Three-dimensional incremental model updating method and device combining grid diagram and boundary optimization

Abstract

The invention discloses a grid diagram and boundary optimization combined three-dimensional incremental model updating method and device, and the method comprises the steps: setting a plane as a grid diagram, and carrying out the fine registration of an incremental model and a stock model based on the registration relation between an incremental model layering line cluster and a stock model layering line cluster on the grid diagram; obtaining a to-be-updated overlapping region through the overlapping part of the incremental model and the stock model; projecting the point cloud in the to-be-updated overlapping region to a two-dimensional plane to obtain a two-dimensional point set, constructing a triangulation network for the two-dimensional point set through triangulation, and forming an initial contour along the boundary of the triangulation network; optimizing the initial contour to obtain an optimized mosaic line; and respectively cutting the triangulation network of the incremental model and the triangulation network of the stock model by using the optimized mosaic line, and merging the triangulation network parts after cutting to complete updating of the stock model. By the adoption of the technical scheme, fine registration is achieved, and the accuracy and smoothness of the complex scene boundary are improved.

Description

Technical Field

[0001] This invention relates to the field of surveying and mapping technology, and in particular to a method and apparatus for updating a three-dimensional incremental model by combining raster maps and boundary optimization. Background Technology

[0002] For detailed 3D modeling of complex scenes (such as undulating terrain, urban building complexes, bridges and tunnels), existing technologies mainly employ multi-source data acquisition, traditional registration algorithms, and manual intervention in model updates.

[0003] However, existing technologies have several shortcomings, including: difficulty in multi-source data fusion; significant differences in data format and coordinate reference between different sensors, making it difficult for newly acquired incremental models to perfectly match the original existing models, affecting the overall consistency of the models; low automation: existing methods rely on manual intervention for boundary adjustment and model stitching, which is inefficient and costly, making it difficult to meet the needs of high-frequency and large-scale model updates; and rough model boundary processing: obvious seams exist at the model fusion boundaries, affecting the visual effect, and the lack of an adaptive boundary generation mechanism leads to unstable model update quality.

[0004] Furthermore, with the widespread application of real-world 3D models in fields such as smart cities and digital twins, higher demands are placed on the real-time performance, accuracy, and automation of model updates, which existing technologies struggle to meet. Summary of the Invention

[0005] Purpose of the invention: The present invention provides a method and apparatus for updating a three-dimensional incremental model by combining raster maps and boundary optimization, aiming to solve the technical problems of insufficient registration accuracy of multi-source data and coarse boundary processing in the prior art.

[0006] Technical Solution: This invention provides a method for updating a 3D incremental model using a combined raster map and boundary optimization, comprising: acquiring local scene images for a local scene to be updated in a target scene, generating a corresponding 3D incremental model for updating the existing 3D model of the target scene; setting a plane as a raster map on the spatial coordinate system shared by the incremental and existing models, and mapping the point clouds of the incremental and existing models to the raster map respectively, obtaining corresponding incremental point sets and existing point sets on the raster map, clustering the point sets to obtain incremental layered line clusters and existing layered line clusters, registering the incremental layered line clusters and existing layered line clusters to obtain corresponding registration relationships, and performing fine registration of the incremental and existing models based on the registration relationships; setting corresponding incremental bounding boxes and existing bounding boxes for the incremental and existing models, and discretizing the bounding boxes into multiple voxel grids, and optimizing the voxel grids according to the voxel grids... The point cloud labels correspond to the incremental and existing point cloud indices of the bounding box voxel grid. The intersection of the incremental and existing point cloud indices is calculated, and the voxel grid in the intersection is the overlay region to be updated. For the point cloud in the overlay region to be updated, the elevation dimension is eliminated and projected onto a two-dimensional plane to obtain a two-dimensional point set. A triangular mesh is constructed on the two-dimensional point set through triangulation, and an initial contour is formed along the boundary of the triangular mesh. The initial contour is optimized to obtain optimized mosaic lines that meet the requirements of smoothness and geometric consistency. The optimized mosaic lines are used to trim the triangular meshes of the incremental model and the existing model respectively, retaining the existing model triangular mesh part outside the overlay region to be updated and the incremental model triangular mesh part inside the overlay region to be updated. The existing model triangular mesh part and the incremental model triangular mesh part are merged to complete the update of the existing model.

[0007] Specifically, the generation of the corresponding three-dimensional incremental model includes: unifying the data format and coordinate system of the incremental model and the existing model.

[0008] Specifically, in the spatial coordinate system O-XYZ, in a direction perpendicular to any coordinate plane, multiple first vertical planes and multiple second vertical planes that are parallel to each other and spaced at a standard distance are set, wherein the first vertical planes and the second vertical planes are also perpendicular to each other.

[0009] Specifically, for each raster image, points with a distance less than a threshold distance are extracted from the point clouds of the incremental and stock models and mapped to the raster image; the threshold distance is less than the standard distance.

[0010] Specifically, within the overlapping region to be updated, the covariance matrix of the point cloud vertex set of the existing model or the incremental model is calculated, and the eigenvectors are decomposed. The point cloud vertices are then projected onto a two-dimensional plane spanned by the two eigenvectors obtained from the decomposition.

[0011] Specifically, traverse all edges of the triangular mesh, mark the edges that belong to only one triangle as boundary edges, connect the boundary edges to obtain multiple closed polygons, and select the closed polygon with the largest area as the initial contour.

[0012] Specifically, the initial contour is constrained and iteratively calculated using an energy function until the energy function converges. The resulting optimized mosaic line is deemed to meet the requirements of smoothness and geometric consistency. The constraints of the energy function include the deviation between the distance from the point on the optimized mosaic line to the surface of the incremental model and the distance to the surface of the existing model; and the distance between the point on the optimized mosaic line and the corresponding point on the initial contour.

[0013] Specifically, the energy function E total The formula is as follows: E total =λ1E error +λ2E aviod +λ3E guide +λ4E reg , Among them, E error For the error field term, E error =Σ p∈C SDF z (p)-SDF c (p)|, where p represents a point on the optimized tessellation line, C represents the optimized tessellation line, and SDF z (p) represents the distance from a point on the optimized mosaic line to the surface of the incremental model, SDF c (p) represents the distance from a point on the optimized mosaic line to the surface of the stock model, E aviod This represents the obstacle avoidance parameter, set to a maximum value in obstacle areas and 0 in non-obstacle areas, E. guide E represents the leading term. guide =Σ p∈C ||p-ΠC(p)|| 2 Where ΠC(p) represents the projection of points on the initial contour onto the optimized tessellation line, and E reg E represents the smoothing term. reg =Σ p∈C ||Δt(p)|| 2 , where Δt(p) represents the tangent vector of the optimized mosaic line at point p, and λ1 to λ4 represent the weights of the corresponding constraint terms; The following formula is used to iterate the optimization of the mosaic lines: p (k+1) =p (k) -μΔE total (p (k) ), Where k represents the number of iterations, μ represents the step size controlling the optimization of the boundary point position of the mosaic line in each iteration, and p (k) and p (k+1) Let ΔE represent the positions of point p in the k-th and (k+1)-th iterations, respectively. total (p (k) ) represents the energy function E total At point p (k) The gradient at that point.

[0014] Specifically, merging the existing model triangular mesh portion and the incremental model triangular mesh portion includes: reconstructing the triangular mesh region along the optimized mosaic line by splicing it together through triangulation.

[0015] This invention also provides a 3D incremental model update device combining raster mapping and boundary optimization, comprising: a model generation unit, a registration unit, an overlay region calculation unit, a contour determination unit, a contour optimization unit, and an update unit, wherein: the model generation unit is used to acquire local scene images for a local scene to be updated in a target scene, generate a corresponding 3D incremental model, and update the existing 3D model of the target scene; the registration unit is used to set a plane as a raster map on the spatial coordinate system shared by the incremental model and the existing model, and map the point clouds of the incremental model and the existing model to the raster map respectively, obtaining corresponding incremental point sets and existing point sets on the raster map, clustering the point sets to obtain incremental layered line clusters and existing layered line clusters, registering the incremental layered line clusters and existing layered line clusters to obtain corresponding registration relationships, and performing fine registration of the incremental model and the existing model based on the registration relationships; the overlay region calculation unit is used to set corresponding incremental bounding boxes and existing bounding boxes for the incremental model and the existing model, and... The bounding box is discretized into multiple voxel grids. Based on the point cloud labels within the voxel grids, the incremental and existing point cloud indices of the bounding box voxel grids are obtained respectively. The intersection between the incremental and existing point cloud indices is calculated, and the voxel grids in the intersection are the overlapping regions to be updated. The contour determination unit is used to eliminate the elevation dimension of the point cloud in the overlapping region to be updated and project it onto a two-dimensional plane to obtain a two-dimensional point set. The two-dimensional point set is triangulated to construct a triangular mesh, and an initial contour is formed along the boundary of the triangular mesh. The contour optimization unit is used to optimize the initial contour to obtain optimized mosaic lines that meet the requirements of smoothness and geometric consistency. The update unit is used to use the optimized mosaic lines to trim the triangular meshes of the incremental model and the existing model respectively, retaining the existing model triangular mesh part outside the overlapping region to be updated and the incremental model triangular mesh part inside the overlapping region to be updated. The existing model triangular mesh part and the incremental model triangular mesh part are merged to complete the update of the existing model.

[0016] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: fine registration of existing and incremental models is achieved by registering layered line clusters on the raster image; the overlapping region to be updated is determined based on voxelization space and point cloud index, and the mosaic lines obtained by triangular mesh boundary optimization are combined to improve the accuracy and smoothness of complex scene boundaries; the overall use of automated algorithms reduces manual intervention. Attached Figure Description

[0017] Figure 1 A schematic diagram illustrating the steps of the three-dimensional incremental model update method for joint raster graph and boundary optimization provided by the present invention; Figure 2 and Figure 3 These are a top plan view and a side perspective view of the fence-like spatial structure provided by the present invention; Figure 4 and Figure 5 These are schematic diagrams showing the projection distribution of the incremental and stock models on the raster diagram, representing the ideal and actual situations provided by this invention, respectively. Figure 6 This is a schematic diagram of the initial contour extraction provided by the present invention. Detailed Implementation

[0018] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0019] See Figure 1 This is a schematic diagram illustrating the steps of the three-dimensional incremental model update method for joint raster graph and boundary optimization provided by the present invention.

[0020] In this embodiment of the invention, for a local scene to be updated in the target scene, a local scene image is acquired, and a corresponding three-dimensional incremental model is generated to update the existing three-dimensional model of the target scene.

[0021] In practice, the existing model refers to the 3D model of the existing real scene; the incremental model refers to the 3D model of the real scene generated from data re-collected for local areas that need to be updated.

[0022] In practice, drones equipped with oblique photography equipment or ordinary close-range optical digital cameras can be used to acquire optical image data of the local scene to be updated. Incremental models are then generated using photogrammetric 3D modeling technology. After data standardization processing, such as unifying the real-scene 3D data format and unifying the coordinate system, the incremental model and the existing model are initially aligned.

[0023] In this embodiment of the invention, a plane is set as a raster graph on the spatial coordinate system shared by the incremental model and the stock model, and the point clouds of the incremental model and the stock model are mapped to the raster graph respectively, so that the corresponding incremental point set and stock point set are obtained on the raster graph respectively.

[0024] See Figure 2 and Figure 3 These are top plan views of the fence-like spatial structure provided by the present invention. Figure 2 ) and side 3D diagram ( Figure 3 ).

[0025] In this embodiment of the invention, in the spatial coordinate system O-XYZ, in the direction perpendicular to any coordinate plane (taking the XOY coordinate plane as an example), a plurality of first vertical planes (referring to those perpendicular to the XOY coordinate plane) (e.g., A1 to A3) and a plurality of second vertical planes (referring to those perpendicular to the XOY coordinate plane) (e.g., B1 to B6) are set up that are parallel to each other and spaced at a standard distance. The first vertical plane and the second vertical plane are also perpendicular to each other (the dot product of the normal vectors is 0), that is, the first vertical plane and the second vertical plane are respectively the horizontal and vertical sections in the direction perpendicular to the XOY coordinate plane.

[0026] In practice, the unified spatial coordinate datum O-XYZ after standardization of the existing model and the incremental model is used to divide the real scene model into vertically intersecting horizontal and vertical sections at equal intervals, forming a fence-like spatial structure composed of two sets of mutually parallel vertical slice planes.

[0027] In this embodiment of the invention, for each raster image, points with a distance less than a threshold distance are extracted from the point clouds of the incremental model and the stock model and mapped to the raster image; the threshold distance is less than the standard distance (both can be set according to the actual application scenario).

[0028] In practical implementation, taking the target raster image as an example, points that are less than a threshold distance from the target raster image are extracted from the point clouds of the incremental model and the stock model, and these points are projected onto the target raster image to obtain the corresponding incremental point set and stock point set respectively on the target raster image.

[0029] In practice, Figure 2 The overlapping area mentioned later is the overlapping area to be updated.

[0030] See Figure 4 and Figure 5 These are the ideal cases provided by the present invention ( Figure 4 ) and actual situation ( Figure 5 A schematic diagram of the projection distribution of the incremental model and the stock model on the raster diagram.

[0031] In this embodiment of the invention, the point set is clustered to obtain incremental hierarchical line clusters and existing hierarchical line clusters, the incremental hierarchical line clusters and existing hierarchical line clusters are registered to obtain the corresponding registration relationship, and the incremental model and existing model are finely registered based on the registration relationship.

[0032] In practice, ideally, the existing model and the incremental model are strictly aligned, so the projection points on each raster map are distributed in a single line. However, in reality, some raster maps located in the overlapping areas of the models may be distributed in double or multiple lines.

[0033] In practice, for the effective area of ​​a meaningful point cloud distribution that requires registration (i.e., the area covered by the point set), incremental hierarchical line clusters and existing hierarchical line clusters are obtained by clustering (connecting) the point set.

[0034] In the specific implementation, a two-dimensional iterative nearest neighbor registration algorithm is adopted, and registration is performed according to two-dimensional rigid body transformation to achieve fine registration of the layered curve clusters in the effective area of ​​each raster image. A series of two-dimensional transformations of all raster images and all effective areas are restored to three-dimensional space, and the global optimal three-dimensional rigid body transformation result is obtained by aggregation optimization (achieving global optimal registration), thus realizing fine registration between the existing model and the incremental model.

[0035] In practice, the minimum common bounding box of the existing model and the incremental model is determined based on the spatial range of the point cloud coordinates. The minimum common bounding box is actually determined by the minimum and maximum coordinates of the union of the two models. It is mainly used to define the size of the bounding box corresponding to each of the two models.

[0036] In this embodiment of the invention, corresponding (minimum) incremental bounding boxes and (minimum) stock bounding boxes are set for the incremental model and the stock model, and the bounding boxes are discretized into multiple voxel grids.

[0037] In practical implementation, the incremental bounding box B is defined as L×W×H. z and existing enclosed box B c Define the voxel sampling interval ΔV, and then sample the bounding box B. z and B c Discretize into i z ×j z ×k z and i c ×j c ×k c The voxel grid, where i, j, and k are the number of voxel grids (spatial blocks) in the x, y, and z directions, respectively, i z =i c =(L / △V)+1,j z =j c =(W / △V)+1,k z =k c =(H / △V)+1.

[0038] In this embodiment of the invention, based on the point cloud labels within the voxel grid, the incremental point cloud index and the existing point cloud index of the bounding box voxel grid are obtained respectively. The intersection between the incremental point cloud index and the existing point cloud index is calculated. The voxel grid in the intersection is the overlapping region to be updated.

[0039] In practice, the point cloud index is used to query the point cloud in each voxel grid, which is equivalent to storing the position label of each point in the voxel grid.

[0040] In practice, the intersection between the incremental point cloud index and the existing point cloud index stores the position labels of points in the voxel grid in each of the two models. For example, voxel grid cells with points are labeled as 1, and those without points are labeled as 0. Finding the intersection means taking all voxel grid cells with labels of 1. This is equivalent to obtaining the overlapping region to be updated, which is the overlapping part of the incremental model and the existing model. However, by adopting the above method of calculating the overlapping part, the accuracy of the overlapping part can be significantly improved.

[0041] In this embodiment of the invention, for the point cloud in the overlapping area to be updated, the elevation dimension is eliminated and projected onto a two-dimensional plane to obtain a two-dimensional point set. A triangular mesh is constructed on the two-dimensional point set through Delaunay triangulation, and an initial contour is formed along the boundary of the triangular mesh.

[0042] In this embodiment of the invention, within the overlapping region to be updated, the covariance matrix of the point cloud vertex set of the existing model or the incremental model is calculated, and the eigenvectors are decomposed. The point cloud vertices are then projected onto a two-dimensional plane spanned by the two eigenvectors obtained from the decomposition.

[0043] In practice, since the two models have been finely registered and are within the overlapping area to be updated, either model can be projected into a two-dimensional point set, and the existing model is preferred.

[0044] In practice, the data dimensionality reduction method based on principal component analysis (PCA) eliminates elevation interference and reduces computational dimensionality.

[0045] See Figure 6 This is a schematic diagram of the initial contour extraction provided by the present invention.

[0046] In this embodiment of the invention, all edges of the triangular mesh are traversed, and edges belonging to only one triangle are marked as boundary edges. Multiple closed polygons are obtained by connecting the boundary edges, and the closed polygon with the largest area is selected as the initial contour.

[0047] In practice, the triangular mesh constructed by triangulation is stored according to the structure of vertices and triangular faces, where a triangular face refers to the triangular face formed by connecting the vertices of the triangular mesh.

[0048] In this embodiment of the invention, the initial contour is optimized to obtain an optimized mosaic line that meets the requirements of smoothness and geometric consistency.

[0049] In practice, "smooth" refers to a smooth surface without drastic changes in the lines, while "geometric consistency" means that the lines are consistent with the boundary direction of the area to be updated, that is, consistent with the boundary direction of the actual scene corresponding to the area to be updated.

[0050] In this embodiment of the invention, an energy function is used to constrain and iteratively calculate the initial contour until the energy function converges (the convergence condition can be set according to the actual application scenario). The resulting optimized mosaic line is deemed to meet the requirements of smoothness and geometric consistency. The constraints of the energy function include the deviation between the distance between points on the optimized mosaic line and the incremental model surface and the distance between points on the optimized mosaic line and the corresponding points on the initial contour.

[0051] In practice, smoothness and geometric consistency are the optimization goals. After multiple iterations, when the energy function converges, it can be considered that the smoothness and geometric consistency requirements are met, rather than requiring absolute smoothness and geometric consistency.

[0052] In this embodiment of the invention, the energy function E total The formula is as follows: E total =λ1E error +λ2E aviod +λ3E guide +λ4E reg , Wherein, λ1 to λ4 represent the weights of the corresponding constraint terms: λ1 controls the importance of the error field term, and the larger the value, the closer the mosaic line is to the geometric boundary between the existing model and the incremental model; λ2 controls the strength of the obstacle avoidance term, which is usually set to a large value to ensure that the mosaic line is located in areas with no significant height jumps, such as the road surface; λ3 controls the strength of the guidance term, and the value directly affects the degree of deviation between the mosaic line and the initial contour; λ4 controls the importance of the smoothing term, and the larger the value, the smoother the mosaic line.

[0053] To minimize the difference in the surface distance fields of the optimized mosaic lines in the incremental and stock models, the error field term E is defined. error =Σ p∈C SDF z (p)-SDF c (p)|, where p represents a point on the optimized tessellation line, C represents the optimized tessellation line, and SDF z (p) represents the directed distance from a point on the optimized tessellation line to the surface of the incremental model, SDF c(p) represents the distance from a point on the optimized mosaic line to the surface of the stock model, specifically the shortest directed vertical distance to the model surface (model fitting surface) (positive and negative indicate inside and outside). SDF is the directed distance function, which quantizes the distance from the point to the model surface.

[0054] To prevent optimized mosaic lines from crossing obstacle areas (such as building outlines, feature boundaries, etc.), an obstacle avoidance term E is defined. aviod It is set to a maximum value (e.g., 1e6) in areas with obstacles (such as building outlines) and to 0 in non-obstacle areas.

[0055] To ensure that the optimized mosaic lines do not deviate excessively from the initial contour, thus affecting the reliability of the update results, an optimization guiding term E is defined. guide =Σ p∈C ||p-ΠC(p)|| 2 , where ΠC(p) represents the projection of points on the optimized tessellation line onto the initial contour. The smaller this term is, the closer the tessellation line is to the initial contour.

[0056] To ensure the smoothness of the tessellation lines and avoid sharp turns or jitter, a smoothness term E is defined. reg =Σ p∈C ||Δt(p)|| 2 , where Δt(p) represents the tangent vector of the optimized mosaic line at point p.

[0057] The optimized mosaic line is iteratively updated using the following formula, i.e., using gradient descent to iteratively update the boundary point positions of the contour: p (k+1) =p (k) -μΔE total (p (k) ), Where k represents the number of iterations, μ represents the step size controlling the optimization of the boundary point position of the mosaic line in each iteration, and p (k) and p (k+1) Let represent the positions of point p in the k-th and (k+1)-th iterations, respectively, used to balance the convergence rate and stability of the iteration process, ΔE. total (p (k) ) represents the energy function E total At point p (k) The gradient at a point represents the direction of the fastest energy change at that point.

[0058] In this embodiment of the invention, optimized mosaic lines are used to trim the triangular mesh of the incremental model and the triangular mesh of the existing model respectively, retaining the triangular mesh portion of the existing model outside the overlapping region to be updated, and retaining the triangular mesh portion of the incremental model inside the overlapping region to be updated, and merging the triangular mesh portion of the existing model and the triangular mesh portion of the incremental model to complete the update of the existing model.

[0059] In practice, the triangular mesh is presented in the form of mesh vertices and triangular patches formed by connecting vertices. There is a correspondence between the vertices and the point cloud vertices. Therefore, the triangular mesh parts of the existing model and the incremental model inside and outside the overlapping region to be updated can be determined by the range of the point cloud in the overlapping region to be updated.

[0060] In practice, merging the existing model triangular mesh and the incremental model triangular mesh involves merging the mesh vertices and triangular faces within the triangular mesh.

[0061] In practice, both the incremental model's triangulation and the stock model's triangulation can be obtained by eliminating the elevation dimension of the incremental or stock model, projecting the point cloud onto a two-dimensional plane to obtain a two-dimensional point set, and then constructing a triangulation from the two-dimensional point set through triangulation.

[0062] In this embodiment of the invention, after the optimized mosaic lines are used for trimming and the existing model triangular mesh portion and the incremental model triangular mesh portion are merged, the triangular mesh area traversed by the optimized mosaic lines is spliced ​​and reconstructed through triangulation.

[0063] In practice, regarding the use of optimized tessellation lines to trim the triangular meshes of the incremental model and the existing model respectively, and to determine whether the triangular facets of the triangular mesh are within the optimized tessellation lines, this can be done by constructing a set of points formed by the center points of each triangular facet. Then, the triangular facets are retained or removed based on the mapping relationship between the center points and the facets. If the center point is within the optimized tessellation line, the corresponding triangular facet is retained, and vice versa.

[0064] In practice, after using optimized mosaic lines to cut the triangular mesh, there will be certain gap areas between the existing model triangular mesh and the incremental model triangular mesh. Therefore, it can be reconstructed by triangulation.

[0065] In practice, the crack area refers to the spatially disconnected gap between two models caused by the trimming process (trimming the two model triangular meshes within the optimized mosaic line).

[0066] In practice, the triangular meshing formed by the triangulation is equivalent to reconstructing and stitching the points of the existing model and the incremental model near the optimized boundary line (optimized mosaic line). Therefore, after meshing, the two models (in the overlapping area to be updated) become one model.

[0067] In practice, updating existing models also includes processing texture resources, including integrating texture images and updating material file references; performing texture harmonization, and eliminating color differences and lighting differences at the seams through color space transformation (such as Lab), histogram matching, and deep learning-based style transfer techniques to achieve a natural visual transition.

[0068] This invention also provides a 3D incremental model update device combining raster mapping and boundary optimization, comprising: a model generation unit, a registration unit, an overlay region calculation unit, a contour determination unit, a contour optimization unit, and an update unit, wherein: the model generation unit is used to acquire local scene images for a local scene to be updated in a target scene, generate a corresponding 3D incremental model, and update the existing 3D model of the target scene; the registration unit is used to set a plane as a raster map on the spatial coordinate system shared by the incremental model and the existing model, and map the point clouds of the incremental model and the existing model to the raster map respectively, obtaining corresponding incremental point sets and existing point sets on the raster map, clustering the point sets to obtain incremental layered line clusters and existing layered line clusters, registering the incremental layered line clusters and existing layered line clusters to obtain corresponding registration relationships, and performing fine registration of the incremental model and the existing model based on the registration relationships; the overlay region calculation unit is used to set corresponding incremental bounding boxes and existing bounding boxes for the incremental model and the existing model, and... The bounding box is discretized into multiple voxel grids. Based on the point cloud labels within the voxel grids, the incremental and existing point cloud indices of the bounding box voxel grids are obtained respectively. The intersection between the incremental and existing point cloud indices is calculated, and the voxel grids in the intersection are the overlapping regions to be updated. The contour determination unit is used to eliminate the elevation dimension of the point cloud in the overlapping region to be updated and project it onto a two-dimensional plane to obtain a two-dimensional point set. The two-dimensional point set is triangulated to construct a triangular mesh, and an initial contour is formed along the boundary of the triangular mesh. The contour optimization unit is used to optimize the initial contour to obtain optimized mosaic lines that meet the requirements of smoothness and geometric consistency. The update unit is used to use the optimized mosaic lines to trim the triangular meshes of the incremental model and the existing model respectively, retaining the existing model triangular mesh part outside the overlapping region to be updated and the incremental model triangular mesh part inside the overlapping region to be updated. The existing model triangular mesh part and the incremental model triangular mesh part are merged to complete the update of the existing model.

[0069] In specific implementations, the methods, steps, or functions executed by the execution units of the three-dimensional incremental model update apparatus for joint raster graph and boundary optimization provided by the present invention can refer to the three-dimensional incremental model update method for joint raster graph and boundary optimization provided by the present invention.

Claims

1. A method for updating a three-dimensional incremental model using a combination of raster graph and boundary optimization, characterized in that, include: For a local scene in the target scene that needs to be updated, acquire local scene images and generate a corresponding 3D incremental model to update the existing 3D model of the target scene. In the spatial coordinate system shared by the incremental model and the stock model, a plane is set as a raster map, and the point clouds of the incremental model and the stock model are mapped to the raster map respectively. The corresponding incremental point set and stock point set are obtained on the raster map respectively. The point sets are clustered to obtain incremental layered line clusters and stock layered line clusters. The incremental layered line clusters and stock layered line clusters are registered to obtain the corresponding registration relationship. Based on the registration relationship, the incremental model and the stock model are finely registered. Set corresponding incremental bounding boxes and stock bounding boxes for incremental and stock models, and discretize the bounding boxes into multiple voxel grids. Based on the point cloud labels in the voxel grids, obtain the incremental point cloud index and stock point cloud index of the bounding box voxel grids respectively. Calculate the intersection between the incremental point cloud index and the stock point cloud index. The voxel grid in the intersection is the overlapping region to be updated. For the point cloud in the overlapping area to be updated, the elevation dimension is eliminated and projected onto a two-dimensional plane to obtain a two-dimensional point set. A triangular mesh is constructed on the two-dimensional point set through triangulation, and an initial contour is formed along the boundary of the triangular mesh. The initial contour is optimized to obtain an optimized mosaic line that meets the requirements of smoothness and geometric consistency. The triangular meshes of the incremental model and the existing model are clipped using optimized mosaic lines. The existing model triangular mesh portion outside the overlapping region to be updated and the incremental model triangular mesh portion inside the overlapping region to be updated are retained. The existing model triangular mesh portion and the incremental model triangular mesh portion are then merged to complete the update of the existing model.

2. The three-dimensional incremental model update method based on joint raster graph and boundary optimization according to claim 1, characterized in that, The generation of the corresponding three-dimensional incremental model then includes: The data format and coordinate system are unified for both incremental and stock models.

3. The three-dimensional incremental model update method based on joint raster graph and boundary optimization according to claim 1, characterized in that, The setting plane as a grid map includes: In the spatial coordinate system O-XYZ, in a direction perpendicular to any coordinate plane, there are multiple first vertical planes that are parallel to each other and spaced at a standard distance, and multiple second vertical planes that are parallel to each other and spaced at a standard distance. The first vertical planes and the second vertical planes are also perpendicular to each other.

4. The three-dimensional incremental model update method based on joint raster graph and boundary optimization according to claim 3, characterized in that, The mapping from point clouds in the incremental model and the stock model to raster images includes: For each raster image, points with a distance less than a threshold distance are extracted from the point clouds of the incremental and stock models and mapped to the raster image; the threshold distance is less than the standard distance.

5. The three-dimensional incremental model update method based on joint raster graph and boundary optimization according to claim 1, characterized in that, The step of eliminating the elevation dimension and projecting the point cloud within the overlapping region to be updated onto a two-dimensional plane includes: Within the overlapping region to be updated, the covariance matrix is ​​calculated for the set of point cloud vertices of the existing model or the incremental model, and the eigenvectors are decomposed. The point cloud vertices are then projected onto a two-dimensional plane spanned by the two eigenvectors obtained from the decomposition.

6. The three-dimensional incremental model update method based on joint raster graph and boundary optimization according to claim 1, characterized in that, The initial contour formation along the boundary of the triangular mesh includes: Traverse all edges of the triangular mesh, mark the edges that belong to only one triangle as boundary edges, connect the boundary edges to obtain multiple closed polygons, and select the closed polygon with the largest area as the initial contour.

7. The three-dimensional incremental model update method based on joint raster graph and boundary optimization according to claim 1, characterized in that, The optimization of the initial contour includes: The initial contour is constrained and iteratively calculated using an energy function until the energy function converges. The resulting optimized mosaic line is deemed to meet the requirements of smoothness and geometric consistency. The constraints of the energy function include the deviation between the distance from the point on the optimized mosaic line to the surface of the incremental model and the distance to the surface of the existing model; and the distance between the point on the optimized mosaic line and the corresponding point on the initial contour.

8. The three-dimensional incremental model update method based on joint raster graph and boundary optimization according to claim 7, characterized in that, The energy function E total The formula is as follows: E total =λ1E error +λ2E aviod +λ3E guide +λ4E reg , Among them, E error For the error field term, E error =Σ p∈C SDF z (p)-SDF c (p)|, where p represents a point on the optimized tessellation line, C represents the optimized tessellation line, and SDF z (p) represents the distance from a point on the optimized mosaic line to the surface of the incremental model, SDF c (p) represents the distance from a point on the optimized mosaic line to the surface of the stock model, E aviod This represents the obstacle avoidance parameter, set to a maximum value in obstacle areas and 0 in non-obstacle areas, E. guide E represents the leading term. guide =Σ p∈C ||p-ΠC(p)|| 2 Where ΠC(p) represents the projection of points on the initial contour onto the optimized tessellation line, and E reg E represents the smoothing term. reg =Σ p∈C ||Δt(p)|| 2 , where Δt(p) represents the tangent vector of the optimized mosaic line at point p, and λ1 to λ4 represent the weights of the corresponding constraint terms; The following formula is used to iterate the optimization of the mosaic lines: p (k+1) =p (k) -μΔE total (p (k) ), Where k represents the number of iterations, μ represents the step size controlling the optimization of the boundary point position of the mosaic line in each iteration, and p (k) and p (k +1) Let ΔE represent the positions of point p in the k-th and (k+1)-th iterations, respectively. total (p (k) ) represents the energy function E total At point p (k) The gradient at that point.

9. The three-dimensional incremental model update method based on joint raster graph and boundary optimization according to claim 1, characterized in that, The merging of the existing model triangular mesh portion and the incremental model triangular mesh portion includes: The triangular mesh area traversed by the optimized mosaic line is reconstructed by triangulation.

10. A three-dimensional incremental model update device combining raster graph and boundary optimization, characterized in that, include: The model generation unit, registration unit, overlapping region calculation unit, contour determination unit, contour optimization unit, and update unit include: The model generation unit is used to acquire local scene images for local scenes to be updated in the target scene, generate corresponding three-dimensional incremental models, and update the existing three-dimensional model of the target scene. The registration unit is used to set a plane as a raster map on the spatial coordinate system where the incremental model and the existing model reside, and to map the point clouds of the incremental model and the existing model to the raster map respectively, to obtain the corresponding incremental point set and existing point set on the raster map respectively, to cluster the point set to obtain incremental layered line clusters and existing layered line clusters, to register the incremental layered line clusters and existing layered line clusters to obtain the corresponding registration relationship, and to perform fine registration of the incremental model and the existing model based on the registration relationship; The overlapping region calculation unit is used to set corresponding incremental bounding boxes and existing bounding boxes for the incremental model and the existing model, and to discretize the bounding boxes into multiple voxel grids. Based on the point cloud labels in the voxel grids, the incremental point cloud index and existing point cloud index of the bounding box voxel grids are obtained respectively. The intersection between the incremental point cloud index and the existing point cloud index is calculated. The voxel grid in the intersection is the overlapping region to be updated. The contour determination unit is used to eliminate the elevation dimension of the point cloud in the overlapping area to be updated, and project it onto a two-dimensional plane to obtain a two-dimensional point set. The two-dimensional point set is then triangulated to construct a triangular mesh, and an initial contour is formed along the boundary of the triangular mesh. The contour optimization unit is used to optimize the initial contour to obtain an optimized mosaic line that meets the requirements of smoothness and geometric consistency. The update unit is used to trim the triangular mesh of the incremental model and the triangular mesh of the existing model using optimized mosaic lines, retaining the triangular mesh portion of the existing model outside the overlapping region to be updated, and retaining the triangular mesh portion of the incremental model inside the overlapping region to be updated, merging the triangular mesh portion of the existing model and the triangular mesh portion of the incremental model to complete the update of the existing model.

Images