation and maintenance management that is adapted to the narrow topology of river systems, has high data loading efficiency and good image stability, in order to solve the problem that conventional grid simplification algorithms in the existing technology deviate from the narrow direction of river systems, resulting in excessive folding and distortion of key water-land boundary geometric features. Summary of the Invention

[0007] To address the technical problem in the existing technology where conventional grid simplification algorithms deviate from the narrow and elongated direction of river systems, resulting in excessive folding and distortion of key geometric features at the water-land interface, this invention provides a three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines, including: a data preprocessing module, a memory scheduling module, and a rendering pipeline intervention module; The data preprocessing module is used to receive raw data collected by UAV inspection equipment, generate multi-level detailed mesh data files, and solidify the multi-level detailed mesh data files into structured data blocks; the raw data includes raw UAV oblique photogrammetry 3D mesh vertex data and river channel centerline discrete vector data in Cartesian coordinate system. The memory scheduling module is used to acquire the spatial position of the virtual camera in the Cartesian coordinate system in real time, perform projection calculations, and obtain the longitudinal projection arc length coordinates corresponding to the current frame. Derivation of the look-ahead velocity scalar Calculate the one-dimensional sliding scheduling interval, determine the index range, and read the data within the index range into the system video memory; The rendering pipeline intervention module includes a fragment shader controller and a vertex shader controller. After reading the data into the system video memory, the fragment shader controller calculates the fragment-level lateral normal distance component based on the structured data block. With fragment-level elevation components Generate nonlinear mapping depth numerical values With stretch mapping depth value The process of overwriting video memory is completed; the vertex shader controller extracts the position deviation vector based on structured data blocks. Combined with the longitudinal projection arc length coordinates corresponding to the current frame And the transition mixing factor derived from the look-ahead velocity scalar Generate a dimension-reduced deformation deviation vector Output the coordinates of the vertices after deformation. .

[0008] Furthermore, the generation of multi-level detailed mesh data files includes: The original data is converted into a reconstructed coordinate vector and written into the underlying memory structure of each original 3D mesh vertex; the reconstructed coordinate vector includes a longitudinal arc length component S, a transverse normal distance component L, and an elevation component H; The reconstructed coordinate vector is obtained from the underlying memory structure, and a dynamic penalty weight is constructed based on the lateral normal distance component L. Calculate the cost of folding anisotropic edges. ; Based on the anisotropic edge folding cost Iterative calculations are performed to generate multi-level detailed mesh data files, which are then solidified into structured data blocks.

[0009] Furthermore, converting the original data into a reconstructed coordinate vector includes: Using discrete vector data of the river channel centerline as control points, a cubic B-spline curve function is constructed to generate a smooth centerline. Based on the smooth central axis, the orthogonal projection point is calculated using the Newton-Raphson iterative algorithm; the orthogonal projection point is the curve point with the smallest Euclidean distance from the vertex data of the original UAV oblique photogrammetry 3D mesh. Based on the Euclidean distance from the vertex data of the original UAV oblique photogrammetry 3D mesh to the orthogonal projection point, calculate the longitudinal arc length component. Horizontal normal distance component and elevation components The original 3D mesh vertices are merged to generate a reconstructed coordinate vector, which is then written as an additional vertex attribute parameter into the underlying memory structure of each original 3D mesh vertex.

[0010] Furthermore, the execution of iterative operations includes: A min-heap data structure is constructed, and the target topological edge at the top of the min-heap data structure is extracted to trigger the vertex fusion instruction. When the total number of remaining faces is reduced to a preset polygon number threshold, the edge folding iteration operation is paused, and the retained mesh topology connection data and vertex coordinate data are extracted, serialized and encapsulated to generate a multi-level detailed mesh data file, which is then solidified into a structured data block. The polygon number threshold is the limit of the number of discrete polygons contained in the hierarchical file.

[0011] Furthermore, the derivation of the look-ahead velocity scalar include: Real-time acquisition of the spatial position of the virtual camera in a Cartesian coordinate system, application of Kalman filtering, and output of smoothed camera coordinate data; Determine the orthogonal projection point of the current frame and calculate the longitudinal projection arc length coordinates corresponding to the current frame. The orthogonal projection point of the current frame is the curve point closest to the current camera coordinate data; Store several longitudinal projection arc length coordinates into a circular buffer, and extract the historical longitudinal projection arc length coordinates from the circular buffer. Vertical projection arc length coordinates corresponding to the current frame Perform difference quotient calculation and output the look-ahead velocity scalar. .

[0012] Furthermore, reading the data within the index range into the system video memory includes: Based on the aforementioned look-ahead velocity scalar, a dynamic bias term is constructed, tilting the preloaded center of gravity towards the future movement direction of the camera, and the starting arc length coordinates of the one-dimensional sliding scheduling interval are calculated. coordinates of the endpoint arc length This yields a one-dimensional sliding scheduling interval. The one-dimensional sliding scheduling interval is a data loading window that dynamically slides along the central axis of the river channel. Based on the starting arc length coordinates coordinates of the endpoint arc length By calling the floor function, the starting file index of the target data block to be loaded is output. With Termination File Index This forms an index range; An asynchronous block transfer instruction is dispatched to read the data within the index range of the structured data block into the system video memory.

[0013] Furthermore, the calculation of the fragment-level lateral normal distance component... With fragment-level elevation components include: The lateral normal distance component and elevation component are extracted from the structured data block and configured as interpolation variables; When entering the rasterization stage, for any element generated inside the patch, extract the homogeneous coordinates of the vertices in the clip space and calculate the centroid projection weights. Based on the centroid projection weight, the perspective correction interpolation algorithm is invoked to calculate the fragment-level lateral normal distance component. With fragment-level elevation components .

[0014] Furthermore, the generation of nonlinear mapping depth numerical values With stretch mapping depth value The process includes: Based on the Gaussian distribution function model, the lateral normal distance component of the structure at the fragment level is described. Build depth offset; Get screen space depth value Combined with the depth offset, a nonlinear mapped depth value is generated. ; Based on the nonlinear mapping depth value Combined with the fragment-level elevation components The polynomial mapping formula is used to generate the stretch mapping depth value. ; Combine the nonlinear mapping depth values and stretch mapping depth values The overwrite instruction is issued to replace the fragment output variable and complete the video memory overwrite.

[0015] Furthermore, the derivation of the transition mixing factor include: Based on structured data blocks, the difference between the current coordinates of the mesh vertices and the target coordinates in the lower level of detail model is extracted as a position deviation vector. ; Extract the hierarchical boundary coordinates of the structured data block Set the basic transition width constant. Combined with forward speed scalar Calculate the dynamic transition band width parameter; Get the vertical projection arc length coordinates of the current frame. with hierarchical boundary coordinates The difference is used to derive the transition mixing factor by combining the dynamic transition band width parameter. .

[0016] Furthermore, the generation of the dimension-reduced deformation deviation vector include: Based on the position deviation vector Extract the unit tangent vector The position deviation vector is subjected to orthogonal decomposition and component filtering operations to output a dimension-reduced deformation deviation vector. The operation expression is: .

[0017] The output deformed vertex coordinates include: Obtain the basic 3D coordinates of the current processing mesh vertex Combined with transition mixing factor and dimension reduction deformation deviation vector Calculate the vertex coordinates after deformation The calculation expression is: .

[0018] The beneficial effects of this invention are: 1. This invention converts Cartesian coordinate data into reconstructed coordinate vectors based on the river channel centerline through a data preprocessing module, and uses the extracted lateral normal distance component to calculate dynamic penalty weights to intervene in the edge folding process of the detailed mesh. This allows the mesh simplification algorithm to directly adapt to the narrow topology of the river system. While generating multi-level detailed mesh data files on demand to reduce the data volume, it preserves the geometric features of the water-land transition zone.

[0019] 2. This invention utilizes a memory scheduling module to calculate the longitudinal projection arc length coordinates of the camera and derive the look-ahead velocity scalar. This constructs a one-dimensional sliding scheduling interval to achieve linear pre-reading of structured data blocks. It reduces the traditional three-dimensional bounding box intersection operation to the index comparison of a one-dimensional arc length interval, improving the data loading efficiency along the river channel and avoiding the problems of video memory overflow and screen stuttering when the terminal accesses large-scale inspection data.

[0020] 3. This invention uses a rendering pipeline intervention module to generate nonlinear mapping depth values ​​using normal distance and elevation components during the fragment shading stage to complete the memory overwriting. In the vertex shading stage, it combines look-ahead velocity to derive a blending factor to output a dimension reduction deformation deviation vector, thus resolving the depth conflict phenomenon caused by insufficient accuracy of distant water surface polygons. At the same time, it filters out the lateral jitter when switching model levels, ensuring the stability of the 3D visualization. Attached Figure Description

[0021] Figure 1 This is an architecture diagram of the three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines provided by the present invention; Figure 2 This is a dynamic penalty weight distribution curve at the water-land interface provided in a specific application embodiment of the present invention; Figure 3 This is a comparison chart of system rendering frame rate stability in a specific application embodiment provided by the present invention. Detailed Implementation

[0022] The technical solution of the present invention is further described below, but the scope of protection is not limited to what is described.

[0023] This invention provides a 3D visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines, such as... Figure 1 As shown, it includes: a data preprocessing module P100, a memory scheduling module P200, and a rendering pipeline intervention module P300; The data preprocessing module P100 is used to receive raw data collected by the UAV inspection equipment and generate a multi-level detailed mesh data file; and to solidify the multi-level detailed mesh data file into structured data blocks; the raw data includes the original UAV oblique photogrammetry 3D mesh vertex data and the discrete vector data of the river channel centerline in the Cartesian coordinate system. The generation of multi-level detailed mesh data files includes: 1. Convert the original data into a reconstructed coordinate vector; the reconstructed coordinate vector includes a longitudinal arc length component S, a lateral normal distance component L, and an elevation component H; wherein, the reconstructed coordinate vector is a curve coordinate data structure; converting the rectangular coordinate system data into a curve coordinate data structure is to extract the spatial topological features of the longitudinal extension and lateral cross-section of the river channel.

[0024] Converting the original data into a reconstructed coordinate vector includes: 1) Using the discrete vector data of the river channel centerline as control points, construct a cubic B-spline curve function to generate a smooth centerline; When generating the smoothed central axis, the data preprocessing module P100 reads the discrete vector data of the river channel central axis and performs filtering and smoothing calculations to obtain the smoothed central axis. The conventionally input river channel central axis data is presented as a discrete vector point sequence. Directly constructing tangent and normal vectors based on this discrete vector point sequence can lead to abrupt changes in the normal direction and undefined phenomena at the junctions of the broken lines. To ensure the continuity of the second derivative and eliminate the undefined normal vector, in this embodiment, the data preprocessing module uses the chord length parameterization method to estimate the node vectors and constructs a cubic B-spline curve function using the input discrete vector point sequence as control points to generate the smoothed central axis. For the cubic B-spline fitting calculation of the discrete vector point sequence, those skilled in the art can use existing computational geometry libraries to implement it, and it will not be elaborated here.

[0025] 2) Based on the smooth central axis, the orthogonal projection point is calculated using the Newton-Raphson iterative algorithm; the orthogonal projection point is the curve point with the smallest Euclidean distance from the vertex data of the original UAV oblique photogrammetry 3D mesh. In practice, a curve coordinate system is established based on the smoothed centerline. The three orthogonal dimensions of the coordinate system are defined as the longitudinal arc length direction, the transverse normal direction, and the elevation direction. The longitudinal arc length direction coincides with the tangent vector direction of the smoothed centerline; the transverse normal direction is parallel to the horizontal plane and perpendicular to the aforementioned tangent vector direction; and the elevation direction is consistent with the vertical upward direction of the world coordinate system.

[0026] 3) Calculate the longitudinal arc length component based on the Euclidean distance from the vertex data of the original UAV oblique photogrammetry 3D mesh to the orthogonal projection point. Horizontal normal distance component and elevation components The original 3D mesh vertices are merged to generate a reconstructed coordinate vector, which is then written as an additional vertex attribute parameter into the underlying memory structure of each original 3D mesh vertex.

[0027] In practice, the system first iterates through and reads the vertex data of the original UAV oblique photogrammetry 3D mesh in Cartesian coordinates, and then performs a geometric spatial mapping transformation calculation from mesh vertices to curvilinear coordinates. For any original 3D mesh vertex, the data preprocessing module P100 searches for the nearest point in the discrete vector point sequence as the initial value for iteration. It then calls Newton's iteration algorithm to find the curve point with the smallest Euclidean distance from the selected original 3D mesh vertex on the smooth central axis, and configures the solved curve point as an orthogonal projection point. After completing the projection operation, the spatial mapping transformation formula is used to calculate the reconstructed coordinate components of the vertex. The transformation formula is expressed as follows: (1) (2) (3) In the formula, It represents the longitudinal arc length component; These are the curve parameters corresponding to the orthogonal projection points, and the parameter values ​​range from... Normalized interval; This is the integral term of the arc length accumulated along the curve from the starting point of the smooth central axis to the orthogonal projection point; To smooth the central axis; These are curve parameters, and their value range is also [missing information]. Normalized interval; This represents the lateral normal distance component; This represents the spatial coordinate vectors of the original 3D mesh vertices in a Cartesian coordinate system. is the spatial coordinate vector of the orthogonal projection point in the rectangular coordinate system; This is the Euclidean geometric distance term from the vertices of the original 3D mesh to the orthogonal projection points; Elevation component; This represents the elevation values ​​of the original 3D mesh vertices along the Z-axis in a Cartesian coordinate system.

[0028] Furthermore, the calculated longitudinal arc length component S, lateral normal distance component L, and elevation component H are extracted and merged into a reconstructed coordinate vector, declared as additional vertex attribute parameters, and the graphics application programming interface is called to allocate and write the additional vertex attribute parameters into the underlying memory structure of each original 3D mesh vertex. After the reconstructed coordinates are written, the mesh vertices are no longer limited by Cartesian coordinate dimensions, and subsequent downgrading operations on the multi-level detailed mesh model are triggered based on the reconstructed coordinate vector stored in the underlying memory structure.

[0029] 2) Obtain the reconstructed coordinate vector from the underlying memory structure, and construct a dynamic penalty weight based on the lateral normal distance component. Calculate the cost of folding anisotropic edges. ; Construct dynamic penalty weights After obtaining the underlying memory structure containing the reconstructed coordinate vectors, the conventional quadratic error metric simplification algorithm is modified using the lateral normal distance component to control the degree of mesh dimensionality reduction in the land-water interface region. The mesh reduction calculation process specifically includes the following sub-steps: A threshold for the distance between the water and land boundaries is set, and a dynamic penalty weight function is constructed. The data preprocessing module P100 reads the basic hydrological design data of the river channel, extracts the average normal distance from the river's central axis to the water-land interfaces on both sides under normal water level conditions, and configures this average normal distance as the water-land boundary distance threshold. Simultaneously, a buffer width threshold is configured to set the spatial span of the topological transition zone between the water surface and land. The lateral normal distance component L stored in the underlying memory structure is extracted and compared with the water-land boundary distance threshold. A simplified basic weight constant with smaller values ​​is assigned to the water surface area, while a retained basic weight constant with larger values ​​is assigned to the inland area. For the water-land transition area, a polynomial function is called to generate smoothly transitioning interpolation weights, and the dynamic penalty weight is output. Dynamic penalty weights The calculation formula is expressed as follows: (4) In the formula, For dynamic penalty weights; This represents the lateral normal distance component; This represents the threshold for the distance between the land and water boundaries. This is a threshold for the width between buffers, limited to avoid division by zero errors. The lower limit of the value is 0.001 meters; The basic simplified weighting constant for the water surface area is configured with a value range between 0.01 and 0.1. The basic weight constant for inland areas is reserved, with a value range of 1.0 to 5.0.

[0030] Next, the cost of the basic quadratic error metric is calculated, combined with the dynamic penalty weight. Calculate the cost of folding anisotropic edges. ; In practice, all topological edges in the 3D mesh are traversed. For the topological edges formed by the interconnected first and second vertices, an error quadratic form symmetric matrix is ​​constructed, and the basic quadratic error metric cost is calculated. When the complete coplanarity of local mesh patches causes the error quadratic form symmetric matrix to exhibit singularity and cannot be directly inverted, the singular value decomposition algorithm is called to calculate the pseudo-inverse matrix and solve for the coordinates of the folding target point. After completing the basic calculation, the first dynamic penalty weight corresponding to the first vertex and the second dynamic penalty weight corresponding to the second vertex are extracted. Combining the basic quadratic error metric cost with the dynamic penalty weights of the two vertices, the formula for calculating the folding cost of anisotropic edges is expressed as follows: (5) In the formula: The cost of folding anisotropic edges; The cost of the basic quadratic error metric; The first dynamic penalty weight is the weight corresponding to the first vertex. This is the second dynamic penalty weight corresponding to the second vertex.

[0031] 3. Based on the anisotropic edge folding cost Iterative calculations are performed to generate multi-level detailed mesh data files, which are then solidified into structured data blocks.

[0032] The iterative computation includes: A min-heap data structure is constructed, and the target topological edge at the top of the min-heap data structure is extracted to trigger a vertex fusion command. When the total number of remaining faces is reduced to a preset polygon count threshold, the edge folding iteration is paused, and the retained mesh topology connection data and vertex coordinate data are extracted, serialized, and encapsulated to generate a multi-level detailed mesh data file, which is then solidified into a structured data block. The structured data block includes a basic mesh data area, a topology index data area, a material parameter data area, and an additional attribute data block. The longitudinal arc length component S, the transverse normal distance component L, and the elevation component H are solidified into the additional attribute data block of the structured data block. The polygon count threshold is the limit of the number of discrete polygons contained in the hierarchical file.

[0033] The additional attribute data block is used to store the longitudinal arc length component S, the lateral normal distance component L, and the elevation component H. Based on the cost of folding anisotropic edges During iterative computation, the cost of folding anisotropic edges of all topological edges within the 3D mesh is calculated. Sort the data in ascending order and construct a min-heap data structure based on the sort order. Extract the target topological edge at the top of the min-heap to trigger a vertex merging instruction, merging the first and second vertices into a single new coordinate vertex and removing adjacent triangle faces that have degenerated due to geometric space compression. After the merging instruction is completed, recalculate the anisotropic edge folding cost of adjacent topological edges affected by the merging operation. The node arrangement within the min-heap data structure is then updated. The local topological updates and degenerate patch removal during edge folding are standard techniques in this field and will not be elaborated upon here.

[0034] The process monitors the progress of iterative computation and outputs multi-level detailed mesh data files. A pre-configured hierarchical file containing multiple discrete polygon count thresholds is used. When the total number of remaining faces in the 3D mesh decreases to the polygon count threshold corresponding to the current level, the edge-folding iterative computation is paused, and the retained mesh topology connectivity data and vertex coordinate data are extracted and serialized. By changing the polygon count threshold and performing truncation operations at different nodes, multi-level detailed mesh data files with decreasing precision are generated. During the data serialization and encapsulation stage, the longitudinal arc length component S, the lateral normal distance component L, and the elevation component H from the underlying memory structure are extracted synchronously and solidified into the additional attribute data block of the multi-level detailed mesh data file, providing support for subsequent rendering pipeline parameter intervention.

[0035] After the multi-level detailed mesh data file is serialized, packaged, and saved to external storage, the memory scheduling module takes over the viewpoint state tracking and data loading tasks during the rendering process, transforming the 3D spatial view frustum intersection test into linear memory addressing based on longitudinal river channel features.

[0036] The memory scheduling module P200 is used to acquire the spatial position of the virtual camera in the Cartesian coordinate system in real time, perform projection calculations, and obtain the longitudinal projection arc length coordinates corresponding to the current frame. Derivation of the look-ahead velocity scalar Calculate the one-dimensional sliding scheduling interval, determine the index range, and read the structured data block within the index range into the system video memory (i.e., main memory); then, calculate the longitudinal projection arc length coordinates corresponding to the current frame. and look-ahead velocity scalar The data is passed to the rendering pipeline intervention module; The derivation of the look-ahead velocity scalar include: Real-time acquisition of the spatial position of the virtual camera in a Cartesian coordinate system, application of Kalman filtering, and output of smoothed camera coordinate data; Determine the orthogonal projection point of the current frame and calculate the longitudinal projection arc length coordinates corresponding to the current frame. The orthogonal projection point of the current frame is the curve point closest to the current camera coordinate data; Store several longitudinal projection arc length coordinates into a circular buffer, and extract the historical longitudinal projection arc length coordinates from the circular buffer. Vertical projection arc length coordinates corresponding to the current frame Perform difference quotient calculation and output the look-ahead velocity scalar. .

[0037] Reading data within the index range into the system video memory includes: Based on the aforementioned look-ahead velocity scalar, a dynamic bias term is constructed, tilting the preloaded center of gravity towards the future movement direction of the camera, and the starting arc length coordinates of the one-dimensional sliding scheduling interval are calculated. coordinates of the endpoint arc length This yields a one-dimensional sliding scheduling interval. The one-dimensional sliding scheduling interval is a data loading window that dynamically slides along the central axis of the river channel. Based on the starting arc length coordinates coordinates of the endpoint arc length By calling the floor function, the starting file index of the target data block to be loaded is output. With Termination File Index This forms an index range; An asynchronous block transfer instruction is dispatched to read the data within the index range of the structured data block into the system video memory.

[0038] In practice, the spatial position of the virtual camera in the Cartesian coordinate system is collected in real time, and the projection calculation of the longitudinal arc length coordinate of the curve coordinate system is performed. During the update cycle of the 3D rendering engine, the current global position coordinate of the virtual camera is read, Kalman filtering is applied to the collected global position coordinate, and the smoothed camera coordinate data is output; the timestamp alignment operation is performed synchronously. After obtaining the smoothed camera coordinate data, since the motion trajectory of the virtual camera is time continuous, the curve parameter corresponding to the orthogonal projection point recorded in the previous rendering frame is retrieved as the initial input value of the Newton iteration algorithm in the current round, and converged to the curve point closest to the current camera coordinate data as the orthogonal projection point of the current frame. Then, the spatial mapping transformation formulas (1)-(3) mentioned above are used to calculate the longitudinal projection arc length coordinate of the current frame. The longitudinal projection arc length coordinate compresses the trajectory of the freely moving viewpoint to the one-dimensional river channel extension direction, stripping away redundant dimensions related to lateral and elevation, and forming a reference position for triggering external memory read commands.

[0039] Derivation of the look-ahead velocity scalar along the smooth central axis Single-frame-level positional difference calculations are susceptible to fluctuations in operational input. To smooth the velocity estimation curve, a fixed-capacity circular buffer is allocated in the system's main memory. Several temporally consecutive longitudinal projection arc length coordinates and their corresponding timestamps are sequentially stored in this buffer. When the storage sequence reaches its maximum capacity, the earliest stored data node is evicted. The earliest recorded historical longitudinal projection arc length coordinate in the circular buffer is extracted and compared with the longitudinal projection arc length coordinate corresponding to the current frame; the difference quotient is then calculated, and the look-ahead velocity scalar is output. The calculation formula is expressed as follows: (6) In the formula: Forward velocity scalar, positive values ​​represent positive movement along the smooth central axis, and negative values ​​represent negative movement along the smooth central axis. The coordinates of the vertical projection arc length corresponding to the current frame; The coordinates of the historical longitudinal projection arc length; The timestamp corresponding to the current frame; This is the timestamp corresponding to the historical longitudinal projection arc length coordinate record; The minimum rendering update cycle is set by the parameter. To maximize the function, the look-ahead velocity scalar is calculated. It is then written into the system's global state variable pool to indicate the instantaneous speed and geometric direction of the viewpoint's movement.

[0040] Combined with the vertical projection arc length coordinates of the current frame With look-ahead speed scalar A one-dimensional sliding scheduling interval is calculated to trigger the linear reading of structured data blocks. Essentially, the one-dimensional sliding scheduling interval is a dynamically sliding data loading window along the river's central axis. Its function is to pre-read the 3D mesh data of the area the camera is about to reach, reducing loading lag and image stuttering. Conventional static preloading ranges are insufficient to cover newly appearing topological meshes in front of the viewpoint during rapid viewpoint shifts. This embodiment uses a look-ahead velocity scalar... A dynamic bias term is constructed, tilting the preloaded center of gravity towards the future direction of camera movement. The formulas for calculating the starting and ending arc length coordinates of the one-dimensional sliding scheduling interval are expressed as follows: (7) (8) In the formula: Let be the coordinates of the starting arc length of the one-dimensional sliding scheduling interval; Let [the coordinates be] the arc length of the endpoint of the one-dimensional sliding interval; the one-dimensional sliding interval is: [ ; The coordinates of the vertical projection arc length corresponding to the current frame; The base preload radius is set between 500 and 1000 to provide the base data coverage range when the camera is stationary; For look-ahead velocity scalar; The prediction time constant is set to a value between 2 and 5, which limits the time span of the advance prediction. and Used to expand the loading range forward or backward according to the camera's forward direction.

[0041] After establishing the upper and lower boundaries, perform linear addressing and data transfer operations on the external storage medium. Extract the starting arc length coordinates. coordinates of the endpoint arc length Divide each index by the spatial span constant of a single data block, and call the floor function to output the starting and ending file indices of the target data block to be loaded: (9) (10) in, The starting file index for the target data block; The terminating file index for the target data block; The spatial span constant of a single data block. This is the floor function.

[0042] Asynchronous block transfer instructions are dispatched to the operating system's underlying bus to linearly read structured data blocks within the index range into the system's main memory. Allocated grid data structure pointers are retrieved synchronously, and memory reclamation instructions are initiated for idle data blocks exceeding the current file index range. One-dimensional file index comparison is used instead of three-dimensional view frustum intersection testing to maintain a dynamic balance in main memory consumption.

[0043] After the data block is read into the video memory, the rendering control module P300 takes over the low-level instructions of the graphics processing pipeline to intervene in the depth writing mechanism, eliminating the depth overlap conflict in the densely intersecting areas of polygons.

[0044] The rendering pipeline intervention module P300 includes a fragment shader controller and a vertex shader controller; after the data is read into the system video memory, the fragment shader controller calculates the fragment-level lateral normal distance component based on the structured data block. With fragment-level elevation components Generate nonlinear mapping depth numerical values With stretch mapping depth value The process of overwriting video memory is completed; the vertex shader controller extracts the position deviation vector based on structured data blocks. Combined with the longitudinal projection arc length coordinates corresponding to the current frame And the transition mixing factor derived from the look-ahead velocity scalar The constrained mesh vertices perform smooth deformation calculations only along the two-dimensional cross-section composed of the transverse normal and elevation, generating a dimension-reduced deformation deviation vector. Output the coordinates of the vertices after deformation. ; The deformed vertex coordinates This is used to transmit data to the 3D visualization and maintenance terminal for primitive assembly and rasterization operations. The rendering pipeline intervention module intervenes in the low-level instructions of the graphics processing pipeline to generate stretch mapping depth values ​​to complete the memory overwrite, and uses the transition blending factor to constrain the mesh vertices to perform smooth deformation calculations.

[0045] The calculation fragment-level lateral normal distance component With fragment-level elevation components include: The lateral normal distance component L and the elevation component H are extracted from the structured data block and configured as interpolation variables; When entering the rasterization stage, for any element generated inside the patch, extract the homogeneous coordinates of the vertices in the clip space and calculate the centroid projection weights. Based on the centroid projection weight, the perspective correction interpolation algorithm is invoked to calculate the fragment-level lateral normal distance component. With fragment-level elevation components .

[0046] Specifically, the lateral normal distance component L and the elevation component H are configured as interpolation variables and passed to the fragment shader hardware runtime environment. Traditional hardware depth testing relies on a single matrix transformation result, which is prone to loss of depth accuracy. An attribute communication interface is established in the vertex shader stage to extract the lateral normal distance component L and the elevation component H from the data file and declare them as transferable vertex output variables. When entering the rasterization stage, for any fragment generated within a face, the homogeneous coordinates of the vertices in clip space are extracted to calculate the centroid projection weight, and a perspective correction interpolation algorithm is called to calculate the fragment-level lateral normal distance component. With fragment-level elevation components The interpolation formula is expressed as follows: (11) (12) In the formula: For the fragment-level lateral normal distance component; The weight of the centroid coordinates corresponding to the first vertex is directly generated by the hardware rasterization unit; The lateral normal distance component carried by the first vertex; The fourth component of the homogeneous coordinates of the first vertex in clip space; The weight of the centroid coordinates corresponding to the second vertex; The lateral normal distance component carried by the second vertex; The fourth component of the homogeneous coordinates of the second vertex in clip space; The weight of the centroid coordinates corresponding to the third vertex; The lateral normal distance component carried by the third vertex; The fourth component of the homogeneous coordinates of the third vertex in clip space; For fragment-level elevation components; The elevation component carried by the first vertex; The elevation component carried by the second vertex; The elevation component carried by the third vertex; the tolerance constant is set to 0.00001. When the denominator of the perspective correction interpolation formula is less than the tolerance constant, conventional linear interpolation is forced to replace the perspective correction calculation.

[0047] The generated nonlinear mapping depth value With stretch mapping depth value The process includes: Based on the Gaussian distribution function model, the lateral normal distance component of the structure at the fragment level is described. Build depth offset; Get screen space depth value Combined with the depth offset, a nonlinear mapped depth value is generated. ; Based on the nonlinear mapping depth value Combined with the fragment-level elevation components The polynomial mapping formula is used to generate the stretch mapping depth value. ; Combine the nonlinear mapping depth values and stretch mapping depth values The overwrite instruction is issued to replace the fragment output variable and complete the video memory overwrite.

[0048] Specifically, based on the fragment-level lateral normal distance component The nonlinear depth mapping function is called to generate offset values, and stretching calculations are performed in conjunction with the elevation components to complete the memory overwrite. The lateral normal distance components at the fragment level are compared. When the distance threshold between land and water is used, and the distance falls within the judgment space limit, a smooth depth offset is constructed using a Gaussian distribution function model to generate a nonlinear mapped depth value. The formula is expressed as follows: (13) In the formula: The depth value represents the nonlinear mapping. This is the default screen space depth value output during the rasterization stage; This is the depth offset strength constant, used to construct the depth offset. The value configuration is dynamically allocated based on the depth buffer bit width specification of the target graphics card device, limiting the maximum numerical range of the depth increase. For the fragment-level lateral normal distance component; The aforementioned threshold for the distance between land and water boundaries; The lower limit is set to 0.1 to ensure that the denominator has a non-zero value, which is the control parameter for distributed decay.

[0049] After obtaining the nonlinear mapping depth values, extract the fragment-level elevation components. The normal hydrological fluctuation range of the river channel is configured as the mapping source data range, and the polynomial mapping formula is called to generate the stretched mapping depth value for the fragments within the range: (14) In the formula: This represents the stretch mapping depth value; The near-end depth constant of the standardized equipment coordinate system is configured to be 0.0 in this embodiment; The far-end depth constant of the standardized equipment coordinate system is configured to be 1.0 in this embodiment; For fragment-level elevation components; The lower limit elevation threshold; The upper limit elevation threshold; The elevation tolerance constant is set to 0.1. Threshold truncation is performed on fragments that do not fall within the mapping source range. The merged parameters are overwritten by issuing an overwrite command. By forcibly replacing the fragment output variables, the underlying floating-point bit width is completely concentrated in the geometric topology near the target interface, improving the graphics pipeline's ability to resolve overlapping fragment conflicts.

[0050] The vertex shader controller derives the transition blending factor. At that time, the difference between the current coordinates of the mesh vertex and the target coordinates in the lower detail level model is extracted and defined as the position deviation vector; the level boundary coordinates are extracted based on the basic transition width constant and the look-ahead velocity scalar. The dynamic transition band width parameter is calculated; the difference between the longitudinal projected arc length coordinate and the hierarchical boundary coordinate is calculated, divided by the dynamic transition band width parameter, and the range truncation function is called to obtain the transition blending factor. .

[0051] The derived transition mixing factor include: Based on structured data blocks, the difference between the current coordinates of the mesh vertices and the target coordinates in the lower level of detail model is extracted as a position deviation vector. ; Extract the hierarchical boundary coordinates of the structured data block Set the basic transition width constant. Combined with forward speed scalar Calculate the dynamic transition band width parameter; Get the vertical projection arc length coordinates of the current frame. with hierarchical boundary coordinates The difference is used to derive the transition mixing factor by combining the dynamic transition band width parameter. .

[0052] Specifically, the rendering pipeline intervention module P300 drives the graphics processor to enter the vertex shading stage, constraining the vertex deformation trajectory based on the motion state of the virtual camera, so as to smooth out the visual abrupt changes that occur when switching between multi-level model layers.

[0053] The memory scheduling module P200 will update the vertical projection arc length coordinates corresponding to the current frame in real time. and look-ahead velocity scalar The values ​​are passed to the vertex shader controller and defined as uniform constant variables bound to the global memory context to ensure that all parallel mesh vertices within the same rendering frame read the same parameters. Simultaneously, the system extracts the difference between the current coordinates of the mesh vertex and the target coordinates in the lower detail level model, defining it as a position deviation vector. This is written as a custom vertex attribute into the vertex buffer object of the graphics processor; Derivation of the transition blending factor for the vertex shader controller to characterize the geometric deformation weights Because the fixed-width transition zone of conventional pipelines is prone to drastic weight jumps during high-speed viewpoint movement, this embodiment adds a velocity adjustment term to the set basic transition width constant to obtain a dynamic transition zone width parameter. The spatial proximity is calculated by extracting the hierarchical boundary coordinates of the current grid data block, and a division operation is performed to obtain the transition blending factor. The calculation formula is as follows: (15) In the formula, For transitional mixing factor; The range truncation function is restricted to Within the closed interval; The coordinates of the longitudinal projection arc length of the camera; These are the coordinates of the hierarchical boundary. The basic transition width constant; This is a scaling weight constant; The lookup speed scalar passed to the memory scheduling module; It is a tiny positive constant, with a value range between 0.0001 and 0.001.

[0054] The vertex shader controller extracts the unit tangent vector T and performs orthogonal decomposition and component filtering operations on the position deviation vector to generate a dimension-reduced deformation deviation vector; it calculates the deformed vertex coordinates by adding the transition blending factor and the dimension-reduced deformation deviation vector to the basic 3D coordinates of the currently processed mesh vertex; and it transmits the deformed vertex coordinates to the 3D visualization operation and maintenance terminal to perform primitive assembly operations.

[0055] Specifically, the generation of the dimension-reduced deformation deviation vector include: Based on the position deviation vector Extract the unit tangent vector The position deviation vector is subjected to orthogonal decomposition and component filtering operations to output a dimension-reduced deformation deviation vector. The operation expression is: (16) The output deformed vertex coordinates include: Obtain the basic 3D coordinates of the current processing mesh vertex Combined with transition mixing factor and dimension reduction deformation deviation vector Calculate the vertex coordinates after deformation The calculation expression is: (17) Specifically, the vertex shader controller utilizes a transition blending factor. Constrained mesh vertex deformation trajectory. To avoid texture stretching caused by longitudinal interpolation along the river channel, the vertex shader controller extracts the unit tangent vector T at the orthogonal projection point on the central axis and performs orthogonal decomposition and component filtering operations on the position deviation vector. The constrained vertices only perform smooth deformation calculations along the two-dimensional cross section composed of the transverse normal and elevation.

[0056] The calculation removes the geometric deformation tendency parallel to the river channel direction, ensuring that the remaining reduced-dimensional deformation deviation vector is strictly distributed within the cross-section formed by the lateral and elevation dimensions. This eliminates rendering jumps that cause visual abrupt changes between multi-level detailed mesh models. The calculated vertex coordinates after deformation are then obtained. The data is transmitted to the graphics processor in the 3D visualization operation and maintenance terminal to perform primitive assembly and rasterization operations.

[0057] The river and lake shoreline UAV inspection 3D visualization operation and maintenance management system provided by this invention converts grid data in a rectangular coordinate system into a curved coordinate data structure based on the river's central axis, making the simplification and scheduling process of the 3D model conform to the spatial characteristics of the river system distribution. By introducing lateral normal distance components and elevation components to intervene in the grid simplification process, and modifying the depth buffer by the fragment shader during the rendering stage, depth conflicts of water surface polygons are handled; the vertex shader uses the derived camera movement speed and one-dimensional arc length coordinates to orthogonally decompose and smoothly transition the mesh deformation between different detail levels, filtering out lateral deformation jitter.

[0058] Specific application examples: Implementation scenario setting: Suppose we are conducting an oblique photographic inspection of a 10-kilometer-long, meandering river in southern my country using a drone. The original 3D mesh data contains as many as 50 million vertices (including high-precision water surface and vegetation / buildings on both banks). The virtual inspection camera flies along the river at a speed of 15 m / s.

[0059] Suppose there is a point on the surface of the water at a certain point in the river. After generating a smooth central axis by fitting a cubic B-spline curve using software, the spatial coordinate vector of its nearest orthogonal projection point in the rectangular coordinate system is found. According to the formula: The lateral normal distance component of the point from the central axis is calculated. Meters. Its longitudinal arc length component along the starting point of the river channel. rice.

[0060] Based on on-site hydrological data, a threshold for the distance between the land and water boundaries was set. Meters, buffer width threshold Meters. Define the basic simplified weighting constants for the water surface region. The basic retention weight constant for inland regions .

[0061] Regarding the above The vertex of the meter, because Substituting into the dynamic penalty weight calculation formula, we get... The weight is calculated to be 0.05.

[0062] For the tops of the trees on the shore Rice, because Substituting this into the formula, we get a weight of 3.0.

[0063] Substituting into the formula for the cost of folding anisotropic edges: Subsequently, the folding cost of the water surface grid was extremely low, and it was simplified on a large scale; while the high fidelity of the shoreline land was forcibly preserved.

[0064] It is known that the current camera is located at At meters. Looking-through velocity scalar. m / s (forward flight). Assume a foundation preload radius. Meters, prediction time constant Seconds. Substitute the coordinates of the starting arc length and the ending arc length into the calculation formula: starting point: rice.

[0065] end: rice.

[0066] The traditional scheduling interval should be [1200, 2800]. This system uses the speed formula to shift it forward to [1200, 2845], which advances the redundant loading by 45 meters and completely eliminates the model loading blocking phenomenon during high-speed flight.

[0067] When the model switches between multiple levels of detail, there is a deviation between the target coordinates and the current coordinates of a shoreline vertex. (Position deviation vector) At this point, the direction (tangent vector) of this section of the river is... Substitute into the formula: .

[0068] After orthogonal decomposition and filtering operations, the 0.5-meter spatial stretching deformation parallel to the longitudinal direction of the river channel is completely eliminated, and the grid vertices only undergo smooth deformation along the two-dimensional cross section composed of the transverse normal and elevation.

[0069] Experimental verification and effect comparison: To verify the technological advancement of this invention, a comparative operational experiment was conducted on the same hardware testing platform (containing 64GB of system main memory and 24GB of device video memory) using the aforementioned 10km river channel dataset. The control group employed the traditional method based on 3D view frustum intersection testing and conventional Z-Buffer depth writing; the experimental group used the method provided by this invention.

[0070] Comparison and verification of grid degradation and scheduling efficiency: See attached document Figure 2 This invention is on the water surface ( ) and land ( A continuous nonlinear transition is formed between the two, avoiding mesh tearing caused by direct truncation. Under the same view distance, the total volume of the multi-level detailed mesh data file processed by this method is reduced by 42% compared with the traditional simplification method. After superimposing a one-dimensional linear scheduling mechanism, the peak main memory usage of the experimental group is stable at 4.5GB, while the peak main memory usage of the control group is as high as 14.2GB due to loading a large amount of redundant background data in non-river corridor areas, frequently triggering system page replacement interruptions.

[0071] Deep conflict elimination verification: For the polygonal, shallow water boundary region, the control group exhibited severe depth conflict flicker (conflicting pixels accounted for 34%) under the drone's overhead view. The experimental group of this invention extracted the fragment-level lateral normal distance component, combined it with a Gaussian distribution model to generate nonlinear mapped depth values, and then performed memory overwriting. Test results show that the boundary pixel flicker rate was reduced to below 0.1%, and the execution time of the single-frame memory intervention command was only 0.2ms, achieving a visual quality improvement with extremely low overhead.

[0072] System rendering performance and stability comparison verification: See attached document Figure 3 The high-speed inspection process was monitored for 60 consecutive seconds. The control group (marked with a dashed cross) was monitored when large-scale file loading and model level switching were triggered (e.g., ...). Figure 3At the 15th, 35th, and 50th seconds, the frame rate drops drastically, falling to a minimum of around 25 FPS, resulting in severe stuttering. The method of this invention (marked with solid lines) benefits from the advance scheduling of the look-ahead speed scalar and the smooth deformation constraint mechanism, eliminating pipeline blockage and abrupt redrawing. The frame rate is extremely smooth throughout the rendering process, averaging around 115 FPS, with a minimum frame rate (1%) not lower than 95 FPS, and the system exhibits significant stability.

[0073] The above-disclosed embodiments are merely specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines, characterized in that, include: Data preprocessing module, memory scheduling module, and rendering pipeline intervention module; The data preprocessing module is used to receive raw data collected by UAV inspection equipment, generate multi-level detailed mesh data files, and solidify the multi-level detailed mesh data files into structured data blocks; the raw data includes raw UAV oblique photogrammetry 3D mesh vertex data and river channel centerline discrete vector data in Cartesian coordinate system. The memory scheduling module is used to acquire the spatial position of the virtual camera in the Cartesian coordinate system in real time, perform projection calculations, and obtain the longitudinal projection arc length coordinates corresponding to the current frame. Derivation of the look-ahead velocity scalar Calculate the one-dimensional sliding scheduling interval, determine the index range, and read the data within the index range into the system video memory; The rendering pipeline intervention module includes a fragment shader controller and a vertex shader controller. After reading the data into the system video memory, the fragment shader controller calculates the fragment-level lateral normal distance component based on the structured data block. With fragment-level elevation components Generate nonlinear mapping depth numerical values With stretch mapping depth value The process of overwriting video memory is completed; the vertex shader controller extracts the position deviation vector based on structured data blocks. Combined with the longitudinal projection arc length coordinates corresponding to the current frame And the transition mixing factor derived from the look-ahead velocity scalar Generate a dimension-reduced deformation deviation vector Output the coordinates of the vertices after deformation. .

2. The three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines as described in claim 1, characterized in that, The generation of multi-level detailed mesh data files includes: The original data is converted into a reconstructed coordinate vector and written into the underlying memory structure of each original 3D mesh vertex; the reconstructed coordinate vector includes a longitudinal arc length component S, a transverse normal distance component L, and an elevation component H; The reconstructed coordinate vector is obtained from the underlying memory structure, and a dynamic penalty weight is constructed based on the lateral normal distance component L. Calculate the cost of folding anisotropic edges. ; Based on the anisotropic edge folding cost Iterative calculations are performed to generate multi-level detailed mesh data files, which are then solidified into structured data blocks.

3. The three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines as described in claim 2, characterized in that, Converting the original data into a reconstructed coordinate vector includes: Using discrete vector data of the river channel centerline as control points, a cubic B-spline curve function is constructed to generate a smooth centerline. Based on the smooth central axis, the orthogonal projection point is calculated using the Newton-Raphson iterative algorithm; the orthogonal projection point is the curve point with the smallest Euclidean distance from the vertex data of the original UAV oblique photogrammetry 3D mesh. Based on the Euclidean distance from the vertex data of the original UAV oblique photogrammetry 3D mesh to the orthogonal projection point, calculate the longitudinal arc length component. Horizontal normal distance component and elevation components The original 3D mesh vertices are merged to generate a reconstructed coordinate vector, which is then written as an additional vertex attribute parameter into the underlying memory structure of each original 3D mesh vertex.

4. The three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines as described in claim 2, characterized in that, The iterative computation includes: A min-heap data structure is constructed, and the target topological edge at the top of the min-heap data structure is extracted to trigger the vertex fusion instruction. When the total number of remaining faces is reduced to a preset polygon number threshold, the edge folding iteration operation is paused, and the retained mesh topology connection data and vertex coordinate data are extracted, serialized and encapsulated to generate a multi-level detailed mesh data file, which is then solidified into a structured data block. The polygon number threshold is the limit of the number of discrete polygons contained in the hierarchical file.

5. The three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines as described in claim 1, characterized in that, The derivation of the look-ahead velocity scalar include: Real-time acquisition of the spatial position of the virtual camera in a Cartesian coordinate system, application of Kalman filtering, and output of smoothed camera coordinate data; Determine the orthogonal projection point of the current frame and calculate the longitudinal projection arc length coordinates corresponding to the current frame. The orthogonal projection point of the current frame is the curve point closest to the current camera coordinate data; Store several longitudinal projection arc length coordinates into a circular buffer, and extract the historical longitudinal projection arc length coordinates from the circular buffer. Vertical projection arc length coordinates corresponding to the current frame Perform difference quotient calculation and output the look-ahead velocity scalar. .

6. The three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines as described in claim 1, characterized in that, Reading data within the index range into the system video memory includes: Based on the aforementioned look-ahead velocity scalar, a dynamic bias term is constructed, tilting the preloaded center of gravity towards the future movement direction of the camera, and the starting arc length coordinates of the one-dimensional sliding scheduling interval are calculated. coordinates of the endpoint arc length This yields a one-dimensional sliding scheduling interval. The one-dimensional sliding scheduling interval is a data loading window that dynamically slides along the central axis of the river channel. Based on the starting arc length coordinates coordinates of the endpoint arc length By calling the floor function, the starting file index of the target data block to be loaded is output. With Termination File Index This forms an index range; An asynchronous block transfer instruction is dispatched to read the data within the index range of the structured data block into the system video memory.

7. The three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines as described in claim 1, characterized in that, The calculation fragment-level lateral normal distance component With fragment-level elevation components include: The lateral normal distance component and elevation component are extracted from the structured data block and configured as interpolation variables; When entering the rasterization stage, for any element generated inside the patch, extract the homogeneous coordinates of the vertices in the clip space and calculate the centroid projection weights. Based on the centroid projection weight, the perspective correction interpolation algorithm is invoked to calculate the fragment-level lateral normal distance component. With fragment-level elevation components .

8. The three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines as described in claim 7, characterized in that, The generated nonlinear mapping depth value With stretch mapping depth value The process includes: Based on the Gaussian distribution function model, the lateral normal distance component of the structure at the fragment level is described. Build depth offset; Get screen space depth value Combined with the depth offset, a nonlinear mapped depth value is generated. ; Based on the nonlinear mapping depth value Combined with the fragment-level elevation components The polynomial mapping formula is used to generate the stretch mapping depth value. ; Combine the nonlinear mapping depth values and stretch mapping depth values The overwrite instruction is issued to replace the fragment output variable and complete the video memory overwrite.

9. The three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines as described in claim 1, characterized in that, The derived transition mixing factor include: Based on structured data blocks, the difference between the current coordinates of the mesh vertices and the target coordinates in the lower level of detail model is extracted as a position deviation vector. ; Extract the hierarchical boundary coordinates of the structured data block Set the basic transition width constant. Combined with forward speed scalar Calculate the dynamic transition band width parameter; Get the vertical projection arc length coordinates of the current frame. with hierarchical boundary coordinates The difference is used to derive the transition mixing factor by combining the dynamic transition band width parameter. .

10. The three-dimensional visualization operation and maintenance management system for unmanned aerial vehicle (UAV) inspection of river and lake shorelines as described in claim 9, characterized in that, The generation of the dimension-reduced deformation deviation vector include: Based on the position deviation vector Extract the unit tangent vector The position deviation vector is subjected to orthogonal decomposition and component filtering operations to output a dimension-reduced deformation deviation vector. The operation expression is: 。 The output deformed vertex coordinates include: Obtain the basic 3D coordinates of the current processing mesh vertex Combined with transition mixing factor and dimension reduction deformation deviation vector Calculate the vertex coordinates after deformation The calculation expression is: 。

Images