Three-dimensional scene updating method

Abstract

The invention belongs to the field of image processing, and provides a three-dimensional scene updating method, which comprises the following steps: acquiring an existing three-dimensional structure model of a three-dimensional scene to be updated; acquiring a first space segment of the plurality of space segments, and acquiring two-dimensional image data of the first space segment; determining a first template index according to the structure of the first space fragment, a basic texture map and the two-dimensional image data; matching the first template index with a template mapping library, and determining an update template of the first space fragment; and extracting real texture content in the two-dimensional image according to the two-dimensional image data and the updating template, mapping the real texture content to a target segment of a corresponding three-dimensional model, and updating the three-dimensional scene.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of image processing, and in particular relates to a three-dimensional scene updating method. BACKGROUND

[0002] With the development of digital economy and smart tourism, three-dimensional modeling technology is increasingly widely used in tourism scenes, especially in smart tourism platforms, where three-dimensional scenes not only serve as the core carrier for virtual roaming, path guidance, and cultural display for tourists, but also play a role in real-time interaction and information fusion. Based on the gradual maturity of traditional city landscape modeling, three-dimensional modeling of rural cultural tourism areas has gradually become an important direction for the digital upgrade of tourism. Among them, the natural rural environment represented by the typical regions of the Sichuan West Plateau and hilly basins, due to its strong topographic continuity, sparse distribution of structures, and small change range, has become one of the key coverage objects of smart tourism.

[0003] In the prior art, the three-dimensional model updating method for the above-mentioned environment mainly relies on manual mapping or parameter-driven rendering means. One method is based on time nodes (such as seasonal segmentation), climate factors (such as temperature, humidity), etc. to set the environment rendering nodes, and to apply uniform light and texture parameters to the given model structure to simulate the visual effects of spring, summer, autumn, and winter environment states. Another method directly collects real scene photos and manually pastes them on three-dimensional mesh patches to achieve high-fidelity texture updating. Both of these methods have obvious shortcomings in practical application.

[0004] For parameter rendering methods, as the texture details of the real environment cannot be accurately obtained, the rendering results often show a strong programmatic and templated style, which cannot truly restore the changes on the ground and local features, especially in detail-sensitive areas such as fields, mountains, and country roads, resulting in unnatural and untrustworthy user experience. For direct mapping methods, although the reality is high, they rely on large-scale manual operation and image cropping, which has a long update cycle, heavy system burden, and is difficult to meet the dynamic and rapid change requirements of smart tourism applications.

[0005] Especially in rural areas such as Sichuan, environmental changes mainly manifest as seasonal vegetation changes, local water fluctuations, and other micro-scale texture evolution, with the overall terrain structure being basically stable and the building volume being limited. Therefore, if traditional piece-by-piece mapping or global rendering methods are still used, it not only causes resource waste, but also fails to meet the dual requirements of real-time updating and high-fidelity display. Therefore, it is urgent to propose a fast and realistic three-dimensional scene updating method to improve the real-time performance, authenticity, and operational efficiency of rural smart tourism systems. SUMMARY

[0006] To solve the problems in the prior art, the present application provides a three-dimensional scene updating method, comprising the following steps:

[0007] Step S10, obtaining an existing three-dimensional structure model of a three-dimensional scene to be updated, the three-dimensional structure model being divided into a plurality of spatial segments, each spatial segment corresponding to a three-dimensional grid region and being pre-provided with a basic texture map;

[0008] Step S20, obtaining a first spatial segment of the plurality of spatial segments, obtaining two-dimensional image data of the first spatial segment, the two-dimensional image data being a set of image frames taken from multiple angles;

[0009] Step S30, determining a first template index according to the structure and the basic texture map of the first spatial segment and the two-dimensional image data;

[0010] Step S40, matching the first template index with a template mapping library to determine an update template of the first spatial segment;

[0011] Step S50, extracting real texture content in a two-dimensional image according to the two-dimensional image data and the update template, and mapping the real texture content to a target segment of a corresponding three-dimensional model to update the three-dimensional scene.

[0012] Further, the update template refers to a standardized image structure description unit for guiding texture update of each spatial segment in a three-dimensional scene, a data object integrating visual features and mapping strategies, and is used to map real texture in a two-dimensional image to a corresponding three-dimensional structure surface. Each template includes at least the following contents:

[0013] Template index feature: reflecting the geometric topological information of the spatial segment matched by the template index;

[0014] Texture adaptation strategy: describing how to select an area from a two-dimensional image for sampling;

[0015] Mapping parameter set: including UV mapping method, illumination adjustment factor, and occlusion processing rule;

[0016] Matching weight rule: a similarity standard for judging whether the template is applicable to the current spatial segment.

[0017] Further, step S10 specifically includes the following sub-steps:

[0018] Step S101, analyzing structure model data of a three-dimensional scene to be updated, identifying a spatial boundary range, and dividing the spatial boundary range into a plurality of spatial segments according to a rule;

[0019] Step S102, establishing a mapping relationship between each spatial segment and a three-dimensional grid region, and loading corresponding basic texture map information based on prior data, the basic texture map including topographic images, historical remote sensing images, or artificial rendering layers.

[0020] Further, the step S20 specifically comprises the following sub-steps:

[0021] Step S201, set the shooting point, the number of view angles and the shooting parameters of the two-dimensional image acquisition, including but not limited to resolution, exposure value and view angle coverage;

[0022] Step S202, collect image frames covering the target space segment from multiple shooting points, and record the shooting time, view angle posture and position information of each image frame;

[0023] Step S203, pre-process the collected image frames, including image denoising, light equalization and color correction, to improve the texture recognition quality;

[0024] Step S204, generate an image frame set and establish an angular projection association between the image frames and the space segment, which is used for frame selection and region matching in the subsequent image extraction process.

[0025] Further, the step S30 specifically comprises the following sub-steps:

[0026] Step S301, extract the structural parameters of the first space segment, including geometric boundary, orientation, surface normal information;

[0027] Step S302, calculate the image feature vector based on the base texture map and the two-dimensional image frame features, the feature vector including texture direction, color histogram and edge distribution map;

[0028] Step S303, jointly encode the structural parameters and image features to form a first template index, the first template index containing category encoding, texture feature encoding and environment label, for uniquely identifying the current state of the space segment.

[0029] Further, the step S40 specifically comprises the following sub-steps:

[0030] Step S401, perform formatting processing on the first template index, match its corresponding classification path and label structure;

[0031] Step S402, access the template mapping library, locate the template classification directory according to the index prefix, and match the closest standard template according to the template feature vector;

[0032] Step S403, extract the matched update template and check the integrity of its texture image file and mapping parameters, if the template is not complete, call the suboptimal template or the standby template.

[0033] Further, the texture content extraction in step S50 specifically comprises the following sub-steps:

[0034] Step S501, based on the projection relationship between the image frame and the target segment, determine the region coordinates of the image covering the segment, and establish the mapping correspondence table of the image to the three-dimensional face;

[0035] Step S502, select the main image frame using the image definition score mechanism, and crop the real texture image of the target region from the main image frame;

[0036] Step S503, affine transform and edge transition processing of the cropped image according to the direction, scaling and mixing parameters in the update template, to generate the final texture tile.

[0037] Further, the texture mapping in step S50 specifically includes the following sub-steps:

[0038] Step S504, generating UV texture coordinates for the mesh patches of the target segment, and binding the processed image texture to the three-dimensional mesh surface according to the mapping rule;

[0039] Step S505, updating the texture mapping data of the target segment in the three-dimensional scene, refreshing the scene rendering state, and realizing the local real-time visualization update of the scene.

[0040] Further, the update template supports multi-segment sharing, if multiple spatial segments correspond to the same or similar template index, then they are uniformly matched to the same update template.

[0041] Further, the method is applicable to natural rural areas, and the update template is preset and classified according to typical land, road, mountain and river scenes.

[0042] The three-dimensional scene updating method provided by the application can realize targeted and efficient texture updating process without changing the original three-dimensional structure model, through the structure slicing and template indexing mechanism. By quickly matching the texture style suitable for the current spatial segment from the standardized template library, and automatically extracting the real texture content combined with multi-angle image data, the method effectively improves the accuracy and automation level of three-dimensional scene updating.

[0043] The template mapping mechanism adopted by the application can automatically identify the geometric category and surface texture evolution state of the spatial segment based on the fusion of three-dimensional structure features and image texture features, realize adaptive matching of texture replacement and image projection, and significantly improve the reality and consistency of the updated scene. The method is especially suitable for rural areas, scenic spots and other scenes with slow structure change but frequent texture detail change, and can complete high-frequency visual content update without relying on large-scale manual operation.

[0044] In addition, the application realizes the modularization and reusable design of texture updating by introducing a template-based texture standard system, significantly reduces data redundancy and map overhead, effectively reduces the running burden and bandwidth consumption of the system, and is suitable for deployment in lightweight tourism terminals or edge computing nodes, thereby enhancing the resource utilization efficiency and deployment flexibility of the smart tourism system. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0046] Figure 1 is the main flowchart of the three-dimensional scene updating method of the present application;

[0047] Figure 2 is the structural level diagram of the template mapping library in the present application;

[0048] Figure 3 is the schematic diagram of image texture fusion and edge transition processing in the present application;

[0049] Figure 4 is the schematic diagram of the fragment effect comparison before and after the three-dimensional scene updating in the present application. DETAILED DESCRIPTION

[0050] Next, the preferred description of the application will be made in combination with the drawings and the specific embodiments.

[0051] The present embodiment solves the above problems by the following steps:

[0052] In one embodiment, as shown in Figure 1 The present application discloses a three-dimensional scene updating method, which is suitable for a three-dimensional digital modeling system for rural natural and cultural environment, and aims to improve the updating efficiency while ensuring the authenticity of the three-dimensional scene updating. In view of the characteristics that the rural scene has small change range, relatively stable spatial structure, and the change mainly reflects in the visual performance of seasonal replacement and climate conditions in the actual operation process, the present application proposes a three-dimensional scene fast updating mechanism based on template driving. The method extracts a unified image updating template for spatial fragments with similar geometric structures, and reuses the same texture mapping parameters, thereby avoiding the high time-consuming operation of reconstructing the entire scene frame by frame and region by region, and improving the overall rendering and updating efficiency. At the same time, compared with the method of selecting the rendering state in a coarse-grained manner according to the season or climate label, the present application directly extracts real texture information based on the real image, and combines the template to accurately control the mapping range and method, effectively balancing the authenticity and speed of updating.

[0053] To achieve the above object, the present application specifically comprises the following steps:

[0054] Step S10, obtaining an existing three-dimensional structure model of a three-dimensional scene to be updated, the three-dimensional structure model being divided into a plurality of spatial segments, each spatial segment corresponding to a three-dimensional mesh region and being preset with a basic texture map;

[0055] Step S20, obtaining two-dimensional image data of the first spatial segment, the two-dimensional image data being a set of image frames taken from a plurality of angles;

[0056] Step S30, obtaining a first spatial segment of the plurality of spatial segments, and determining a first template index according to the structure of the first spatial segment, the basic texture map and the two-dimensional image data;

[0057] Step S40, matching the first template index with a template mapping library to determine an update template of the first spatial segment;

[0058] Step S50, extracting real texture content in the two-dimensional image according to the two-dimensional image data and the update template, and mapping the real texture content to a target segment of the corresponding three-dimensional model to update the three-dimensional scene.

[0059] In the present application, a template refers to a standardized image structure description unit for guiding texture update of each spatial segment in a three-dimensional scene, and its essence is a data object integrating typical visual features and mapping strategies, which is used to quickly and accurately map real texture in a two-dimensional image to a corresponding three-dimensional structure surface.

[0060] Each template at least includes the following contents:

[0061] Template index feature: reflecting geometric topological information (such as mesh shape, boundary contour) of the spatial segment matched by the template;

[0062] Texture adaptation strategy: describing how to select an area from a two-dimensional image for sampling;

[0063] Mapping parameter set: including UV mapping mode, illumination adjustment factor, occlusion processing rule, etc.

[0064] Matching weight rule: a similarity standard for judging whether the template is applicable to the current spatial segment.

[0065] The three-dimensional scene updating method disclosed by the application realizes efficient fusion of structure recognition and texture updating based on a template index. The core principle is to decompose the complex spatial updating task into three stages of structure recognition, template calling and image driving, forming a set of reusable and automated three-dimensional local rendering framework. In the method, the system first acquires the three-dimensional structure model to be updated, and divides it into multiple spatial segments, each segment being a three-dimensional mesh region with specific geometric shape and mapping characteristics. By analyzing the basic texture map and geometric structure of each spatial segment, the system constructs a representative template index, which can be regarded as the unique identification of the spatial segment in the template space. Then, the system uses the index to find the most matched update template in the preset template mapping library. The template contains information such as typical spatial configuration, texture layout strategy, UV mapping parameter, illumination compensation rule, and is determined by the preset similarity matching rule. Next, the system collects two-dimensional image data corresponding to the spatial segment, and the image comes from multiple perspectives to enhance the comprehensiveness of the texture. In actual updating, the system extracts the real texture content corresponding to the three-dimensional model surface structure from the image according to the texture extraction rule specified by the template, and maps it to the corresponding mesh region of the three-dimensional model through projection transformation, UV mapping and other ways, thereby completing the local updating operation with high reality and consistency. The method does not depend on point cloud reconstruction, nor does it need to re-render the whole three-dimensional model, and is particularly suitable for efficient texture updating and rendering replacement of rural scenes with stable structure and repetitive configuration.

[0066] The method of the present application effectively solves the problems of low texture reconstruction efficiency, excessive manual intervention and poor update consistency in traditional three-dimensional scene updating by introducing a structured template mechanism. In actual modeling process, there are often a large number of structural units with similar structures but different positions in three-dimensional scenes, such as multiple farmland, mountain, road or river segments. Through the template mechanism in the present application, similar geometric structures can be classified into the same category, and a unified update template is called by the system to extract and map the texture, thereby significantly improving the consistency and rendering quality of the map. In addition, each template not only includes spatial feature information, but also encapsulates corresponding texture extraction rules and map parameters, such as UV stretching ratio, sampling area size, illumination correction factor, and occlusion area processing method. These parameters can be directly reused for multiple spatial segments, avoiding repeated manual setting and improving the automation level of the entire modeling process. Compared with traditional methods that rely only on manual modeling or image projection, the template mechanism not only speeds up the update, but also ensures the style consistency and texture accuracy, especially suitable for rural scenes with simple structure and high repetition.

[0067] Compared with traditional three-dimensional modeling or texture updating methods, the present application has significant advantages. In traditional methods, the update of three-dimensional scenes often needs to obtain the latest point cloud data through three-dimensional scanning devices, or to perform image projection and manual mapping for each region one by one, which not only complicates the process, but also easily causes a work bottleneck when dealing with large-scale scenes. In the method described in the present application, relying on the template mapping mechanism, the system can quickly identify the geometric type corresponding to the spatial segment and automatically call the matching template for texture replacement, without starting from zero each time. The traditional method also has the problem of difficulty in unifying the style. Even if multiple regions have similar structures, due to different texture sources, map scale and lighting conditions, visual jumps may still occur in the modeling results. The present method ensures that all spatial segments belonging to the same template category finally present a style-consistent and detailed texture performance through the parameter setting and image screening standard under the unified template. Especially in rural digital scenes, many structural units have a high degree of repetition, such as batches of brick houses, farmland irrigation channels, farmland, mountains, etc. Figure 2 As shown in the structure level diagram of the template mapping library in the present application (only schematic, not complete structure), the traditional method performs a complete modeling process for each unit, which is extremely inefficient, while the present method can greatly improve the update efficiency and response speed of three-dimensional scenes through one template matching and multiple template reuse.

[0068] In a specific example, it is necessary to dynamically express the state of farmland in different seasons. The farmland area is divided into regular spatial segments by the system, with geometric structures similar to rectangles or trapezoids, flat terrain, and regular boundaries. The system first constructs a standardized "farmland planting area template F-T001", which includes the following: the spatial projection mode is planar orthographic projection; the texture extraction area is an 80% square range in the center of the image; the texture sampling scale is 0.1 meters / pixel; and the accompanying texture parameters include color saturation enhancement, strip structure detection factor, vegetation NDVI threshold mask, etc. In the actual updating process, the operator uses a drone to take multi-angle low-altitude aerial photography over the farmland. The system obtains the current image frame set of each farmland, and quickly extracts the main texture image area through the "F-T001" template, adjusts it to the preset UV coordinate system, and then attaches it to the farmland three-dimensional grid model, realizing the rapid scene update from "sowing period bare land" to "heading period green plants". Traditional methods require independent reconstruction and mapping for each farmland, while the present invention can efficiently reuse one template to support automatic updating of hundreds of farmlands within a few minutes.

[0069] In another specific example, in the smart tourism platform, the digital three-dimensional scene of mountain tourism attractions often faces seasonal changes, such as changes in mountain color, snow conditions, and vegetation coverage differences. To address this issue, the system establishes a "mountain gentle slope rock body template M-T008", which is suitable for areas with a slope of less than 45 degrees, continuous surface structure, and rock-soil mixed mountain bodies. The template includes: slope average normal vector extraction rules, color and shade level matching standards, UV map compression rate set to 0.75, and occlusion area transparent edge smoothing function, etc. In the running process, the system extracts mountain photo information from remote sensing images or photos taken by climbers, and uses the template to update the mountain in segments, especially for quickly and accurately replacing the mountain surface color (such as autumn red and yellow mixed forest), texture (such as spring new green texture), and snow conditions (such as winter white coverage on the top). Compared to the traditional method of re-scanning the entire mountain model, the present invention only extracts the significantly changed area and uses the template to guide the texture projection, greatly improving the update speed and authenticity.

[0070] In another specific example, the road surface evolves from a dirt road to a gravel road, and then to a cement or asphalt road. In the system, the road is divided into regular linear spatial segments, and two typical templates, "Rural Asphalt Road Template R-T015" and "Gravel Compacted Road Template R-T009", are constructed. Taking "R-T015" as an example, it contains: the linear texture tiling direction is the main axis direction, the texture repeating unit is 2.5 meters, a 10 cm texture transition zone is provided between the road curb stone and the main road, and the vehicle wheel mark weighted mask factor is 0.3. After the village road reconstruction construction is completed, the system collects the post-construction road surface photos, extracts the image features in the new asphalt state, performs texture projection through the "R-T015" template, and applies it to all spatial segments belonging to the road section in the three-dimensional model, realizing the visual state replacement of the entire road. In the traditional method, a long strip texture needs to be redrawn and manually adapted to the road curve shape, while in the present method, through the alignment rules and texture shrinking algorithm in the template, the texture adaptive adjustment of the curved road or intersection area can be automatically completed.

[0071] In another specific example, the river or stream often changes in visual state due to changes in rainfall, water level, season, etc. In the present method, two representative templates, "Low Water Level Clear River Surface Template W-T002" and "Rainy Season Turbid Water Body Template W-T003", are preset. In the templates, the water surface normal disturbance frequency, the reflected light shielding factor, the color depth distribution map, and the dynamic UV flow animation control parameters are defined. After the system performs geometric recognition on the river segment, the corresponding template is called, and the current texture features are extracted and updated in the water surface area of the three-dimensional model in combination with the water surface photos collected by the camera monitoring or real-time water quality monitoring images. When the water quality is detected to be mixed and the waves are increased, it can be automatically switched from "W-T002" to "W-T003", and the real water body images are fused in multiple frames to generate a dynamic visual map by superimposing the flow vector field. In the traditional method, the flow direction and color difference need to be manually rendered by modeling software, which is difficult to respond to water condition changes in real time. In the present method, the state change of the river is quickly, automatically and in real time reflected through template driving, which is particularly suitable for rural flood prevention warning, landscape simulation, real-time river patrol, etc.

[0072] In a further implementation, in order to further explain the working principle of the present application, a detailed implementation of each step will be described.

[0073] Step S10, an existing three-dimensional structure model of a three-dimensional scene to be updated is obtained, the three-dimensional structure model is divided into a plurality of spatial segments, each spatial segment corresponds to a three-dimensional grid area, and a basic texture map is preset.

[0074] In the present application, step S10 aims to obtain an existing three-dimensional structure model of the three-dimensional scene to be updated, and to structure the division thereof, so as to improve the processing efficiency of subsequent texture matching, template retrieval and image updating. Since a three-dimensional scene often contains a large number of heterogeneous objects, directly updating the entire scene not only has large computational overhead, but also easily causes problems such as texture inconsistency and mapping errors. Therefore, by dividing the three-dimensional structure model into multiple spatial segments and presetting a basic texture map, texture identification and replacement operations at the segment level can be realized on the basis of maintaining the stability of the three-dimensional geometric structure, thereby improving the local accuracy and response speed of the overall update. In addition, the basic texture map, as an expression of the original state, can provide a reference basis for subsequent template index generation.

[0075] In the present application, the three-dimensional structure model refers to a meshed geometric structure set representing the construction of a real scene in a physical space, which is usually composed of multiple triangular facets, and is used to accurately restore the spatial form of an object in a virtual space. The spatial segment refers to a local area divided in the three-dimensional structure model according to geometric features, functional areas or texture continuity, and each spatial segment usually corresponds to a mesh subset with independent semantic attributes, such as a wall, a fence or a road. The three-dimensional mesh region is the actual expression of the spatial segment in a computer graphics system, and is usually stored in the form of a facet set composed of vertices, edges and facets. The basic texture map refers to the texture image data attached to the surface of the three-dimensional mesh region when the system is initially modeled or built, which is used to represent the default visual state of the current structure, and can also be regarded as the original reference layer of texture update.

[0076] In an alternative specific implementation, the step S10 specifically includes the following sub-steps:

[0077] Step S101, reading the structure model data of the three-dimensional scene to be updated, which can be derived from a previously modeled data file, a BIM (Building Information Modeling) export format, a point cloud reconstruction model or a three-dimensional scanning result, and the reading process can use common three-dimensional format analysis tools such as OBJ, FBX, PLY or gLTF for format identification and data import.

[0078] Step S102, spatially dividing the three-dimensional structure model based on geometric features, and identifying the segments of the overall structure model according to shape boundary continuity, normal vector variation degree, mesh density and other dimensions. Alternative implementation schemes include: clustering the normal direction and position coordinates of adjacent facets based on clustering algorithms such as K-means or DBSCAN to form spatial segments with strong structural continuity; logically dividing based on artificial preset area labels or region boundaries defined by a geographic information system (GIS).

[0079] Step S103, extract the corresponding mesh region information of each identified space segment, including its own vertex, boundary index, normal vector set and patch texture coordinates, and establish an index mark inside the segment, which is used to indicate the scope of the subsequent texture update operation.

[0080] Step S104, attach the initial or basic texture map information of each space segment, which can be the real scene image collected at the beginning of modeling, the texture layer left by the historical update stage, or the general texture under the default template. The loading method of texture map includes but is not limited to image binding through UV mapping coordinates, or realizing the corresponding mapping of image and geometric region through orthogonal projection.

[0081] Step S105, establish a unified data structure representation for all space segments after division, which records the spatial position, mesh topology, texture state, texture map path and mapping relationship with the template library of each segment, so as to facilitate template matching, image recognition and three-dimensional update operation in the subsequent steps.

[0082] Through the processing of this step, the original three-dimensional structure model is clearly divided into several space segments with clear structure logic and known texture state, so that the subsequent template recognition and texture replacement operation can be executed at the segment level, greatly improving the processing efficiency and reducing the risk of update error propagation. This processing strategy makes the three-dimensional scene update have good local controllability and scalability, while avoiding the high resource overhead of overall reconstruction. It is a kind of efficient scene management method for rural scenes with stable structure, repeated form and slow texture change.

[0083] In a specific example, the system first imports the three-dimensional model data of the whole village agricultural area through step S101, and identifies a plurality of rectangular farmland segments composed of regular meshes in step S102. In step S103, the system extracts the mesh boundary and vertex set of each farmland, and constructs the number index. In step S104, the spring collected farmland basic texture is attached to each farmland as the current state identifier. Finally, in step S105, all segments form a unified segment structure mapping table, realizing the linkage management of structure, texture and geographical position, laying a data foundation for subsequent template matching and texture replacement. This example fully illustrates the necessity and practicality of three-dimensional structure model division and basic texture loading in the invention.

[0084] Step S20, acquiring two-dimensional image data of the first space segment, the two-dimensional image data being a set of image frames taken from multiple angles.

[0085] In the present application, step S20 is used to obtain two-dimensional image data corresponding to the first spatial segment, which is a set of image frames taken from multiple different angles, and the main purpose is to provide an image basis with sufficient angle coverage and detail accuracy in the subsequent texture extraction and three-dimensional mapping process. Due to the problems of structural occlusion, uneven lighting, and angle ambiguity in any spatial segment in a three-dimensional scene, relying on a single angle image may result in texture loss or mapping error. Therefore, by collecting image frames from multiple angles, the system can construct a more complete and accurate two-dimensional image description space, thereby enhancing the integrity of texture extraction and the reliability of geometric alignment, and improving the visual authenticity and spatial matching degree of three-dimensional update effect.

[0086] In the present application, two-dimensional image data refers to digital image frames obtained from real scenes by image acquisition devices, and the image frame set refers to an ordered set of multiple images obtained at different shooting angles, positions or times for the same spatial segment. These image frames can come from continuous pictures collected by unmanned aerial vehicle low-altitude cruise shooting, ground standing photography, fixed point monitoring cameras or vehicle-mounted camera equipment. The multi-angle characteristic of the image frame set means that each image has a different camera position and orientation in order to cover all possible occluded or inclined areas on the surface of the spatial segment from multiple angles.

[0087] In an alternative specific implementation, step S20 specifically includes the following sub-steps:

[0088] Step S201, set image acquisition parameters, including shooting distance, focal length, resolution, exposure level, number of shooting angles and angle distribution range of the camera or image acquisition device. The system can automatically or manually set multiple shooting points to cover 360-degree horizontal angles and appropriate elevation or depression angle ranges of the area where the spatial segment is located. Alternative implementation schemes include: using a drone to fly around above the spatial segment to collect multiple image frames covering the full angle of view at equidistant shooting points; or using a tripod fixed camera equipment to shoot multiple static images at specified angles.

[0089] Step S202, perform image acquisition operation, including image shooting, image numbering and time stamp annotation. Image frames are stored in sequence according to the shooting order, with position and attitude data attached, including three-dimensional coordinate position and angle direction information of the shooting device, which can be obtained by joint calibration of GPS and IMU (Inertial Measurement Unit).

[0090] Step S203, image preprocessing operations are performed, which include image denoising, brightness equalization, color correction and lens distortion correction. The denoising can be achieved by a median filter algorithm, the brightness equalization can use a histogram equalization method, the color correction can be based on a standard color card for white balance adjustment, and the lens distortion correction can be based on a preset lens model parameter for reverse mapping processing.

[0091] Step S204, angle registration and frame set generation are performed on the collected image frames. The system identifies all image frames according to the spatial angle distribution of the shooting position, ensures that the images in the set cover all areas of the surface of the spatial segment, and reduces the occluded repeated areas as much as possible. The system can establish an image angle matrix or a view coverage map to record the projection range of each image that can be covered, to guide the subsequent texture extraction process.

[0092] Through the above steps, the system can efficiently obtain a two-dimensional image frame set with multi-view and full coverage characteristics, providing a comprehensive and reliable data basis for subsequent template matching, texture extraction and three-dimensional mapping. Compared with the traditional method of relying only on single-view images for modeling, this step significantly improves the spatial integrity and texture accuracy of image data, avoiding the problems of occluded dead angles, texture stretching and color distortion.

[0093] In a specific example, after identifying a rectangular farmland segment, the system calls step S201 to set the shooting parameters, sets six equally spaced shooting points using a drone, the flight height is 20 meters, one overhead view and two oblique views at different angles are taken at each point, forming a total of 18 image sets. Then in step S202, image acquisition and position labeling are completed, in step S203, image brightness standardization and color uniformity processing are performed, and finally in step S204, an image frame set containing image angle information and coverage matrix is generated, ensuring that the complete and high-precision farmland surface texture can be spliced from different angles when extracting texture, realizing real and distortion-free three-dimensional scene updating. This method effectively solves the problem of texture gaps caused by single shooting angle in traditional methods, improving the application accuracy and practicality of the system in agricultural scenes.

[0094] Step S30, a first spatial segment of the plurality of spatial segments is obtained, and a first template index is determined according to the structure of the first spatial segment, a base texture map and the two-dimensional image data.

[0095] In the present application, the purpose of step S30 is to select the first spatial segment from multiple spatial segments, and determine the most suitable update template index for the segment in combination with the structural information, base texture map and corresponding two-dimensional image data of the segment. The determination of the template index is one of the key steps in the entire three-dimensional scene updating process, and its main role is to establish the correspondence between the segment and the existing templates in the template library. Due to the existence of a large number of shape similar and texture similar regions in actual three-dimensional models, if the texture content is analyzed independently and the texture mapping rule is reconstructed every time, not only the efficiency is low, but also the style is inconsistent. By identifying the structural features and image features of the spatial segment, a unified template index is generated, which can enable the system to call the corresponding template structure and mapping rule in subsequent updates, thereby realizing efficient and consistent batch texture updating.

[0096] In the present application, the first spatial segment refers to the target segment that is first selected for processing in the current update round, and is usually the region with the highest priority or the most significant change identified by the system. Structure refers to the geometric topological properties of the spatial segment in three-dimensional space, including mesh shape, vertex coordinates, normal vector distribution, surface curvature and other information. The base texture map is the original texture image currently attached to the spatial segment, which is used to provide visual reference in the historical or initial state. The two-dimensional image data refers to the multi-view image set covering the spatial segment obtained by an external camera device. The template index refers to a unique identifier generated based on the structure, texture and image information, which is used to retrieve the corresponding template in the template mapping library.

[0097] In an alternative specific implementation, the step S30 specifically comprises the following sub-steps:

[0098] Step S301: Extract spatial structure features

[0099] The system performs mesh structure analysis on the first spatial segment and extracts the following parameters:

[0100] Area A: the total area of all faces is accumulated;

[0101] Boundary length L: the sum of all edge lengths of the outer contour is calculated;

[0102] Normal vector direction N: the unit vectors of all face normal vectors are averaged to obtain the main direction;

[0103] Surface curvature mean K: the second derivative estimation of the normal vector change rate at each vertex is performed, and the mean of all points is taken;

[0104] Face density D: represents the number of mesh faces per unit area, D=N_f / A.

[0105] The geometric feature vector Vg = [A, L, Nx, Ny, Nz, K, D] is constructed, wherein Nx, Ny, and Nz are respectively the components of the normal vector direction on the x, y, and z axes.

[0106] Step S302: Extracting basic texture statistical features

[0107] The system analyzes the content of the basic texture map image and extracts the following indexes:

[0108] Color mean C_avg: the pixel mean of R, G, and B channels is respectively calculated;

[0109] Color variance σ_c: representing the degree of color variation;

[0110] Gray level co-occurrence matrix feature GLCM: including texture contrast, correlation, energy, and entropy;

[0111] Main direction θ: after converting the image to the frequency domain, the maximum energy direction in the Fourier spectrum is calculated as the texture directionality feature.

[0112] The texture feature vector Vt = [R_avg, G_avg, B_avg, σ_c, GLCM_1, GLCM_2, θ] is constructed.

[0113] Step S303: Extracting image frame texture matching features

[0114] For each image frame, the system performs the following processing:

[0115] Projecting the three-dimensional coordinates of the spatial segment to the image plane to determine the position in the image;

[0116] Extracting local texture feature points in the coverage area, which can use the ORB algorithm to obtain a set of descriptor vectors for each image frame;

[0117] Performing K-means clustering on the descriptors of all image frames to generate K representative cluster centers to form the image feature vector Vi.

[0118] For example, assuming that 100 feature points are extracted from each image frame and K = 10 is set, then a final image feature matrix with a dimension of K x D (D is the descriptor dimension, for example, 256) is formed.

[0119] Step S304: Fusing three types of feature vectors to construct a joint feature vector

[0120] The system concatenates the aforementioned geometric features Vg, texture features Vt, and image features Vi into a joint vector V:

[0121] V = [Vg || Vt || flatten(Vi)]

[0122] To ensure that each dimension of the vector is in the same numerical scale, normalization processing is performed:

[0123] For each dimension v_i, the standardization formula is used:

[0124] v_i' = (v_i - μ_i) / σ_i

[0125] Where μ_i is the mean value of this dimension in the template library, and σ_i is the standard deviation.

[0126] Step S305: Perform template index matching

[0127] The system performs similarity matching between the standardized feature vector and the standard vector of all templates in the template library, and calculates the Euclidean distance using the following formula:

[0128] Dist(V, V_templ) = sqrt(Σ_i (v_i' - v_templ_i')^2)

[0129] Select the template with the smallest distance as the matching template, and return its template index T_index.

[0130] If multiple templates have close distances and are less than a set threshold ε, the system can further compare the texture direction or color distribution difference rate to finally determine the optimal matching template.

[0131] Through the above series of sub-steps, the system starts from the first spatial segment, comprehensively collects structural information, original texture, and current image data, and forms a highly expressive feature vector through standardization processing and feature aggregation mechanism, and then quickly finds the best matching template in the template library through an efficient similarity calculation method. This process not only improves the accuracy and robustness of index matching, but also has strong universality and automation capabilities, and can support parallel processing and template reuse in large-scale three-dimensional models.

[0132] In a specific example, in a rural agricultural area, the system detects that a rectangular farmland area needs to be updated with three-dimensional texture.

[0133] Spatial structure feature extraction:

[0134] The farmland segment is divided into a rectangular spatial segment with a length of about 90 meters and a width of about 30 meters, with a total area A = 2700 square meters.

[0135] The grid is composed of dense rectangular facets, with a number of facets N_f = 3240.

[0136] The average normal vector direction is (0, 0, 1), indicating that the terrain is flat.

[0137] The surface curvature K ≈ 0, and the boundary is complete and has no missing parts.

[0138] Basic texture feature extraction:

[0139] The original map shows light brown bare soil, R_avg≈180, G_avg≈160, B_avg≈120.

[0140] The gray level co-occurrence matrix (GLCM) shows low contrast and weak directionality.

[0141] The texture has few details, with a standard deviation σ_c≈15, indicating that the texture changes little.

[0142] Two-dimensional image feature extraction:

[0143] Six directional image frames are obtained by UAV flight, a total of 18 photos;

[0144] The new image frame shows light green strip-shaped vegetation texture with a certain directionality, with a main direction θ≈45°(northeast by east);

[0145] High-density linear region feature points are identified in the image.

[0146] Index matching process:

[0147] After the system fuses the structural features, texture features, and image features, it generates a joint feature vector;

[0148] Match with the existing "Spring Sowing Period Farmland Texture Template (T-AF-001)" in the template library, the Euclidean distance is the smallest, and the matching score is 0.94;

[0149] The system finally determines to use the template index "TAF001".

[0150] Update template content:

[0151] Template "TAF001" contains strip-shaped directional vegetation texture patterns, based on flat grid strip mapping, with uniform texture direction distribution and color transition strategy, suitable for the initial stage of spring sowing.

[0152] In another specific example, in the northwest mountainous edge area of a rural scenic spot, the system identifies that the mountain surface needs to be updated.

[0153] Spatial structure feature extraction:

[0154] This spatial segment is an irregular surface with a grid area A≈620 square meters;

[0155] The surface curvature distribution range is K∈[0.02, 0.06], showing a gentle undulating slope;

[0156] The average normal vector direction is (-0.4, -0.3, 0.85), indicating a southwest direction with an upward bias;

[0157] The mesh patch density D ≈ 0.19.

[0158] Basic texture feature extraction:

[0159] The original map is bare rock texture, with gray and white as the main color;

[0160] The color standard deviation σ_c ≈ 40, and the GLCM shows moderate directionality and high energy value.

[0161] The texture is moderately repetitive, with relatively coarse patterns.

[0162] Two-dimensional image feature extraction:

[0163] A total of 12 image frames are obtained from the perspective of the two side slopes and the top view;

[0164] The new image frame shows a large number of vine plants attached to the mountain surface, with a green coverage of more than 60%;

[0165] The main direction θ ≈ 90° (developing vertically along the rock mass direction), and the texture detail density increases.

[0166] Index matching process:

[0167] After generating the joint feature vector and comparing it with the "summer green period mountain template (M-GR-009)" in the template library, the matching degree is 0.91;

[0168] The matching degree of the similar template "rock rainy season template M-WT-003" is only 0.86, and the former is finally selected;

[0169] Output template index "MGR009".

[0170] Update template content:

[0171] Template "MGR009" is defined as a rock-based mesh + green plant coverage layered structure, and the mapping area distribution automatically adjusts the texture density according to the curvature, suitable for summer mountain surface plant growth scenes.

[0172] In another specific example:

[0173] In a certain village main road section, the system detects a road segment that needs to be updated.

[0174] Spatial structure feature extraction:

[0175] This segment is a long rectangular mesh with a length of 150 meters and a width of 4.5 meters;

[0176] The total area of the grid is about A = 675 square meters;

[0177] The grid normal vector direction is (0, 0, 1), indicating that the road surface is flat;

[0178] The surface curvature K is approximately 0, and the number of grid patches is N_f = 1800.

[0179] Basic texture feature extraction:

[0180] The original map is a dark gray cement floor, and the main color tone RGB is (80, 80, 80);

[0181] The texture contrast is moderate, and the standard deviation σ_c is approximately 20;

[0182] The GLCM shows part of the transverse crack pattern, and the directionality θ is approximately 0°.

[0183] Two-dimensional image feature extraction:

[0184] The image source is a total of 10 groups of data from road side monitoring and vehicle-mounted recorders;

[0185] The image shows that the road surface has obvious ruts and water reflection, and there are fine sand and gravel deposits;

[0186] The image feature descriptor shows a high-frequency texture pattern, reflecting high detail differences.

[0187] Index matching process:

[0188] After comprehensive analysis, the system matches the template "Rainy Road Damage Texture Template (R-RN-012)", and the matching score is 0.96;

[0189] After comparison with the conventional dry road template "R-DR-004", it is obviously preferred;

[0190] Determine the index "RRN012".

[0191] Update the template content:

[0192] The template "RRN012" adopts a standard asphalt grid structure, generates two high reflectivity channel textures in the center of the road, simulates ruts and water reflections, and randomly distributes sand and gravel map blocks in the edge area to achieve real rain after the road texture effect reconstruction.

[0193] Step S40, match the first template index with the template mapping library to determine the update template of the first space segment.

[0194] In the three-dimensional scene updating method, in order to ensure the accuracy and efficiency of the space segment texture updating, a preset template index system is used to quickly locate the updating template with the highest matching degree with the target space segment in the unified template mapping library. The updating template not only includes texture image information, but also includes mapping method, texture direction, transparency rule and texture transition strategy, so that the updating effect can achieve a good balance between local reality and global coordination. Through the matching mechanism of the template index and the template mapping library, redundant manual judgment and repeated texture collection can be avoided, and the overall updating efficiency can be improved and standardized expansion can be supported.

[0195] In an alternative implementation, the step S40 specifically comprises the following sub-steps:

[0196] Step S401: receiving the first template index and performing format standardization

[0197] The system receives the first template index T_index generated in the step S30. In order to ensure that the index is consistent with the template identification format in the template library, the system performs format standardization operation, including removing redundant identifiers, converting case to uniform form, and confirming whether the prefix conforms to the template classification specification, such as "AF" representing farmland type and "GR" representing mountain green type.

[0198] Step S402: accessing the template mapping library and locating the template classification directory

[0199] The system accesses the template mapping library deployed locally or in the cloud. The library is organized in a hierarchical structure, with the first level classification being the landform type and the second level classification being the time / state feature. The system matches the corresponding directory structure according to the standardized index prefix, and quickly locates the target subset. For example, the index "TAF001" will first be located to the "farmland scene template" first level directory, and then matched to the specific template under the "spring planting state" second level classification.

[0200] Step S403: calling the index matching function to retrieve the corresponding updating template content

[0201] The system finds the updating template item corresponding to the index T_index through the mapping table. If the index is a string matching type, the hash table can be directly used for retrieval. If the index includes a vector pointer, the nearest neighbor search algorithm such as KD-Tree or HNSW is used to find the nearest item in the template feature space. The retrieval result is the updating template T_model, and its content includes:

[0202] Template texture image;

[0203] Mapping coordinate rule;

[0204] Normal direction adaptation strategy;

[0205] transparency or blending factor;

[0206] Optional edge transition parameters.

[0207] Step S404: Verify the validity of the updated template

[0208] The system verifies that the retrieved template contains a complete structure, primarily checking whether the texture file exists, whether mapping parameters are missing, and whether the mapping rules are compatible with the current mesh structure. If the check passes, the template is confirmed as the updated template. If the template is incomplete or conflicts exist, the options include falling back to a suboptimal template, triggering a manual approval process, or using the default base texture.

[0209] By executing these sub-steps in tandem, the system can quickly and accurately retrieve the update template that best matches the spatial fragment in complex and diverse scenes. This method supports highly automated processing while allowing for manual review, making it suitable for both batch updates and real-time adjustments. The template indexing and library mapping mechanism not only ensures texture visual consistency and stability, but also significantly reduces data redundancy and storage pressure, forming a key foundation for a lightweight 3D update system.

[0210] In a specific example, the system determines that the template index is "TAF001" in step S30. In step S401, it is standardized to the lowercase prefix "taf001", and the system recognizes that "taf" corresponds to the "Farmland Spring Sowing Template" category. In step S402, the system enters the "Template_Library / Farmland / Spring" directory, and calls a hash search therein to match the updated template "TAF001". The template contains a green striped texture image with a diagonal texture direction, a transparency of 1.0, and an edge blending width of 0.2 meters. After verifying that the texture file exists and the mapping parameters are complete in step S404, the system confirms that "TAF001" is a valid update template. Finally, the template is mapped to the original space fragment, quickly replacing the bare soil map, and achieving a visual update effect of the farmland during the sowing period. If the index is wrong or the template is missing, the system will automatically return to "TAF000 (standard spring sowing general template)" to ensure that the update process is not interrupted.

[0211] Step S50: extracting real texture content in the two-dimensional image according to the two-dimensional image data and the update template, and mapping it to a target segment of the corresponding three-dimensional model, thereby updating the three-dimensional scene.

[0212] To achieve high-fidelity and rapid updates of the visual effects of spatial fragments in a 3D scene, it is necessary to combine actual captured 2D image data with a standardized update template. Texture content representing the current real-world scene state must be extracted from the image frame and accurately mapped to the corresponding 3D target fragment area. This mapping process not only ensures precise geometric correspondence between the image texture and the 3D fragment, but also matches rendering parameters such as direction, blending method, and edge transition rules in the update template. This allows the appearance of the 3D model to be updated without changing its structure, improving the scene's real-time performance and visual quality.

[0213] In an optional specific implementation, step S50 specifically includes the following sub-steps:

[0214] Step S501: Determine the projection correspondence between the image frame and the 3D segment

[0215] Based on the camera position parameters and shooting angle, the system projects each image frame onto the 3D model through the extrinsic matrix. Specifically, the perspective projection model is used to calculate the mapping relationship between the pixel point P_i in the image and the surface point P_3d of the 3D segment. The formula is as follows:

[0216] P_i=K×[R|t]×P_3d

[0217] Among them, K is the camera intrinsic parameter matrix, R is the rotation matrix, t is the translation vector, P_3d is the homogeneous coordinate of the three-dimensional point, and P_i is the image pixel coordinate.

[0218] Step S502: Extracting high-quality texture areas from image frames

[0219] like Figure 3 As shown in the figure, based on the established projection correspondence, the system identifies the set of pixel regions covering the target segment. To ensure texture clarity and lighting consistency, the system prioritizes image frames shot at near-vertical angles and free of strong light interference as the primary texture source. Optional implementation options include using the Structural Similarity (SSIM) algorithm to evaluate the local clarity of each frame, prioritizing the regions with the highest scores for extraction.

[0220] Step S503: Fusing the updated template's mapping parameters to perform coordinate mapping

[0221] Further parameters Figure 3 After determining the primary texture region (i.e., the region with the highest score), the system transforms the extracted image texture according to the texture direction θ, texture scaling ratio s, and edge blending width α defined in the updated template. Specifically, this involves rotating it to the template direction θ, scaling it to the target fragment size by the ratio s, and applying a transparency gradient to the edge α region to achieve a smooth transition.

[0222] Step S504: Projecting and mapping the processed texture image to the three-dimensional segment mesh

[0223] The system generates texture coordinates (UV coordinates) for the three-dimensional target segment mesh, and maps the two-dimensional image texture according to the adjusted direction and size. If the three-dimensional segment contains multiple irregular surfaces, the system uses a face-by-face mapping method to calculate the texture coordinates and bind the texture image region for each mesh surface.

[0224] Step S505: Performing texture mapping synthesis and scene update

[0225] As shown in Figure 4 After the texture mapping of all target segments is completed, the system performs texture mapping synthesis operation to replace the original base texture with the newly generated real texture image, and updates the data structure of the three-dimensional scene model, refreshes the rendering state to ensure that the new texture content takes effect immediately in the interactive or display system.

[0226] Through the above steps, the system can realize high-fidelity texture update based on real image data without changing the three-dimensional structure model. Combined with the guidance rules of the standardized update template, the update process is not only accurate but also has high repeatability and system scalability, which is suitable for batch processing in rural areas, scenic spots and other natural scenes. Compared with traditional texture reconstruction methods, the present application has significant advantages in processing efficiency, texture consistency and automation degree.

[0227] In a specific example, the system has matched the template "TAF001" and extracted a set of light green strip-shaped texture patterns from the northeast direction unmanned aerial vehicle image frames. In step S501, the system projects the image to the segment plane according to the camera pose, and locates the 1280x720 region as the target texture region. Then, in step S502, the system evaluates the local sharpness of each frame image and identifies the 5th frame as the best quality. Next, in step S503, the system rotates the texture by 45 degrees (i.e. θ = 45°), scales the proportion s = 1.2, and sets the blending width α = 0.3 meters in the edge area. After the transformation is completed, in step S504, the texture is bound to the mesh surface corresponding to the farmland segment. Finally, in step S505, the system refreshes the texture data of the region to generate a green new farmland texture image with spring planting strips, realizing efficient and rapid scene update. This process from image acquisition to final rendering does not require human intervention, fully embodying the intelligence and operability of the present application in practical application.

[0228] The part of the module structure not specifically defined in the present application shall be subject to the content described in the prior art. The prior art mentioned in the foregoing background section and the specific embodiment section of the present application can be used as a part of the present application to understand the meaning of some technical features or parameters.

Claims

1. A three-dimensional scene updating method, characterized in that: The following steps are involved: Step S10: obtaining an existing three-dimensional structural model of the three-dimensional scene to be updated, wherein the three-dimensional structural model is divided into a plurality of spatial segments, each of which corresponds to a three-dimensional grid area and has a preset basic texture map; Step S20: Acquire a first spatial segment of the multiple spatial segments, and acquire two-dimensional image data of the first spatial segment, where the two-dimensional image data is a set of image frames captured from multiple angles; Step S30: determining a first template index according to the structure of the first spatial segment, the basic texture map, and the two-dimensional image data; Step S40: Match the first template index with a template mapping library to determine an updated template for the first spatial segment; Step S50: extracting real texture content in the two-dimensional image according to the two-dimensional image data and the update template, and mapping it to a target segment of the corresponding three-dimensional model, thereby updating the three-dimensional scene.

2. The three-dimensional scene updating method according to claim 1, characterized in that: The update template refers to a standardized image structure description unit used to guide the texture update of each spatial segment in a three-dimensional scene. It is a data object that integrates visual features and mapping strategies and is used to map the real texture in the two-dimensional image to the corresponding three-dimensional structure surface. Each template includes at least the following: Template index feature: reflects the geometric topological information of the matching spatial fragment; Texture adaptation strategy: describes how to select areas for sampling from a two-dimensional image; Mapping parameter set: including UV mapping method, lighting adjustment factor, and occlusion processing rules; Matching weight rule: A similarity standard used to determine whether a template is applicable to the current spatial segment.

3. The three-dimensional scene updating method according to claim 1, characterized in that: The step S10 specifically includes the following sub-steps: Step S101: parsing the structural model data of the 3D scene to be updated, identifying its spatial boundary range, and dividing it into multiple spatial segments according to rules; Step S102 : establishing a mapping relationship between each spatial segment and a three-dimensional grid area, and loading corresponding basic texture map information based on prior data, wherein the basic texture map includes a topographic image, a historical remote sensing image, or an artificial rendering layer.

4. The three-dimensional scene updating method according to claim 1, characterized in that: The step S20 specifically includes the following sub-steps: Step S201, setting the shooting points, number of viewing angles and shooting parameters for 2D image acquisition, including but not limited to resolution, exposure value and viewing angle coverage; Step S202: collecting image frames covering the target space segment from multiple shooting points, and recording the shooting time, viewing angle posture and position information of each frame; Step S203, pre-processing the collected image frames, including image denoising, illumination balancing and color correction, to improve texture recognition quality; Step S204 : generating an image frame set and establishing an angular projection association between the image frames and the spatial segments, for use in the subsequent frame and region matching in the image extraction process.

5. The three-dimensional scene updating method according to claim 1, characterized in that: The step S30 specifically includes the following sub-steps: Step S301, extracting structural parameters of the first spatial segment, including geometric boundary, orientation, and surface normal information; Step S302, calculating an image feature vector based on the basic texture map and the two-dimensional image frame features, wherein the feature vector includes a texture direction, a color histogram, and an edge distribution map; Step S303: jointly encode the structural parameters and the image features to form a first template index, where the first template index includes a category code, a texture feature code, and an environment label, and is used to uniquely identify the current state of the spatial segment.

6. The three-dimensional scene updating method according to claim 1, characterized in that: The step S40 specifically includes the following sub-steps: Step S401: Formatting the first template index to match its corresponding classification path and label structure; Step S402: access the template mapping library, locate the template classification directory according to the index prefix, and match the closest standard template according to the template feature vector; Step S403: extract the matched updated template and verify the integrity of its texture image file and mapping parameters. If the template is incomplete, call a suboptimal template or a backup template.

7. The three-dimensional scene updating method according to claim 1, characterized in that: The texture content extraction in step S50 specifically includes the following sub-steps: Step S501, based on the projection relationship between the image frame and the target segment, determine the coordinates of the region covering the segment in the image, and establish a mapping table from the image to the three-dimensional surface; Step S502 , selecting a main image frame using an image clarity scoring mechanism, and cropping a real texture image of a target area from the main image frame; Step S503 , performing affine transformation and edge transition processing on the cropped image according to the direction, scaling and blending parameters in the updated template to generate a final texture block.

8. The three-dimensional scene updating method according to claim 1, characterized in that: The texture mapping in step S50 specifically includes the following sub-steps: Step S504 , generating UV texture coordinates for the mesh surface of the target segment, and binding the processed image texture to the three-dimensional mesh surface according to the mapping rule; Step S505 , updating the texture map data of the target fragment in the three-dimensional scene, refreshing the scene rendering state, and realizing a local real-time visual update of the scene.

9. The three-dimensional scene updating method according to claim 1, characterized in that: The update template supports sharing of multiple segments. If multiple spatial segments correspond to the same or similar template indexes, they are all matched to the same update template.

10. The three-dimensional scene updating method according to claim 1, characterized in that: The method is applicable to natural rural areas, and the update template is preset and classified according to typical landform scenes such as farmland, roads, mountains, and rivers.

Images