Gaussian sputtering orthographic image generation method, device, equipment, medium and product
By using a Gaussian rendering model trained with an anisotropic penalty term and 3D Gaussian sputtering technology, the problems of high computational cost and distortion in generating high-precision orthophotos in existing technologies are solved, achieving higher fidelity and geometric fidelity.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUANGZHOU URBAN PLANNING & DESIGN SURVEY RES INST
- Filing Date
- 2026-02-02
- Publication Date
- 2026-06-05
Smart Images
Figure CN122156419A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of three-dimensional reconstruction technology, and in particular to a method, apparatus, device, medium, and product for generating orthophotos by Gaussian sputtering. Background Technology
[0002] Orthophotos are widely used in surveying, agricultural monitoring, urban planning, and disaster emergency response. They can also serve as base maps for Geographic Information Systems (GIS), overlaid with line maps and annotations to form image maps, or combined with DEMs to generate 3D landscapes, thereby improving urban management efficiency and decision support capabilities.
[0003] Currently, existing orthophoto generation methods can be divided into four main categories: satellite image processing, aerial photogrammetry, UAV aerial surveying, and oblique photogrammetry modeling. Satellite image processing, aerial photogrammetry, and UAV aerial surveying all require geometric correction and digital elevation models (DEM / DSM) for terrain compensation. The accuracy of the DEM directly affects the quality of the generated orthophoto. Generating a high-precision DEM and performing orthophoto correction drastically increases the computational load and production costs. Oblique photogrammetry modeling is limited by the distortion and deformation inherent in oblique models compared to the actual scene, which can affect the quality of the generated orthophoto to some extent. Summary of the Invention
[0004] This invention provides a method, apparatus, device, medium, and product for generating orthophotos by Gaussian sputtering. By introducing an anisotropic Gaussian rendering model trained with an anisotropic penalty term, and combining air-ground joint imagery for Gaussian rendering, an orthophoto of the target area is generated. This method can fuse multi-view image data, improve the fidelity of the target area, and suppress horizontal stretching through the anisotropic penalty term to reduce distortion and improve geometric fidelity.
[0005] To achieve the above objectives, embodiments of the present invention provide a method for generating orthophotos by Gaussian sputtering, comprising: Acquire aerial and ground images of the target area, and construct an air-ground joint dense point cloud of the target area based on the aerial and ground images; Based on the air-ground joint dense point cloud and the trained anisotropic Gaussian rendering model, a Gaussian rendering scene of the target area is generated; wherein, the trained anisotropic Gaussian rendering model is obtained by 3D Gaussian training by introducing anisotropic penalty terms and then using a phased optimization method. Based on the Gaussian rendering scene and the viewing range of the target area, 3D Gaussian sputtering technology is used for rendering to generate an orthophoto image within the viewing range of the target area.
[0006] As an improvement to the above scheme, the trained anisotropic Gaussian rendering model is obtained by performing 3D Gaussian training through a staged optimization method after introducing anisotropic penalty terms, including: Obtain the Gaussian distribution set of the air-ground joint dense point cloud, decompose the covariance matrix of each Gaussian distribution in the Gaussian distribution set, and obtain the scaling degree of each Gaussian distribution. Based on the scaling degree, an anisotropy penalty term is determined for each Gaussian distribution; Based on the anisotropic penalty term, a phased optimization method is used to perform 3D Gaussian training to obtain the trained anisotropic Gaussian rendering model.
[0007] As an improvement to the above scheme, the step of performing 3D Gaussian training using a phased optimization approach based on the anisotropic penalty term to obtain the trained anisotropic Gaussian rendering model includes: Set initial values for the penalty weight coefficients of the anisotropic penalty term to determine the total loss function for 3D Gaussian training; 3D Gaussian training is performed based on the total loss function until the total loss function is less than a preset loss threshold. Then, the total loss function is updated according to a dynamic weight adjustment strategy. The updated total loss function is used for 3D Gaussian training until the model converges, resulting in a trained anisotropic Gaussian rendering model.
[0008] As an improvement to the above solution, the step of rendering using 3D Gaussian sputtering technology based on the Gaussian rendering scene and the viewing angle range of the target area to generate an orthophoto image within the viewing angle range of the target area includes: Set the image parameters of the orthophoto of the target area, and determine the viewing angle range of the target area based on the image parameters; The Gaussian rendering scene is converted into Gaussian sputtering points of the Gaussian rendering scene using 3D Gaussian sputtering technology. Orthophoto rendering is performed on the Gaussian sputtering points within the specified viewing angle range to generate an orthophoto of the target area within the specified viewing angle range.
[0009] As an improvement to the above solution, the step of performing orthophoto rendering on the Gaussian sputtering points within the viewing angle range to generate an orthophoto of the target area within the viewing angle range includes: The Gaussian center coordinates of the Gaussian sputtering points within the specified viewing angle range are transformed into planar pixels using the orthogonal projection matrix of the specified viewing angle range. The color and opacity of planar pixels are calculated using spherical harmonic functions to generate an orthophoto of the target region within its spectral range.
[0010] As an improvement to the above scheme, the step of acquiring aerial and ground imagery of the target area, and constructing a joint air-ground dense point cloud of the target area based on the aerial and ground imagery, includes: Acquire aerial and ground images of the target area, and spatially align the aerial and ground images to obtain a combined air-ground image; A sparse point cloud model of the target area is constructed based on the air-ground joint imagery, and the sparse point cloud model is transformed into a geographic coordinate system. Using multi-view stereo vision technology, a combined air-ground dense point cloud of the target area is constructed based on the converted sparse point cloud model.
[0011] To achieve the above objectives, embodiments of the present invention provide an orthophoto generation apparatus for Gaussian sputtering, comprising: The dense point cloud construction module is used to acquire aerial and ground images of the target area and construct an air-ground joint dense point cloud of the target area based on the aerial and ground images. The rendering scene generation module is used to generate a Gaussian rendering scene of the target area based on the air-ground joint dense point cloud and the trained anisotropic Gaussian rendering model; wherein, the trained anisotropic Gaussian rendering model is obtained by 3D Gaussian training by introducing anisotropic penalty terms and then using a phased optimization method. The orthophoto generation module is used to generate an orthophoto of the target area within its field of view by rendering the Gaussian rendering scene and the view range of the target area using 3D Gaussian sputtering technology.
[0012] To achieve the above objectives, embodiments of the present invention provide an orthophoto generation device for Gaussian sputtering, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the above-described orthophoto generation method for Gaussian sputtering.
[0013] To achieve the above objectives, embodiments of the present invention also provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform the above-described method for generating orthophotos by Gaussian sputtering.
[0014] To achieve the above objectives, embodiments of the present invention also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the above-described method for generating orthophotos by Gaussian sputtering.
[0015] Compared with existing technologies, the present invention discloses a method, apparatus, device, medium, and product for generating orthorectified images by Gaussian sputtering. This involves acquiring aerial and ground images of a target area, constructing a joint air-ground dense point cloud of the target area based on the aerial and ground images, generating a Gaussian rendered scene of the target area based on the joint air-ground dense point cloud and a trained anisotropic Gaussian rendering model, wherein the trained anisotropic Gaussian rendering model is obtained through 3D Gaussian training using a phased optimization method after introducing anisotropic penalty terms, and rendering the orthorectified image within the spectral range of the target area using 3D Gaussian sputtering technology based on the Gaussian rendered scene and the spectral range of the target area. It can integrate multi-view image data and combine aerial and ground images to supplement missing perspectives, solve the problem of detail loss caused by occlusion and shadows, generate more complete orthophotos, improve the restoration of target areas, and realistically restore the scene through an optimizable Gaussian ellipsoid set to avoid model distortion and improve detail accuracy. By suppressing horizontal stretching through anisotropic penalty terms, it ensures that the Gaussian distribution fits the geometric requirements of vertical projection, effectively reducing building facade artifacts and roof texture deformation, and improving geometric fidelity. Attached Figure Description
[0016] Figure 1 This is a flowchart illustrating a method for generating orthophotos by Gaussian sputtering according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of an orthophoto generation device for Gaussian sputtering provided in an embodiment of the present invention; Figure 3 This is a structural block diagram of an orthophoto generation device using Gaussian sputtering, provided in an embodiment of the present invention. Detailed Implementation
[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0018] It should be noted that the terms "comprising" and "specific" in this invention, and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such process, method, product, or device.
[0019] Please see Figure 1 , Figure 1This is a flowchart illustrating a method for generating orthophotos from Gaussian sputtering according to an embodiment of the present invention. The method includes: S1, acquire aerial and ground images of the target area, and construct an air-ground joint dense point cloud of the target area based on the aerial and ground images; S2, based on the air-ground joint dense point cloud and the trained anisotropic Gaussian rendering model, generate a Gaussian rendering scene of the target area; wherein, the trained anisotropic Gaussian rendering model is obtained by 3D Gaussian training by introducing anisotropic penalty terms and then using a phased optimization method. S3, based on the Gaussian rendering scene and the viewing range of the target area, 3D Gaussian sputtering technology is used for rendering to generate an orthophoto image within the viewing range of the target area.
[0020] For example, the orthorectified image generation method using Gaussian sputtering described in this embodiment of the invention can be implemented by an orthorectified image generation server, which is capable of interacting with the target user. The orthorectified image generation server acquires aerial and ground images of the target area, and constructs an air-ground joint dense point cloud of the target area based on the aerial and ground images. Based on the air-ground joint dense point cloud and a trained anisotropic Gaussian rendering model, a Gaussian rendering scene of the target area is generated. The trained anisotropic Gaussian rendering model is obtained through 3D Gaussian training using a phased optimization method after introducing anisotropic penalty terms. Based on the Gaussian rendering scene and the viewing angle range of the target area, 3D Gaussian sputtering technology is used for rendering to generate an orthorectified image within the viewing angle range of the target area. This invention employs 3D Gaussian sputtering technology to realistically reconstruct scenes, generating orthophotos from a given perspective. This results in a scene model with higher fidelity and more accurate details, enabling the rendering of orthophotos with higher precision. By combining aerial and ground imagery for joint 3D Gaussian scene construction, the missing perspectives and details in aerial imagery are effectively supplemented, thereby generating orthophotos with more complete details. Anisotropic constraints suppress excessively stretched Gaussian distributions in the horizontal direction, effectively avoiding problems such as building facade artifacts and roof texture distortion caused by excessive horizontal covariance, thus improving the geometric fidelity of vertical projection.
[0021] Specifically, the trained anisotropic Gaussian rendering model is obtained through 3D Gaussian training using a staged optimization approach after introducing anisotropic penalty terms, including: S201, Obtain the Gaussian distribution set of the air-ground joint dense point cloud, decompose the covariance matrix of each Gaussian distribution in the Gaussian distribution set, and obtain the scaling degree of each Gaussian distribution. S202, Based on the scaling degree, determine the anisotropy penalty term for each Gaussian distribution; S203, based on the anisotropic penalty term, 3D Gaussian training is performed using a phased optimization method to obtain the trained anisotropic Gaussian rendering model.
[0022] For example, the aerial and ground imagery of the target area and its pose, along with the joint aerial and ground dense point cloud, are used as training data for 3D Gaussian training. The joint aerial and ground dense point cloud provides a good initialization for the 3D Gaussian distribution, with each 3D point initialized as a Gaussian ellipsoid (3D Gaussian distribution). Each 3D Gaussian distribution is defined by the following parameters: mean (location). Covariance matrix Opacity These Gaussian distributions can be represented as: , In the formula, For the first in 3D space The probability density of the Gaussian distribution at each point. It is a natural constant. This is the transpose symbol.
[0023] The input for training 3D Gaussian Sputtering (3DGS) includes a set of images of a static scene along with their corresponding camera intrinsic and extrinsic parameters and point cloud data of the scene. During 3DGS training, parameters optimized include position... Covariance matrix Opacity In addition, it includes spherical harmonics (SH) coefficients to represent color. After training, it can be used to render images from new perspectives. To generate better orthophotos, an anisotropic penalty term is introduced in the 3D Gaussian training to suppress the excessively stretched Gaussian distribution in the horizontal direction, making it more consistent with the vertical projection geometry requirements of orthophotos. This operation can effectively avoid problems such as building facade artifacts and roof texture distortion caused by excessive horizontal covariance, thus improving the geometric fidelity of vertical projection.
[0024] For example, in a 3D Gaussian, the covariance matrix The geometry of the Gaussian distribution is described. The covariance matrix for each 3D Gaussian distribution is given. Perform orthogonal decomposition to obtain the covariance matrix of each 3D Gaussian distribution. The corresponding rotation matrix and diagonal matrix are used to determine the scaling (stretching or compression) degree of each Gaussian distribution based on the eigenvalues in the diagonal matrix. Analyzing the eigenvalue ratio can quantify the degree of anisotropy and determine the anisotropy penalty term for each Gaussian distribution. For example, the eigenvalue ratio is compared with the eigenvalue threshold (which can be set as needed) to determine whether the Gaussian is overstretched in the horizontal direction. Based on the eigenvalues, a phased optimization method is used to perform 3D Gaussian training to obtain the trained anisotropic Gaussian rendering model.
[0025] Wherein, for each 3D Gaussian distribution, the covariance matrix The expression for orthogonal decomposition is: , In the formula, For rotation matrix, It is a diagonal matrix. This is the transpose symbol.
[0026] The expression for the diagonal matrix is: , In the formula, These represent the variances of the Gaussian distribution along the three principal axes (x, y, and z), indicating the degree of scaling (stretching or compression) along each axis. Off-diagonal elements are 0, indicating no covariance correlation between directions. Analyze the eigenvalues. The proportion can quantify the degree of anisotropy.
[0027] Eigenvalue ratio The calculation formula is: , Assuming the feature value threshold is ,like If so, the Gaussian distribution is determined to be excessively stretched in the horizontal direction. It is usually set to 3 or 5.
[0028] More specifically, step S203 includes: S2031, Set the initial value of the penalty weight coefficient of the anisotropic penalty term to determine the total loss function of 3D Gaussian training; S2032, Perform 3D Gaussian training based on the total loss function until the total loss function is less than a preset loss threshold. Update the total loss function according to a dynamic weight adjustment strategy and use the updated total loss function to perform 3D Gaussian training until the model converges, thus obtaining the trained anisotropic Gaussian rendering model.
[0029] For example, an anisotropic penalty term (energy function) is introduced to suppress horizontal covariance by introducing a penalty weight coefficient. An initial value is set for the penalty weight coefficient of this anisotropic penalty term. This penalty term forces the principal axis of the Gaussian to align with the vertical axis, ensuring that the projected geometry meets the requirements of orthophotos. This penalty term is added to the loss function of 3D Gaussian training to determine the total loss function of 3D Gaussian training. A phased optimization approach is adopted when training the 3D Gaussian. In the initial stage, a small penalty weight coefficient (the initial value of the penalty weight coefficient is usually 0.01-0.1) is used to constrain the distribution of the Gaussian, allowing the Gaussian to optimize in a direction that better fits the scene surface, prioritizing global reconstruction accuracy. In the later stages of training (until the total loss function is less than a preset loss threshold), the penalty weight coefficient can be adjusted adaptively (using a dynamic weight adjustment strategy). Adjustments are made to obtain updated penalty weight coefficients, which are then used to update the total loss function. 3D Gaussian training is performed using the updated total loss function until the model converges, resulting in a trained anisotropic Gaussian rendering model. It can be understood that the penalty weight coefficients... Adjustments can be made to dynamically increase the penalty weights according to the degree of anisotropy, specifically addressing severely stretched Gaussians and strengthening anisotropic constraints. Based on the above settings, a 3D Gaussian model is trained on a given scene to obtain a trained anisotropic Gaussian rendering model.
[0030] Wherein, the energy function The formula is: , In the formula, For the penalty weighting coefficient, It is a local minimum value, typically taking the value of 1. This is used to prevent the denominator from being zero.
[0031] The total loss function The expression is: , In the formula, It is a rendering loss function, which is commonly used in 3D Gaussian rendering tasks to measure the difference between the rendered image and the real image.
[0032] The penalty weight coefficient The following formula can be used for adaptive adjustment: , In the formula, The updated penalty weight coefficient. This is the initial penalty weight coefficient.
[0033] Specifically, step S3 includes: S31, Set the image parameters of the orthophoto of the target area, and determine the viewing angle range of the target area based on the image parameters; S32, using 3D Gaussian sputtering technology to convert the Gaussian rendering scene into Gaussian sputtering points of the Gaussian rendering scene; S33, orthophoto rendering is performed on the Gaussian sputtering points within the viewing angle range to generate an orthophoto of the target area within the viewing angle range.
[0034] More specifically, step S33 includes: S331, the Gaussian center coordinates of the Gaussian sputtering points within the viewing angle range are transformed into planar pixels through the orthogonal projection matrix of the viewing angle range; S332, use spherical harmonic functions to calculate the color and opacity of planar pixels to generate an orthophoto of the target region within its spectral range.
[0035] For example, orthographic projection is a vertical projection, with projected rays parallel and perpendicular to the ground (z-axis aligned). Image parameters of the orthographic image of the target area are set, including width in pixels, height in pixels, ground resolution, and geographic coordinates of the orthographic image center point. The viewing angle range of the target area is determined based on these image parameters. A 3D Gaussian sputtering technique is used to convert the Gaussian rendered scene into Gaussian sputtering points within the Gaussian rendered scene. The Gaussian center coordinates of the Gaussian sputtering points within the viewing angle range are transformed into planar pixels using the orthogonal projection matrix of the viewing angle range. A spherical harmonic function is used to calculate the color and opacity of the planar pixels, generating an orthographic image within the viewing angle range of the target area. For example, the image width... Pixels, High Pixels, ground resolution (meters / pixel), and the geographic coordinates of the center point of the orthophoto. The coordinates of the upper left corner of the orthophoto are: and the coordinates of the bottom right corner are The field of view of the target area is obtained as follows: , So, what orthogonal projection matrix is needed for orthophoto rendering? for: , In the formula, and The near and far clipping planes in the vertical direction of the scene.
[0036] Using the horizontal center position to satisfy , A Gaussian ellipsoid is used for rendering orthophotos. The coordinates of the Gaussian center of all elements involved in the rendering are... Through orthogonal projection matrix Transformed into 2D image plane coordinates, i.e., plane pixels The transformation formula is: , During rendering, calculations are performed sequentially from closest to farthest from the 3D Gaussian source, using spherical harmonic functions to simultaneously calculate the color of the current pixel. and opacity The rendering formula is: , , In the formula, and It represents the color and opacity of the current pixel. and It's the updated color and opacity. When the opacity... If this happens, the calculation of the pixel's color will stop. This is the final color value of the pixel. This is the opacity threshold, which is usually set to a value close to 1, such as 0.99 or 0.999.
[0037] Specifically, step S1 includes: S11, acquire aerial and ground images of the target area, spatially align the aerial and ground images to obtain a combined air-ground image; S12, construct a sparse point cloud model of the target area based on the air-ground joint imagery, and convert the sparse point cloud model to a geographic coordinate system; S13. Using multi-view stereo vision technology, a combined air-ground dense point cloud of the target area is constructed based on the converted sparse point cloud model.
[0038] For example, acquiring aerial and ground images of a target area requires feature point extraction and matching, followed by spatial alignment to obtain a combined air-ground image. Generating a dense point cloud provides a better initialization for the 3D Gaussian image, facilitating a better fit to the target scene. Aerial images include, but are not limited to, aerial view images taken by aircraft, drones, etc. First, feature points are extracted from all aerial images. These feature points can be traditional handcrafted features, such as SIFT (Scale-Invariant Feature Transform) feature points, or deep learning features, such as Superpoint. After feature extraction, feature point matching between images can be performed based on the similarity of image feature point descriptors to obtain matching relationships. Ground images include, but are not limited to, ground view images taken by cameras, mobile phones, etc. Similar to aerial images, feature point extraction and matching are required for all ground images to obtain matching relationships. Air-ground connection points are added using photogrammetry software. Due to significant differences in perspective, automated feature matching between air-ground images is often difficult to achieve directly through descriptor similarity. However, existing photogrammetry software can easily add matching relationships between aerial and ground images. Specifically, identical ground features in the aerial and ground images can be identified and marked manually. After adding a sufficient number of matching point pairs between aerial and ground images, the matching point pairs between aerial images, ground images, and aerial and ground images can be combined using SfM (Structure from Motion) technology to perform joint positioning and pose determination of the aerial and ground images, obtaining the pose (position and attitude) of all images, and simultaneously generating a sparse point cloud model of the scene. Orthophotos require correct geographic coordinates, so the obtained aerial triangulation model needs to be converted to the required geographic coordinate system. The conversion basis includes, but is not limited to, high-precision control points acquired in the early stage and high-precision image GPS positioning information. The results of SfM can then be used for 3D Gaussian training. However, the sparse point cloud of the scene obtained in this step is usually insufficient to describe the complete structure of the scene due to the small number of points, which will affect the initialization of the 3D Gaussian model and the final training results. Multi-View Stereo (MVS) can be used to obtain denser point clouds with higher density, which can more clearly represent the structure of the scene, provide a better initialization basis for 3D Gaussian mapping, and thus achieve better training results, which helps to render more detailed orthophotos. Therefore, MVS is applied to the aerial triangulation results to construct a combined aerial and ground dense point cloud of the scene.
[0039] This invention discloses a method for generating orthorectified images using Gaussian sputtering. The method involves acquiring aerial and ground images of a target area, constructing a combined air-ground dense point cloud of the target area based on these images, and generating a Gaussian rendered scene of the target area using the combined air-ground dense point cloud and a trained anisotropic Gaussian rendering model. The trained anisotropic Gaussian rendering model is obtained through 3D Gaussian training using a phased optimization approach after introducing an anisotropic penalty term. Based on the Gaussian rendered scene and the viewing angle range of the target area, 3D Gaussian sputtering technology is used for rendering to generate an orthorectified image within the viewing angle range of the target area. This method can fuse multi-view image data, supplement missing views by combining air-ground images, solve the problem of detail loss caused by occlusion and shadows, generate a more complete orthorectified image, improve the target area's fidelity, realistically restore the scene using an optimizable Gaussian ellipsoid set, avoid model distortion, and improve detail accuracy. The anisotropic penalty term suppresses horizontal stretching, ensuring that the Gaussian distribution conforms to the vertical projection geometry requirements, effectively reducing building facade artifacts and roof texture deformation, and improving geometric fidelity.
[0040] See Figure 2 , Figure 2 This is a schematic diagram of the structure of an orthophoto generation apparatus 10 for Gaussian sputtering provided in an embodiment of the present invention. The orthophoto generation apparatus 10 for Gaussian sputtering includes: The dense point cloud construction module 11 is used to acquire aerial and ground images of the target area and construct an air-ground joint dense point cloud of the target area based on the aerial and ground images. The rendering scene generation module 12 is used to generate a Gaussian rendering scene of the target area based on the air-ground joint dense point cloud and the trained anisotropic Gaussian rendering model; wherein, the trained anisotropic Gaussian rendering model is obtained by 3D Gaussian training by introducing anisotropic penalty terms and then using a phased optimization method. The orthophoto generation module 13 is used to generate an orthophoto of the target area by rendering using 3D Gaussian sputtering technology based on the Gaussian rendering scene and the viewing angle range of the target area.
[0041] The Gaussian sputtering orthophoto generation apparatus 10 provided in this embodiment of the invention can realize all the processes of the Gaussian sputtering orthophoto generation method of the above embodiment. The functions and technical effects of each module in the apparatus are the same as those of the Gaussian sputtering orthophoto generation method of the above embodiment, and will not be repeated here.
[0042] See Figure 3 , Figure 3This is a schematic diagram of the structure of a Gaussian sputtering orthophoto generation device 20 provided in an embodiment of the present invention. The Gaussian sputtering orthophoto generation device 20 of this embodiment includes: a processor 21, a memory 22, and a computer program stored in the memory 22 and executable on the processor 21. When the processor 21 executes the computer program, it implements the steps in the above-described Gaussian sputtering orthophoto generation method embodiment. Alternatively, when the processor 21 executes the computer program, it implements the functions of each module in the above-described Gaussian sputtering orthophoto generation apparatus embodiment.
[0043] For example, the computer program may be divided into one or more modules, which are stored in the memory 22 and executed by the processor 21 to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the Gaussian sputtering orthophoto generation device 20.
[0044] The Gaussian sputtering orthophoto generation device 20 can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. The Gaussian sputtering orthophoto generation device 20 may include, but is not limited to, a processor 21 and a memory 22. Those skilled in the art will understand that the schematic diagram is merely an example of the Gaussian sputtering orthophoto generation device 20 and does not constitute a limitation on the Gaussian sputtering orthophoto generation device 20. It may include more or fewer components than illustrated, or combine certain components, or different components. For example, the Gaussian sputtering orthophoto generation device 20 may also include input / output devices, network access devices, buses, etc.
[0045] The processor 21 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor. The processor 21 is the control center of the Gaussian sputtering orthophoto generation device 20, connecting all parts of the Gaussian sputtering orthophoto generation device 20 via various interfaces and lines.
[0046] The memory 22 can be used to store the computer program and / or modules. The processor 21 implements various functions of the Gaussian sputtering orthophoto generation device 20 by running or executing the computer program and / or modules stored in the memory 22 and calling the data stored in the memory 22. The memory 22 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phone book, etc.). In addition, the memory 22 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.
[0047] The module integrated into the Gaussian sputtering orthophoto generation device 20, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by the processor 21, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content contained in the computer-readable medium may be appropriately added to or subtracted from the content as required by the legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium may not include electrical carrier signals and telecommunication signals.
[0048] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.
[0049] This invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the Gaussian sputtering orthophoto generation method as described above.
[0050] Furthermore, embodiments of the present invention also provide a computer program product, which is stored in a storage medium and executed by at least one processor to implement the steps of the Gaussian sputtering orthophoto generation method of the above embodiments.
[0051] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.
Claims
1. A method for generating orthophotos by Gaussian sputtering, characterized in that, include: Acquire aerial and ground images of the target area, and construct an air-ground joint dense point cloud of the target area based on the aerial and ground images; Based on the air-ground joint dense point cloud and the trained anisotropic Gaussian rendering model, a Gaussian rendering scene of the target area is generated; wherein, the trained anisotropic Gaussian rendering model is obtained by 3D Gaussian training by introducing anisotropic penalty terms and then using a phased optimization method. Based on the Gaussian rendering scene and the viewing range of the target area, 3D Gaussian sputtering technology is used for rendering to generate an orthophoto image within the viewing range of the target area.
2. The method for generating orthophotos by Gaussian sputtering as described in claim 1, characterized in that, The trained anisotropic Gaussian rendering model is obtained through 3D Gaussian training using a staged optimization approach after introducing anisotropic penalty terms, including: Obtain the Gaussian distribution set of the air-ground joint dense point cloud, decompose the covariance matrix of each Gaussian distribution in the Gaussian distribution set, and obtain the scaling degree of each Gaussian distribution. Based on the scaling degree, an anisotropy penalty term is determined for each Gaussian distribution; Based on the anisotropic penalty term, a phased optimization method is used to perform 3D Gaussian training to obtain the trained anisotropic Gaussian rendering model.
3. The method for generating orthophotos by Gaussian sputtering as described in claim 2, characterized in that, The step of performing 3D Gaussian training using a phased optimization approach based on the anisotropic penalty term to obtain the trained anisotropic Gaussian rendering model includes: Set initial values for the penalty weight coefficients of the anisotropic penalty term to determine the total loss function for 3D Gaussian training; 3D Gaussian training is performed based on the total loss function until the total loss function is less than a preset loss threshold. Then, the total loss function is updated according to a dynamic weight adjustment strategy. The updated total loss function is used for 3D Gaussian training until the model converges, resulting in a trained anisotropic Gaussian rendering model.
4. The method for generating orthophotos by Gaussian sputtering as described in claim 1, characterized in that, The step of rendering the target area using 3D Gaussian sputtering technology based on the Gaussian rendering scene and the viewpoint range of the target area to generate an orthophoto image within the viewpoint range of the target area includes: Set the image parameters of the orthophoto of the target area, and determine the viewing angle range of the target area based on the image parameters; The Gaussian rendering scene is converted into Gaussian sputtering points of the Gaussian rendering scene using 3D Gaussian sputtering technology. Orthophoto rendering is performed on the Gaussian sputtering points within the specified viewing angle range to generate an orthophoto of the target area within the specified viewing angle range.
5. The method for generating orthophotos by Gaussian sputtering as described in claim 4, characterized in that, The step of rendering orthorectified images of the Gaussian sputtering points within the viewing angle range to generate an orthorectified image of the target region within the viewing angle range includes: The Gaussian center coordinates of the Gaussian sputtering points within the specified viewing angle range are transformed into planar pixels using the orthogonal projection matrix of the specified viewing angle range. The color and opacity of planar pixels are calculated using spherical harmonic functions to generate an orthophoto of the target region within its spectral range.
6. The method for generating orthophotos by Gaussian sputtering as described in claim 1, characterized in that, The acquisition of aerial and ground imagery of the target area, and the construction of a joint air-ground dense point cloud of the target area based on the aerial and ground imagery, includes: Acquire aerial and ground images of the target area, and spatially align the aerial and ground images to obtain a combined air-ground image; A sparse point cloud model of the target area is constructed based on the air-ground joint imagery, and the sparse point cloud model is transformed into a geographic coordinate system. Using multi-view stereo vision technology, a combined air-ground dense point cloud of the target area is constructed based on the converted sparse point cloud model.
7. An apparatus for generating orthophotos by Gaussian sputtering, characterized in that, include: The dense point cloud construction module is used to acquire aerial and ground images of the target area and construct an air-ground joint dense point cloud of the target area based on the aerial and ground images. The rendering scene generation module is used to generate a Gaussian rendering scene of the target area based on the air-ground joint dense point cloud and the trained anisotropic Gaussian rendering model; wherein, the trained anisotropic Gaussian rendering model is obtained by 3D Gaussian training by introducing anisotropic penalty terms and then using a phased optimization method. The orthophoto generation module is used to generate an orthophoto of the target area within its field of view by rendering the Gaussian rendering scene and the view range of the target area using 3D Gaussian sputtering technology.
8. An orthophoto generation apparatus for Gaussian sputtering, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the method for generating orthophotos by Gaussian sputtering as described in any one of claims 1-6.
9. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the Gaussian sputtering orthophoto generation method as described in any one of claims 1-6.
10. A computer program product, characterized in that, The computer program product is stored in a storage medium and is executed by at least one processor to implement the steps of the Gaussian sputtering orthophoto generation method as described in any one of claims 1-6.