A multi-source geographic information three-dimensional scene animation construction method, device and medium

Abstract

The application discloses a multi-source geographic information three-dimensional scene animation construction method and device and medium, comprising: in response to the input instruction of multi-source geographic data, the standardized preprocessing is carried out to the obtained multi-source heterogeneous geographic data, and the standardized geographic data set is generated; based on the standardized geographic data set, the neural network three-dimensional scene reconstruction is executed, and the real three-dimensional geographic base scene is generated, wherein the neural network three-dimensional scene reconstruction comprises adopting a multi-view stereo matching network to extract image depth information to generate a dense point cloud, and adopting a Gaussian splatting neural rendering network to perform implicit scene reconstruction on the dense point cloud and apply a geographic coordinate constraint; in response to the three-dimensional scene animation generation instruction, based on the real three-dimensional geographic base scene and the time sequence monitoring geographic data, the space-time evolution of dynamic geographic elements is driven, and the three-dimensional scene animation is output along a preset camera path through a differentiable rendering pipeline.

Description

Technical Field

[0001] This invention discloses a method, apparatus, and medium for constructing three-dimensional scene animations from multi-source geographic information, belonging to the field of geographic information technology. Background Technology

[0002] Traditional 3D scene animation construction in geographic information mainly relies on manual modeling and texture mapping, which is costly, inefficient, and lacks realism. Multi-source heterogeneous geographic data, including satellite remote sensing imagery, aerial oblique photography imagery, DEM topographic data, DOM orthophotos, GIS vector data, laser point clouds, and time-series monitoring data, are difficult to integrate efficiently and align accurately due to inconsistent formats, coordinate bases, and significant differences in spatiotemporal resolution. While existing 3D reconstruction methods based on general neural rendering can generate highly realistic scenes, their reconstruction results lack geographic coordinate system definitions, making them unsuitable for direct integration with the spatial analysis needs of geographic information devices. Furthermore, traditional 3D animation only supports fixed preset path roaming and lacks the ability to drive the dynamic evolution of geographic elements by incorporating time-series monitoring data. This makes it difficult to meet the needs of applications such as digital twins, disaster simulation, and dynamic display of national land space for constructing integrated 3D scene animations that combine static geographic bases and dynamic geographic elements.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This invention provides a method, apparatus, and medium for constructing 3D scene animations from multi-source geographic information, which at least solves the technical problems of difficulty in fusing multi-source heterogeneous geographic data and high cost and low efficiency caused by reliance on manual modeling for 3D scene construction.

[0005] According to one aspect of the present invention, in order to achieve the above objective, a method for constructing a three-dimensional scene animation based on multi-source geographic information is provided, comprising:

[0006] In response to input commands for multi-source geographic data, the acquired multi-source heterogeneous geographic data is standardized and preprocessed to generate a standardized geographic dataset. The multi-source heterogeneous geographic data includes satellite remote sensing images, aerial oblique photogrammetry images, DEM topographic data, DOM orthophotos, GIS vector data, laser point cloud data, and time-series monitoring geographic data.

[0007] Based on a standardized geographic dataset, a neural network 3D scene reconstruction is performed to generate a real-world 3D geographic base scene. The neural network 3D scene reconstruction includes using a multi-view stereo matching network to extract image depth information to generate a dense point cloud, and using a Gaussian splash neural rendering network to perform implicit scene reconstruction on the dense point cloud and apply geographic coordinate constraints.

[0008] Responding to the command to generate 3D scene animation, based on the real-world 3D geographic base scene and time-series monitoring geographic data, it drives the spatiotemporal evolution of dynamic geographic elements and outputs 3D scene animation along the preset camera path through the differentiable rendering pipeline.

[0009] Furthermore, the acquired multi-source heterogeneous geographic data undergoes standardization preprocessing, including:

[0010] Perform unified spatial coordinate registration on multi-source heterogeneous geographic data to transform each data source to the target coordinate system;

[0011] Perform image denoising, point cloud denoising, and missing data imputation on multi-source heterogeneous geographic data;

[0012] Align multi-source heterogeneous geographic data that have undergone unified spatial coordinate registration and processing under a unified spatiotemporal reference.

[0013] Furthermore, a multi-view stereo matching network is used to extract image depth information to generate a dense point cloud, including:

[0014] Input multi-view remote sensing images and aerial oblique photography images from standardized geographic datasets into a multi-view stereo matching network;

[0015] Global features of images from various perspectives are extracted using a multi-view stereo matching network, and corresponding feature points are matched between images.

[0016] Based on the matching results, the scene depth information is solved, and a dense point cloud that fits the real terrain and features is output.

[0017] Furthermore, a Gaussian splash neural rendering network is used to perform implicit scene reconstruction of dense point clouds and impose geographic coordinate constraints, including:

[0018] Initialize the spatial position, scale, and color parameters of three-dimensional Gaussian particles based on dense point clouds;

[0019] Using DEM terrain data as elevation constraints, the vertical distribution range of three-dimensional Gaussian particles is constrained.

[0020] Using the boundaries of ground features in GIS vector data as horizontal constraints, the distribution density of three-dimensional Gaussian particles on both sides of the ground feature boundaries is constrained.

[0021] Construct a geographic coordinate constraint loss function, which includes a spatial location loss term and a feature boundary consistency loss term;

[0022] Through iterative optimization, the geographic coordinate constraint loss function is minimized, so that the deviation between the spatial position of the 3D Gaussian particle and the real geographic coordinates is less than or equal to the preset geographic accuracy threshold, thereby generating a real-world 3D geographic base scene with geographic locationability.

[0023] Furthermore, the spatiotemporal evolution of dynamic geographic elements includes:

[0024] Overlaying and fusing static geographic features from GIS vector data into a real-world 3D geographic base scene;

[0025] A dynamic evolution model is constructed based on time-series monitoring geographic data to perform time-series interpolation fitting on geographic dynamic processes;

[0026] By integrating static geographic features with the evolution results of dynamic geographic processes, a static-dynamic fusion scene is generated.

[0027] Furthermore, the time-series monitoring geographic data includes at least one of hydrological monitoring data, vegetation monitoring data, traffic flow monitoring data, and disaster evolution monitoring data; the geographic dynamic processes include at least one of flood spread processes, vegetation growth and decline processes, traffic flow processes, and disaster diffusion processes.

[0028] Furthermore, a 3D scene animation is output along a preset camera path via a differentiable rendering pipeline, including:

[0029] Obtain user-defined camera motion keyframes and roaming path parameters;

[0030] Smoothly interpolate the pose between camera motion keyframes to generate continuous viewpoint motion trajectories;

[0031] Along a continuous viewpoint motion trajectory, the real-world 3D geographic base scene and dynamic geographic elements are rendered frame by frame through a differentiable rendering pipeline to generate a 3D scene animation.

[0032] Furthermore, when rendering frame by frame along the motion trajectory of a continuous viewpoint, it also includes:

[0033] Determine the distance between scene features and the viewpoint in the current rendering frame;

[0034] Based on distance, different rendering precision is applied to distant and near-field areas, with the rendering precision of near-field areas being higher than that of distant areas.

[0035] According to one embodiment of the present invention, a multi-source geographic information 3D scene animation construction device is also provided, comprising: a processing module, configured to, in response to input instructions for multi-source geographic data, perform standardized preprocessing on the acquired multi-source heterogeneous geographic data to generate a standardized geographic dataset, wherein the multi-source heterogeneous geographic data includes satellite remote sensing imagery, aerial oblique photography imagery, DEM topographic data, DOM orthophoto imagery, GIS vector data, laser point cloud data, and time-series monitoring geographic data; a reconstruction module, configured to, based on the standardized geographic dataset, perform neural network 3D scene reconstruction to generate a real-world 3D geographic base scene, wherein the neural network 3D scene reconstruction includes using a multi-view stereo matching network to extract image depth information to generate dense point clouds, and using a Gaussian splash neural rendering network to perform implicit scene reconstruction on the dense point clouds and apply geographic coordinate constraints; and a generation module, configured to, in response to 3D scene animation generation instructions, drive the spatiotemporal evolution of dynamic geographic elements based on the real-world 3D geographic base scene and the time-series monitoring geographic data, and output a 3D scene animation along a preset camera path through a differentiable rendering pipeline.

[0036] According to another aspect of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the multi-source geographic information three-dimensional scene animation construction method of various embodiments of the present invention.

[0037] In this embodiment of the invention, standardized preprocessing of multi-source heterogeneous geographic data enables the unified spatial benchmark and efficient fusion of satellite remote sensing, aerial imagery, DEM, DOM, GIS vector, laser point cloud, and time-series monitoring data. A dual-network collaborative architecture of multi-view stereo matching network and Gaussian splash neural rendering network enables the automated construction of realistic 3D scenes without manual fine modeling. Geographic coordinate constraint loss function is used to endow the scene with geographic locationability to connect with GIS devices. Time-series monitoring data drives the evolution of dynamic geographic elements and combines differentiable rendering pipelines to output free-view continuous animation along preset camera paths. Finally, the automated construction of multi-source geographic information 3D scene animation with high realism, high geographic accuracy, and dynamic evolution is achieved. Attached Figure Description

[0038] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0039] Figure 1 This is a flowchart of a method for constructing a three-dimensional scene animation based on multi-source geographic information according to one embodiment of the present invention;

[0040] Figure 2This is a structural block diagram of a multi-source geographic information 3D scene animation construction device according to one embodiment of the present invention. Detailed Implementation

[0041] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. 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 should fall within the scope of protection of the present invention.

[0042] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0043] According to an embodiment of the present invention, a method for constructing a three-dimensional scene animation based on multi-source geographic information is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer device such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0044] This method embodiment can be executed in an electronic device or similar computing device that includes a memory and a processor. Taking data processing as an example, the data processing may include one or more processors (processors may include, but are not limited to, central processing units (CPUs), graphics processing units (GPUs), digital signal processing (DSP) chips, microcontroller units (MCUs), field-programmable gate arrays (FPGAs), neural network processors (NPUs), tensor processors (TPUs), artificial intelligence (AI) type processors, etc.) and a memory for storing data. Optionally, the above data processing may also include transmission devices, input / output devices, and display devices for communication functions. Those skilled in the art will understand that the above structural description is merely illustrative and does not limit the structure of the above data processing. For example, the data processing may include more or fewer components than described above, or have a different configuration than described above.

[0045] The memory can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the multi-source geographic information 3D scene animation construction method in this embodiment of the invention. The processor executes various functional applications and data processing by running the computer program stored in the memory, thereby realizing the aforementioned multi-source geographic information 3D scene animation construction method. The memory may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include memory remotely located relative to the processor, and these remote memories can be connected to the mobile terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0046] The transmission device is used to receive or send data via a network. Specific examples of the network mentioned above may include a wireless network provided by the mobile terminal's communication provider. In one example, the transmission device includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device may be a Radio Frequency (RF) module, used for wireless communication with the Internet.

[0047] Display devices can be, for example, touchscreen liquid crystal displays (LCDs) and touch displays (also referred to as "touchscreens" or "touch displays"). The LCD allows users to interact with the user interface of the mobile terminal. In some embodiments, the mobile terminal has a graphical user interface (GUI), which allows users to interact with the GUI through finger contact and / or gestures on a touch-sensitive surface. Optional human-computer interaction functions include: creating web pages, drawing, word processing, creating electronic documents, playing games, video conferencing, instant messaging, sending and receiving emails, call interfaces, playing digital video, playing digital music, and / or web browsing, etc. Executable instructions for performing the above human-computer interaction functions are configured / stored in one or more processor-executable computer program products or readable storage media.

[0048] Figure 1 According to one embodiment of the present invention, a method for constructing a three-dimensional scene animation using multi-source geographic information includes:

[0049] Step S110: In response to the input command for multi-source geographic data, the acquired multi-source heterogeneous geographic data is subjected to standardization preprocessing to generate a standardized geographic dataset. The multi-source heterogeneous geographic data includes satellite remote sensing imagery, aerial oblique photogrammetry imagery, DEM topographic data, DOM orthophoto imagery, GIS vector data, laser point cloud data, and time-series monitoring geographic data, the specific contents of which are as follows:

[0050] In step S110, firstly, in response to the input command of multi-source geographic data issued by the user or the host device, the 3D scene animation construction process is initiated. After receiving the input command, the device control device begins to execute the acquisition and standardization preprocessing operations of multi-source heterogeneous geographic data.

[0051] The multi-source heterogeneous geographic data includes at least the following seven data sources: satellite remote sensing imagery, aerial oblique photogrammetry imagery, DEM topographic data (Digital Elevation Model data), DOM orthophoto imagery (Digital Orthophoto Map data), GIS vector data, laser point cloud data, and time-series monitoring geographic data. Specifically, satellite remote sensing imagery provides large-scale land cover and spectral texture information; aerial oblique photogrammetry imagery provides multi-view, high-resolution side texture and structural information of land features; DEM topographic data provides surface elevation undulation information for the target area, serving as a crucial constraint for subsequent 3D scene geometric construction; DOM orthophoto imagery provides geometrically corrected orthophoto texture base maps with a unified scale; GIS vector data provides linear and area features with clear geographic semantics, such as roads, water systems, building boundaries, and pipe networks; laser point cloud data provides high-precision 3D geometric sampling information of the land surface; and time-series monitoring geographic data provides process data reflecting the dynamic changes of geographic features over time.

[0052] After acquiring the aforementioned multi-source heterogeneous geographic data, the first step is to perform a unified spatial coordinate registration operation on all types of data. Since geographic data acquired by different sensors and acquisition platforms may use different spatial reference systems—for example, some satellite imagery uses the WGS84 coordinate system, while aerial imagery and laser point clouds may use the 2000 National Geodetic Coordinate System—it is necessary to uniformly convert all data sources to the target coordinate system. Specifically, a projection transformation method is used to complete the unified registration between the WGS84 coordinate system and the 2000 National Geodetic Coordinate System, ensuring precise spatial alignment between topographic image data and GIS vector feature data.

[0053] Then, targeted preprocessing was performed to address the specific data quality issues of each data source. For satellite remote sensing imagery and aerial oblique photography, filtering algorithms were used to remove noise caused by atmospheric scattering, sensor noise, and other factors. Super-resolution inpainting was performed on low-resolution image areas to improve texture clarity, and missing or cloud-obscured areas were filled with textures using nearby temporal or multi-angle imagery. For laser point cloud data, statistical filtering or radius filtering algorithms were used to remove outliers, and areas with uneven point cloud density were uniformly resampled. For DEM terrain data, interpolation was used to fill in raster cells with holes or anomalies to ensure the continuity and integrity of the terrain surface. For DOM orthophotos, geometric edge consistency was checked, and color equalization was performed.

[0054] Finally, the multi-source heterogeneous geographic data, after undergoing unified spatial coordinate registration and various preprocessing operations, are aligned under a unified spatiotemporal reference to form a standardized geographic dataset. All data in this standardized geographic dataset share a consistent geographic coordinate reference in space and are temporally aligned based on the collection timestamps, ensuring accurate spatial correspondence and feature fusion between various data sources during subsequent neural network 3D reconstruction.

[0055] Step S120: Based on a standardized geographic dataset, perform neural network 3D scene reconstruction to generate a real-world 3D geographic base scene. The neural network 3D scene reconstruction includes using a multi-view stereo matching network to extract image depth information and generate dense point clouds, and using a Gaussian splash neural rendering network to perform implicit scene reconstruction on the dense point clouds and apply geographic coordinate constraints. The specific details are as follows:

[0056] In step S120, after generating the standardized geographic dataset in step S110, a neural network 3D scene reconstruction operation is performed to generate a realistic 3D geographic base scene. This step is the core of the entire 3D scene animation construction method. It abandons the traditional manual modeling and mesh stitching approach and adopts a deep learning-based dual-network collaborative architecture to achieve automated 3D reconstruction of the geographic scene.

[0057] The neural network 3D scene reconstruction includes a first stage of dense point cloud generation and a second stage of implicit scene reconstruction.

[0058] In the first stage, a multi-view stereo matching network is used to extract image depth information and generate a dense point cloud. Specifically, multi-view remote sensing images and aerial oblique photographic images from a standardized geographic dataset are input into a pre-trained multi-view stereo matching network. This network adopts an improved cascaded MVSNet architecture. First, a feature extraction network with shared weights performs multi-scale global feature extraction on images from various views, obtaining depth feature representations for each image at different scales. Then, the network's built-in self-attention mechanism is used to match corresponding features between images from different views, establishing dense correspondences between multi-view images. Based on this, the network constructs a 3D cost volume and performs cost aggregation and depth regression to accurately solve for the scene depth information corresponding to each pixel. Since the network has learned the depth distribution patterns of a large number of complex scenes during training, it can adaptively adapt to areas with changes in lighting, terrain occlusion, and complex vegetation textures, significantly improving the quality and completeness of the depth map compared to traditional stereo matching algorithms. Finally, based on the solved multi-view depth maps, the network generates a high-density geographic dense point cloud that closely matches the real terrain through a depth map fusion algorithm.

[0059] In the second stage, a Gaussian splash neural rendering network is used to perform implicit scene reconstruction on the dense point cloud generated in the first stage, and geographic coordinate constraints are applied during the reconstruction process. Specifically, the dense point cloud output from the first stage is used as the initialization basis for 3D Gaussian particles. Each point is assigned a 3D Gaussian function, which has initial spatial position parameters, scale parameters, and color parameters to describe the geometric structure, distribution range, and color texture information of that local region in the scene.

[0060] After initialization, multi-source geographic information is further introduced as spatial constraints to guide the optimization process of Gaussian particles. On the one hand, DEM terrain data from the standardized geographic dataset is used as an elevation constraint to constrain the vertical distribution range of the 3D Gaussian particles, ensuring that the Gaussian particles do not cross the terrain surface, thereby guaranteeing the geometric realism of the reconstructed scene in terms of terrain undulation. On the other hand, the boundaries of features in the GIS vector data from the standardized geographic dataset are used as horizontal constraints to constrain the distribution density of the 3D Gaussian particles on both sides of the feature boundaries, ensuring that the boundaries of key features such as building outlines and road edges are clear and sharp rather than blurred transitions in the reconstruction results.

[0061] Based on this, a geographic coordinate constraint loss function is constructed and embedded into the overall loss function of the Gaussian splash neural rendering network. This geographic coordinate constraint loss function includes at least two parts: a spatial location loss term and a feature boundary consistency loss term. The spatial location loss term measures the deviation between the spatial position of the 3D Gaussian particles and the actual geographic coordinates, ensuring that the particle positions strictly correspond to the GIS coordinate system. The feature boundary consistency loss term constrains the distribution of Gaussian particles on both sides of the vector feature boundary to exhibit abrupt or transitional characteristics consistent with the actual geographic features, thereby ensuring the accuracy of the reconstructed scene at semantic boundaries.

[0062] Finally, through iterative optimization, the position, scale, and color parameters of the 3D Gaussian particles were jointly optimized. Simultaneously, the photometric reconstruction loss and geographic coordinate constraint loss function were minimized. This ensured that while maintaining visual realism, the deviation between the spatial location of the 3D Gaussian particle ensemble and its real geographic coordinates was controlled within a preset geographic accuracy threshold, generating a realistic 3D geographic base scene that combines geometric realism, high texture accuracy, and geographic locationability. This base scene not only possesses cinematic-level visual rendering effects, but more importantly, each spatial location can precisely correspond to the real geographic coordinate system, seamlessly integrating with various spatial analysis and query functions of geographic information devices. This lays the foundation for subsequent dynamic geographic element overlay and fusion, and the generation of 3D scene animations.

[0063] Step S140: In response to the 3D scene animation generation command, based on the real-world 3D geographic base scene and time-series monitoring geographic data, the spatiotemporal evolution of dynamic geographic elements is driven, and the 3D scene animation is output along the preset camera path through the differentiable rendering pipeline. The specific content is as follows:

[0064] In step S140, after generating the real-world 3D geographic base scene in step S120, the scene dynamization and animation output process is initiated in response to the 3D scene animation generation command issued by the user or the host device. This step is the core output link of the entire 3D scene animation construction method. On the static real-world 3D geographic base scene established in step two, dynamic geographic information is further integrated and the scene evolution over time is driven. At the same time, continuous animation generation under a free viewpoint is achieved through neural rendering technology.

[0065] First, static geographic features contained in the GIS vector data of the standardized geographic dataset are overlaid and fused into a real-world 3D geographic base scene. These static features include at least linear and planar features with clear geographic semantics and fixed spatial locations, such as building outlines, road networks, water system boundaries, pipeline distribution, and land parcel boundaries. During the overlay and fusion process, on the one hand, the precise geographic coordinates of the vector features are used to correct potential boundary deviations in the neural network reconstruction scene. For example, when there is a spatial offset between the building edge reconstructed by Gaussian splashing and the vector building outline, the spatial distribution of Gaussian particles in the corresponding area is adjusted to ensure strict alignment. On the other hand, the attribute information of the vector features is associated and bound to the corresponding areas in the base scene, so that each feature in the base scene not only has a geometric shape and texture appearance but also queryable geographic attribute information, thereby ensuring the standardization and accuracy of geographic features. After this overlay and fusion process, the static features and the base scene achieve a dual unification of spatial location and semantic attributes.

[0066] Then, based on time-series monitoring geographic data, the spatiotemporal evolution of dynamic geographic elements is driven, constructing the dynamic change capability of the scene. The time-series monitoring geographic data refers to process data reflecting the changes of geographic elements over time, obtained from repeated observations of the same geographic area at different time points. This data includes, but is not limited to, hydrological monitoring data, vegetation monitoring data, traffic flow monitoring data, and disaster evolution monitoring data. These data record the state changes of various geographic elements at different times, such as the rise and fall of water levels over time, the seasonal changes in vegetation indices, the time-varying distribution of traffic flow, and the dynamic diffusion process of disaster impact areas.

[0067] Based on this, a dynamic evolution model is constructed using time-series monitoring geographic data. Specifically, for each type of geographic dynamic process, the state parameters of geographic elements corresponding to each discrete time point in the time-series monitoring data are first extracted. Then, a continuous interpolation fitting method is used to smoothly transition the states between discrete time points, generating a state change sequence with continuous temporal resolution. For example, for a flood spread process, the location of the water body boundary is extracted using remote sensing monitoring data at different times, and the inundation range at any intermediate time is generated through time-series interpolation. For a traffic flow process, the traffic flow density distribution at any time is generated using road segment traffic flow monitoring data at different time periods through interpolation fitting. Through the above time-series interpolation fitting operation, the discrete monitoring time state is extended into a continuous time evolution process, enabling dynamic geographic elements to present their corresponding states at any animation time point, thereby ensuring the continuity and smoothness of spatiotemporal evolution.

[0068] After completing the overlay and fusion of static geographic features and the temporal evolution modeling of dynamic geographic features, the two are integrated into a unified 3D scene space. Specifically, using a real-world 3D geographic base scene as the underlying framework, static geographic features are fixed to their corresponding spatial positions in the base scene with precise alignment, while the temporal evolution state of dynamic geographic features is overlaid on the base scene as a dynamic layer. Static geographic features provide a stable and unchanging geographic structural skeleton for the scene, while dynamic geographic features endow the scene with the dynamism of change over time. The two are collaboratively expressed under the same spatial reference and time axis, forming a static-dynamic fused scene that encompasses both static landforms and dynamic evolution processes. This fused scene not only contains high-precision static geographic information but also possesses dynamic characteristics that evolve continuously over time, achieving a leap from a static 3D model to a dynamic 4D spatiotemporal scene.

[0069] Finally, a 3D scene animation is output along a preset camera path using a differentiable rendering pipeline. During animation generation, user-defined camera motion keyframes and roaming path parameters are first acquired. The camera motion keyframes consist of several viewpoint positions and orientations specified by the user in the 3D scene; each keyframe defines the camera's pose at a specific moment, including its spatial coordinates and viewing orientation. The roaming path parameters define the camera's movement speed and transition method between adjacent keyframes.

[0070] Then, smooth interpolation is performed on the poses between camera motion keyframes to generate a continuous viewpoint motion trajectory. Specifically, after acquiring the camera poses of two adjacent keyframes, a smooth interpolation algorithm is used to continuously transition the camera position and orientation between the two keyframes, allowing the camera to move smoothly from the pose of the starting keyframe to the pose of the ending keyframe, avoiding screen jumps and visual discomfort caused by abrupt pose changes. By sequentially performing the above interpolation process on all adjacent keyframes, a complete and continuous viewpoint motion trajectory covering the entire animation timeline is generated.

[0071] Based on this, along the continuous viewpoint motion trajectory, the differentiable rendering pipeline renders the real-world 3D geographic base scene and its overlaid static and dynamic geographic features frame by frame. The differentiable rendering pipeline, based on the Gaussian splash rendering principle, projects a 3D Gaussian particle set onto a 2D imaging plane and calculates the color value and opacity of each pixel, generating a highly realistic 3D scene. Because the differentiable rendering pipeline possesses complete end-to-end differentiability, it maintains the continuity and consistency of scene geometry and texture during rendering, ensuring smooth and natural transitions between adjacent frames without flickering or jitter. Simultaneously with frame-by-frame rendering, the camera pose in each frame strictly follows the pre-set continuous viewpoint motion trajectory, thus forming a continuously changing 3D scene animation.

[0072] To improve the rendering efficiency of large-scale geographical scenes during frame-by-frame rendering, a differentiated rendering strategy based on viewpoint distance was introduced. Specifically, during each frame rendering, the distance between the scene objects and the viewpoint in the current frame is first determined, and this distance is used as the basis for determining rendering priority. Based on this distance, differentiated rendering precision is applied to distant and near-field areas. Near-field areas, located near the viewpoint, have objects occupying a large projected area in the image and require higher detail representation; therefore, a high-precision rendering configuration is used, including the use of a larger number of 3D Gaussian particles and high-resolution texture mapping. Distant areas, far from the viewpoint, have objects occupying a smaller projected area in the image, and the human eye's ability to perceive details is limited; therefore, a low-precision rendering configuration is used, including the use of a smaller number of 3D Gaussian particles and low-resolution texture mapping. Through this differentiated rendering strategy, the rendering computation per frame is significantly reduced while maintaining visual quality, thus improving overall rendering efficiency.

[0073] After the above frame-by-frame rendering process, all the rendered single frames are encoded and synthesized in chronological order, and finally output a high-definition, continuous, and customizable 3D scene animation, completing the fully automated construction process from input of multi-source heterogeneous geographic data to output of 3D scene animation.

[0074] Based on steps S110 to S140 above, in this embodiment of the invention, standardized preprocessing of multi-source heterogeneous geographic data achieves unified spatial reference and efficient fusion of satellite remote sensing, aerial imagery, DEM, DOM, GIS vector, laser point cloud, and time-series monitoring data. A dual-network collaborative architecture of multi-view stereo matching network and Gaussian splash neural rendering network enables automated construction of realistic 3D scenes without manual fine modeling. Geographic coordinate constraint loss function is used to give the scene geographic locationability to connect with GIS devices. Time-series monitoring data drives the evolution of dynamic geographic elements and combines differentiable rendering pipelines to output free-view continuous animation along preset camera paths. Finally, the automated construction of multi-source geographic information 3D scene animation with high realism, high geographic accuracy, and dynamic evolution is achieved.

[0075] The method of this invention, in an embodiment, performs standardization preprocessing on acquired multi-source heterogeneous geographic data, including:

[0076] Perform unified spatial coordinate registration on multi-source heterogeneous geographic data to transform each data source to the target coordinate system;

[0077] Perform image denoising, point cloud denoising, and missing data imputation on multi-source heterogeneous geographic data;

[0078] Align multi-source heterogeneous geographic data that have undergone unified spatial coordinate registration and processing under a unified spatiotemporal reference.

[0079] Furthermore, a multi-view stereo matching network is used to extract image depth information to generate a dense point cloud, including:

[0080] Input multi-view remote sensing images and aerial oblique photography images from standardized geographic datasets into a multi-view stereo matching network;

[0081] Global features of images from various perspectives are extracted using a multi-view stereo matching network, and corresponding feature points are matched between images.

[0082] Based on the matching results, the scene depth information is solved, and a dense point cloud that fits the real terrain and features is output.

[0083] Furthermore, a Gaussian splash neural rendering network is used to perform implicit scene reconstruction of dense point clouds and impose geographic coordinate constraints, including:

[0084] Initialize the spatial position, scale, and color parameters of three-dimensional Gaussian particles based on dense point clouds;

[0085] Using DEM terrain data as elevation constraints, the vertical distribution range of three-dimensional Gaussian particles is constrained.

[0086] Using the boundaries of ground features in GIS vector data as horizontal constraints, the distribution density of three-dimensional Gaussian particles on both sides of the ground feature boundaries is constrained.

[0087] Construct a geographic coordinate constraint loss function, which includes a spatial location loss term and a feature boundary consistency loss term;

[0088] Through iterative optimization, the geographic coordinate constraint loss function is minimized, so that the deviation between the spatial position of the 3D Gaussian particle and the real geographic coordinates is less than or equal to the preset geographic accuracy threshold, thereby generating a real-world 3D geographic base scene with geographic locationability.

[0089] Furthermore, the spatiotemporal evolution of dynamic geographic elements includes:

[0090] Overlaying and fusing static geographic features from GIS vector data into a real-world 3D geographic base scene;

[0091] A dynamic evolution model is constructed based on time-series monitoring geographic data to perform time-series interpolation fitting on geographic dynamic processes;

[0092] By integrating static geographic features with the evolution results of dynamic geographic processes, a static-dynamic fusion scene is generated.

[0093] Furthermore, the time-series monitoring geographic data includes at least one of hydrological monitoring data, vegetation monitoring data, traffic flow monitoring data, and disaster evolution monitoring data; the geographic dynamic processes include at least one of flood spread processes, vegetation growth and decline processes, traffic flow processes, and disaster diffusion processes.

[0094] Furthermore, a 3D scene animation is output along a preset camera path via a differentiable rendering pipeline, including:

[0095] Obtain user-defined camera motion keyframes and roaming path parameters;

[0096] Smoothly interpolate the pose between camera motion keyframes to generate continuous viewpoint motion trajectories;

[0097] Along a continuous viewpoint motion trajectory, the real-world 3D geographic base scene and dynamic geographic elements are rendered frame by frame through a differentiable rendering pipeline to generate a 3D scene animation.

[0098] Furthermore, when rendering frame by frame along the motion trajectory of a continuous viewpoint, it also includes:

[0099] Determine the distance between scene features and the viewpoint in the current rendering frame;

[0100] Based on distance, different rendering precision is applied to distant and near-field areas, with the rendering precision of near-field areas being higher than that of distant areas.

[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0102] This invention also provides a multi-source geographic information 3D scene animation construction device, which is used to implement the above embodiments and preferred embodiments, and will not be repeated as already described. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0103] Figure 2 According to one embodiment of the present invention, a multi-source geographic information 3D scene animation construction device includes:

[0104] The processing module 201 is used to respond to the input command of multi-source geographic data, perform standardized preprocessing on the acquired multi-source heterogeneous geographic data, and generate a standardized geographic dataset. The multi-source heterogeneous geographic data includes satellite remote sensing images, aerial oblique photography images, DEM topographic data, DOM orthophoto images, GIS vector data, laser point cloud data, and time-series monitoring geographic data.

[0105] The reconstruction module 202 is used to perform neural network 3D scene reconstruction based on the standardized geographic dataset to generate a real-world 3D geographic base scene. The neural network 3D scene reconstruction includes using a multi-view stereo matching network to extract image depth information to generate a dense point cloud, and using a Gaussian splash neural rendering network to perform implicit scene reconstruction on the dense point cloud and apply geographic coordinate constraints.

[0106] The generation module 203 is used to respond to the three-dimensional scene animation generation command, drive the spatiotemporal evolution of dynamic geographic elements based on the real-world three-dimensional geographic base scene and the time-series monitoring geographic data, and output the three-dimensional scene animation along the preset camera path through the differentiable rendering pipeline.

[0107] It should be noted that the above modules can be implemented by software or hardware. For the latter, they can be implemented in the following ways, but are not limited to: all the above modules are located in the same processor; or, the above modules are located in different processors in any combination.

[0108] According to one embodiment of the present invention, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed by a processor, implements the above-described method for constructing multi-source geographic information three-dimensional scene animation.

[0109] Optionally, in this embodiment, the processor can be configured to perform the following steps via a computer program:

[0110] Step S1: In response to the input command of multi-source geographic data, perform standardized preprocessing on the acquired multi-source heterogeneous geographic data to generate a standardized geographic dataset. The multi-source heterogeneous geographic data includes satellite remote sensing images, aerial oblique photography images, DEM topographic data, DOM orthophoto images, GIS vector data, laser point cloud data and time-series monitoring geographic data.

[0111] Step S2: Based on a standardized geographic dataset, perform neural network 3D scene reconstruction to generate a real-world 3D geographic base scene. The neural network 3D scene reconstruction includes using a multi-view stereo matching network to extract image depth information to generate a dense point cloud, and using a Gaussian splash neural rendering network to perform implicit scene reconstruction on the dense point cloud and apply geographic coordinate constraints.

[0112] Step S3: In response to the 3D scene animation generation command, based on the real-world 3D geographic base scene and time-series monitoring geographic data, drive the spatiotemporal evolution of dynamic geographic elements, and output the 3D scene animation along the preset camera path through the differentiable rendering pipeline.

[0113] According to one embodiment of the present invention, a computer-readable storage medium is also provided, the computer-readable storage medium including a stored executable program, wherein, when the executable program is running, it controls the device where the storage medium is located to execute the above-described method for constructing a three-dimensional scene animation based on multi-source geographic information.

[0114] Optionally, in this embodiment, the storage medium may be configured to store a computer program for performing the following steps:

[0115] Step S1: In response to the input command of multi-source geographic data, perform standardized preprocessing on the acquired multi-source heterogeneous geographic data to generate a standardized geographic dataset. The multi-source heterogeneous geographic data includes satellite remote sensing images, aerial oblique photography images, DEM topographic data, DOM orthophoto images, GIS vector data, laser point cloud data and time-series monitoring geographic data.

[0116] Step S2: Based on a standardized geographic dataset, perform neural network 3D scene reconstruction to generate a real-world 3D geographic base scene. The neural network 3D scene reconstruction includes using a multi-view stereo matching network to extract image depth information to generate a dense point cloud, and using a Gaussian splash neural rendering network to perform implicit scene reconstruction on the dense point cloud and apply geographic coordinate constraints.

[0117] Step S3: In response to the 3D scene animation generation command, based on the real-world 3D geographic base scene and time-series monitoring geographic data, drive the spatiotemporal evolution of dynamic geographic elements, and output the 3D scene animation along the preset camera path through the differentiable rendering pipeline.

[0118] Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0119] According to one embodiment of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the above-described method for constructing multi-source geographic information three-dimensional scene animation.

[0120] Optionally, in this embodiment, the above-mentioned computer program product can be configured as a computer program that performs the following steps:

[0121] Step S1: In response to the input command of multi-source geographic data, perform standardized preprocessing on the acquired multi-source heterogeneous geographic data to generate a standardized geographic dataset. The multi-source heterogeneous geographic data includes satellite remote sensing images, aerial oblique photography images, DEM topographic data, DOM orthophoto images, GIS vector data, laser point cloud data and time-series monitoring geographic data.

[0122] Step S2: Based on a standardized geographic dataset, perform neural network 3D scene reconstruction to generate a real-world 3D geographic base scene. The neural network 3D scene reconstruction includes using a multi-view stereo matching network to extract image depth information to generate a dense point cloud, and using a Gaussian splash neural rendering network to perform implicit scene reconstruction on the dense point cloud and apply geographic coordinate constraints.

[0123] Step S3: In response to the 3D scene animation generation command, based on the real-world 3D geographic base scene and time-series monitoring geographic data, drive the spatiotemporal evolution of dynamic geographic elements, and output the 3D scene animation along the preset camera path through the differentiable rendering pipeline.

[0124] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0125] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0126] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between units or modules may be electrical or other forms.

[0127] The units described as separate components may or may not be physically separate. Similarly, the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0128] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0129] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0130] The above are merely 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 principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for constructing a 3D scene animation using multi-source geographic information, characterized in that, include: In response to the input command of multi-source geographic data, the acquired multi-source heterogeneous geographic data is standardized and preprocessed to generate a standardized geographic dataset. The multi-source heterogeneous geographic data includes satellite remote sensing images, aerial oblique photography images, DEM topographic data, DOM orthophoto images, GIS vector data, laser point cloud data and time-series monitoring geographic data. Based on the standardized geographic dataset, a neural network 3D scene reconstruction is performed to generate a real-world 3D geographic base scene. The neural network 3D scene reconstruction includes using a multi-view stereo matching network to extract image depth information to generate a dense point cloud, and using a Gaussian splash neural rendering network to perform implicit scene reconstruction on the dense point cloud and apply geographic coordinate constraints. In response to the 3D scene animation generation command, based on the real-world 3D geographic base scene and the time-series monitoring geographic data, the spatiotemporal evolution of dynamic geographic elements is driven, and the 3D scene animation is output along the preset camera path through the differentiable rendering pipeline.

2. The method for constructing a three-dimensional scene animation using multi-source geographic information according to claim 1, characterized in that, Standardization preprocessing is performed on the acquired multi-source heterogeneous geographic data, including: Perform unified spatial coordinate registration on the multi-source heterogeneous geographic data to transform each data source to the target coordinate system; Image denoising, point cloud denoising, and missing data imputation are performed on the multi-source heterogeneous geographic data; Align multi-source heterogeneous geographic data that have undergone unified spatial coordinate registration and processing under a unified spatiotemporal reference.

3. The method for constructing a three-dimensional scene animation using multi-source geographic information according to claim 1, characterized in that, A multi-view stereo matching network is used to extract image depth information and generate a dense point cloud, including: The multi-view remote sensing images and aerial oblique photography images in the standardized geographic dataset are input into the multi-view stereo matching network. The multi-view stereo matching network extracts global features from images from each viewpoint and matches corresponding feature points between images. Based on the matching results, the scene depth information is solved, and the dense point cloud that fits the real terrain and features is output.

4. The method for constructing a three-dimensional scene animation using multi-source geographic information according to claim 1, characterized in that, The dense point cloud is implicitly reconstructed using a Gaussian splash neural rendering network, and geographic coordinate constraints are applied, including: The spatial position, scale, and color parameters of the three-dimensional Gaussian particles are initialized based on the dense point cloud. The DEM terrain data is used as an elevation constraint to constrain the vertical distribution range of the three-dimensional Gaussian particles. The boundaries of the ground features in the GIS vector data are used as horizontal constraints to constrain the distribution density of the three-dimensional Gaussian particles on both sides of the ground feature boundaries. Construct a geographic coordinate constraint loss function, which includes a spatial location loss term and a feature boundary consistency loss term; Through iterative optimization, the geographic coordinate constraint loss function is minimized, so that the deviation between the spatial position of the three-dimensional Gaussian particle and the real geographic coordinates is less than or equal to a preset geographic accuracy threshold, thereby generating the real-world three-dimensional geographic base scene with geographic locationability.

5. The method for constructing a three-dimensional scene animation using multi-source geographic information according to claim 1, characterized in that, The spatiotemporal evolution of dynamic geographic features includes: The static geographic features in the GIS vector data are overlaid and fused into the real-world 3D geographic base scene; A dynamic evolution model is constructed based on the aforementioned time-series monitoring geographic data, and time-series interpolation fitting is performed on the geographic dynamic process. The static geographic features are integrated with the evolution results of the geographic dynamic process to generate a static-dynamic fusion scene.

6. The method for constructing a three-dimensional scene animation using multi-source geographic information according to claim 5, characterized in that, The time-series monitoring geographic data includes at least one of hydrological monitoring data, vegetation monitoring data, traffic flow monitoring data, and disaster evolution monitoring data; the geographic dynamic process includes at least one of flood spread process, vegetation growth and decline process, traffic flow process, and disaster diffusion process.

7. The method for constructing a three-dimensional scene animation using multi-source geographic information according to claim 1, characterized in that, Output 3D scene animation along a preset camera path using a differentiable rendering pipeline, including: Obtain user-defined camera motion keyframes and roaming path parameters; Smooth interpolation is performed on the pose between the camera motion keyframes to generate a continuous viewpoint motion trajectory; Along the continuous viewpoint motion trajectory, the real-world 3D geographic base scene and the dynamic geographic elements are rendered frame by frame through the differentiable rendering pipeline to generate the 3D scene animation.

8. The method for constructing a three-dimensional scene animation using multi-source geographic information according to claim 7, characterized in that, When rendering frame by frame along the continuous viewpoint motion trajectory, the method further includes: Determine the distance between scene features and the viewpoint in the current rendering frame; Based on the distance, different rendering precision is applied to the distant and near areas, with the rendering precision of the near area being higher than that of the distant area.

9. A device for constructing three-dimensional scene animation using multi-source geographic information, characterized in that, include: The processing module is used to respond to input instructions for multi-source geographic data, perform standardized preprocessing on the acquired multi-source heterogeneous geographic data, and generate a standardized geographic dataset. The multi-source heterogeneous geographic data includes satellite remote sensing images, aerial oblique photography images, DEM topographic data, DOM orthophoto images, GIS vector data, laser point cloud data, and time-series monitoring geographic data. The reconstruction module is used to perform neural network 3D scene reconstruction based on the standardized geographic dataset to generate a real-world 3D geographic base scene. The neural network 3D scene reconstruction includes using a multi-view stereo matching network to extract image depth information to generate a dense point cloud, and using a Gaussian splash neural rendering network to perform implicit scene reconstruction on the dense point cloud and apply geographic coordinate constraints. The generation module is used to respond to the 3D scene animation generation command, drive the spatiotemporal evolution of dynamic geographic elements based on the real-world 3D geographic base scene and the time-series monitoring geographic data, and output the 3D scene animation along the preset camera path through the differentiable rendering pipeline.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored executable program, wherein, when the executable program is executed, it controls the device where the computer-readable storage medium is located to perform the multi-source geographic information three-dimensional scene animation construction method as described in any one of claims 1 to 8.

Images