Real scene three-dimensional modeling method and management system for karst landform scenic area

Abstract

The invention discloses a karst landform scenic spot-oriented live-action three-dimensional modeling method and management system, and relates to the technical field of digital travel, and the method comprises the following steps: employing an unmanned plane, a three-dimensional laser scanner and the like to collect scenic spot inclined image data, point cloud data and detail texture data; constructing a hierarchical fusion algorithm to perform fusion processing on the multi-source data, and generating a live-action three-dimensional model; generating text travel thematic data by utilizing a web crawler and a geocoding technology; the cross-terminal visualization of the live-action three-dimensional model is realized by applying adaptive LOD and GPU rendering acceleration; based on video monitoring and tourist flow data, a risk assessment model is constructed to realize people flow density calculation, flow prediction and risk early warning; and integrating the live-action three-dimensional model, the text travel thematic data and the real-time passenger flow data to realize the global three-dimensional display and management of the scenic area. According to the invention, for tourism management departments and tourists, real-scene three-dimensional modeling and intelligent management are provided, and text and tourism fusion development is assisted.

Description

Technical Field

[0001] This invention relates to the field of digital cultural tourism technology, specifically to a method and management system for real-scene 3D modeling of karst landform scenic areas. Background Technology

[0002] In recent years, Guizhou Province has actively promoted the modernization of its tourism industry. With the development of technologies such as artificial intelligence and digital twins, the cultural tourism industry has entered a highly intelligent development stage of interconnectedness. Real-scene 3D data, as a link between the physical world and digital space, can construct the spatial foundation of cultural tourism platforms and plays an important role in the construction of smart cultural tourism.

[0003] However, in scenic areas like Xijiang Qianhu Miao Village, the terrain has significant elevation differences, and the stilted houses within the village are densely packed and stacked. The resolution and overlap of images collected by drones using oblique photography cannot guarantee the accuracy requirements for architectural modeling. This results in 3D models that are prone to sticking together, having narrow, distorted bottoms, and lacking texture in the decorative patterns. Furthermore, the cultural and tourism data mainly consists of tabular ledger data, scenic area browsing routes in image format, or introductions to core attractions in document format. If traditional methods are used for manual location collection, the workload would be substantial. In addition, due to the large volume and variety of data in the Xijiang Qianhu Miao Village cultural and tourism demonstration system, traditional technologies cannot handle the massive amounts of data on mobile mini-programs to achieve smooth rendering, and there is a lack of real-time visitor flow monitoring and risk assessment.

[0004] For example, Chinese patent CN120182510A discloses a method and system for 3D modeling of scenic spots for tourism management. The method includes: acquiring images of the target scenic spot; determining the probability information of people in the scenic spot image and a Gaussian pyramid; using the probability information of people, correcting the difference images between adjacent layers of the Gaussian pyramid to obtain a target Gaussian difference pyramid; and determining a multi-view stereo model of the target scenic spot based on the target Gaussian difference pyramid. This invention constructs a Gaussian difference pyramid that removes the influence of tourists, so that the feature points selected by SIFT include only the key points of the scenic spot as much as possible, and do not include key points that cause interference from tourists, thereby improving the accuracy and reliability of 3D modeling of scenic spots and avoiding distortion, missing parts, and inaccuracies during 3D reconstruction. Summary of the Invention

[0005] The technical problem to be solved by this invention is to address the shortcomings of existing technologies by using new basic surveying and mapping technologies such as 3D laser scanning, oblique photogrammetry, and close-range photogrammetry to integrate multi-source data to establish a real-world 3D modeling method and management system for tourist attractions in the karst landform of Guizhou. This system aims to reproduce and effectively disseminate tourism culture through digital twins, and provides a real-world 3D modeling method and management system for karst landform scenic areas.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for real-world 3D modeling of karst landscape scenic areas includes the following steps:

[0008] Step S1: Obtain image data of the target scenic area through drone oblique photography, obtain point cloud data through 3D laser scanning, and obtain detailed texture data through close-range photogrammetry.

[0009] Step S2: Construct a hierarchical fusion algorithm to fuse multi-source data and generate a real-world 3D model;

[0010] Step S3: Generate cultural tourism thematic data using web crawling and geocoding technology;

[0011] Step S4: Apply adaptive LOD and GPU rendering acceleration to achieve cross-terminal visualization of real-world 3D models;

[0012] Step S5: Based on video surveillance and tourist flow data, construct a risk assessment model to realize the calculation of crowd density, flow prediction and risk warning;

[0013] Step S6: Integrate real-scene 3D models, cultural tourism data, and real-time visitor flow data to achieve 3D three-dimensional display and management of the entire scenic area.

[0014] Furthermore, step S2 specifically includes the following steps:

[0015] Step S2.1: Perform surface reconstruction on the point cloud data acquired by 3D laser scanning to generate a 3D mesh model;

[0016] Step S2.2: Perform resolution hierarchical decomposition on the three-dimensional mesh model;

[0017] Step S2.3: For each resolution level, perform texture fusion on the data from drone photography and close-up photography to obtain a real-world 3D model.

[0018] Furthermore, in step S2.3, the texture fusion is based on an energy minimization method, specifically including the following steps:

[0019] The energy function is constructed using the following formula:

[0020]

[0021] in, Represents texture The energy function, This indicates the blended texture in texture coordinates. RGB vector values ​​at that location Indicates from the first An image source in texture coordinates Texture values ​​obtained from sampling at that location, Indicates the image source index. Indicates the total number of image sources. Indicates the first An image source in texture coordinates The fusion weight at the point, Represents the regularization parameter. Represents texture In texture coordinates gradient at, Represents the square of the Euclidean norm;

[0022] The optimal fused texture in texture coordinates is obtained by minimizing the energy function using gradient descent. The RGB vector value at that location.

[0023] Furthermore, the specific formula for the fusion weight in the energy function is as follows:

[0024]

[0025] in, Indicates image source exist gradient magnitude at that point This represents the gradient control parameter, used to adjust the intensity of the impact on sharpness. This represents the angle control parameter, used to adjust the intensity of the geometric consistency effect. Indicates image source The angle between the shooting angle and the surface normal vector of the 3D model.

[0026] Furthermore, step S3 specifically includes the following steps:

[0027] Step S3.1: Automatically retrieve relevant text data of the target scenic area from the geographic information service platform using web crawler technology;

[0028] Step S3.2: Perform structured parsing on the collected text data, automatically identify and extract geographical entities including scenic spot names, building signs and street names in the text data;

[0029] Step S3.3: Convert the identified geographic entities into latitude and longitude coordinates to achieve a one-to-one mapping between text data and spatial coordinates;

[0030] Step S3.4: Merge and bind the text data with spatial coordinates with the corresponding positions in the real-world 3D model to generate cultural tourism thematic data.

[0031] Furthermore, step S4 specifically includes the following steps:

[0032] Step S4.1: Establish a multi-level detail model and dynamically adjust the model according to the view distance;

[0033] Step S4.2: Utilize hardware-accelerated rendering technology to improve rendering efficiency through instantiation rendering, texture compression, and shader optimization;

[0034] The multi-level detail model includes a first level, a second level, and a third level. The first level is applicable to scenes with a viewpoint distance greater than 200 meters, the second level is applicable to scenes with a viewpoint distance of 30-200 meters, and the third level is applicable to scenes with a viewpoint distance less than 30 meters.

[0035] Furthermore, in step S5, the calculation of pedestrian density specifically includes:

[0036] The scenic area is divided into multiple zones based on the type of building structure.

[0037] Based on real-time video surveillance data of each area of ​​the scenic area, the number of tourists in each area is output using the YOLOv8 model.

[0038] The population density of each area is calculated based on the number of tourists and the area of ​​each region.

[0039] Furthermore, in step S5, the traffic prediction employs a spatiotemporal graph convolutional network model, specifically including the following steps:

[0040] Model the scenic area as a graph structure ,in, This represents a set of nodes, where each node represents a scenic area. Let represent the set of edges, which represents the spatial connectivity between regions. The feature vector of each node at each time step includes: the current time step's pedestrian flow, region area, region carrying capacity, region type code, and weather influence factor.

[0041] The node feature vectors from past time steps are input into the spatiotemporal graph convolutional network model, and the output is the predicted pedestrian flow for future time steps.

[0042] The access preference index for each area is calculated based on historical visitor traffic data. The specific formula is as follows:

[0043]

[0044] in, Indicates the region Access preference index, area At a historical moment The actual number of people, This indicates the length of the historical time window, taken as 24 hours. Indicates time Maximum pedestrian traffic in all areas Indicates the current moment. This represents the time index within the historical time window. Indicates the region Variance of pedestrian flow over 24 hours This represents the maximum variance of pedestrian flow across all regions. A constant that prevents division by zero;

[0045] The output of the spatiotemporal graph convolutional network model is combined with the access preference index to obtain the final traffic prediction result, specifically using the following formula:

[0046]

[0047] in, This represents the final flow prediction result at time t+1. This represents the original predicted output vector of the spatiotemporal graph convolutional network model at time t+1. This represents the normalized access preference index vector for each region. This represents the carrying capacity vector for each region. It represents the Hadamah accumulation. and These represent the corresponding weights of the spatiotemporal graph convolutional network model output and the access preference index, respectively.

[0048] A real-scene 3D management system for karst landform scenic areas, implemented based on any one of the aforementioned real-scene 3D modeling methods for karst landform scenic areas, including:

[0049] The multi-source data acquisition module is used to collect aerial multi-angle images, point cloud data, and detailed texture data of the scenic area;

[0050] The hierarchical fusion processing module is used to perform three-dimensional mesh reconstruction on the point cloud data and execute multi-resolution geometric decomposition and texture fusion algorithms to generate a real-world three-dimensional model by minimizing the energy function.

[0051] The cultural and tourism themed data generation module is used to crawl text data using web crawlers, extract place name entities through a named entity recognition model, and bind them with three-dimensional spatial coordinates to generate cultural and tourism themed data.

[0052] The risk assessment and early warning module is used to access video surveillance data and historical passenger flow data, predict passenger flow in each area by constructing a spatiotemporal graph convolutional network model, and estimate passenger flow density and issue early warnings of exceeding limits by combining the calculated access preference index.

[0053] The 3D integration and management module is used to integrate real-scene 3D models, cultural tourism thematic data, and real-time visitor flow data, and present them in a 3D visualization platform.

[0054] The data storage and management module is used to store and manage all data and provides a unified data access interface.

[0055] Furthermore, the system also includes a mobile terminal access module, used to provide tourists with real-time scenic area visitor heat maps, personalized route recommendations, and early warning information reception functions.

[0056] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0057] 1. This invention effectively solves the problems of missing details and texture distortion in the modeling of karst landforms and scenic areas by adopting a multi-source data acquisition method that combines UAV oblique photography, 3D laser scanning and close-range photogrammetry, and proposes a hierarchical fusion algorithm.

[0058] 2. This invention achieves accurate prediction of crowd density and traffic trends by constructing a traffic prediction model that integrates video surveillance, tourist flow, and spatiotemporal graph convolutional networks.

[0059] 3. This invention establishes a comprehensive three-dimensional digital twin model by deeply integrating real-world 3D models, cultural tourism data, and real-time monitoring data. Managers can intuitively grasp the real-time status of each area within the 3D scene.

[0060] 4. By introducing an access preference index, this invention can accurately identify tourist behavior patterns and trends in hotspot areas, supporting scenic spots in conducting scientific tourist guidance, resource allocation, and facility planning. Attached Figure Description

[0061] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0062] Figure 1 This is a flowchart illustrating an embodiment of the present invention;

[0063] Figure 2 This is a system schematic diagram according to an embodiment of the present invention;

[0064] Figure 3 This is a real-scene tour map of the scenic area according to an embodiment of the present invention;

[0065] Figure 4 This is a mobile terminal browsing diagram according to an embodiment of the present invention. Detailed Implementation

[0066] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0067] like Figure 1 As shown, the method for real-world 3D modeling of karst landform scenic areas includes the following steps:

[0068] Step S1: Obtain image data of the target scenic area through drone oblique photography, obtain point cloud data through 3D laser scanning, and obtain detailed texture data through close-range photogrammetry.

[0069] Step S2: Construct a hierarchical fusion algorithm to fuse multi-source data and generate a real-world 3D model;

[0070] Step S3: Generate cultural tourism thematic data using web crawling and geocoding technology;

[0071] Step S4: Apply adaptive LOD and GPU rendering acceleration to achieve cross-terminal visualization of real-world 3D models;

[0072] Step S5: Based on video surveillance and tourist flow data, construct a risk assessment model to realize the calculation of crowd density, flow prediction and risk warning;

[0073] Step S6: Integrate real-scene 3D models, cultural tourism data, and real-time visitor flow data to achieve 3D three-dimensional display and management of the entire scenic area.

[0074] The web crawler technology involved in this invention strictly complies with relevant laws and regulations in all data acquisition activities, and follows the following principles: It is only used to obtain publicly available, non-sensitive basic information about scenic spots on relevant geographic information service platforms for the non-commercial scenic spot management purposes described in this patent. When acquiring data, we will strictly adhere to the Robots Exclusion Protocol and related terms of service of the target website. For specific interfaces or services that explicitly prohibit non-human access, this patented technology solution will circumvent them and will not scrape data from such interfaces. The data crawled is limited to the minimum dataset necessary to achieve the purpose of this patented technology, and does not involve any personal privacy, trade secrets or deep content that is strictly protected by copyright; The acquired data will only be used to generate a cultural and tourism-themed data layer for the public management of the scenic area, and will not cause any undue interference to the normal operation of the target website.

[0075] The implementation of this patented technical solution is based on the aforementioned legality and compliance.

[0076] Step S2 specifically includes the following steps:

[0077] Step S2.1: Perform surface reconstruction on the point cloud data acquired by 3D laser scanning to generate a 3D mesh model;

[0078] Step S2.2: Perform resolution hierarchical decomposition on the three-dimensional mesh model;

[0079] Step S2.3: For each resolution level, perform texture fusion on the data from drone photography and close-up photography to obtain a real-world 3D model.

[0080] The surface reconstruction employs the Poisson surface reconstruction algorithm; the resolution hierarchy decomposition uses the Laplacian pyramid for multi-resolution decomposition, with specific steps including:

[0081] The 3D mesh model generates multiple resolution levels through progressive downsampling;

[0082] For each resolution level, the texture image is downsampled to match the geometric resolution of the current level.

[0083] In step S2.3, the texture fusion is based on an energy minimization method, specifically including the following steps:

[0084] The energy function is constructed using the following formula:

[0085]

[0086] in, Represents texture The energy function, This indicates the blended texture in texture coordinates. RGB vector values ​​at that location Indicates from the first An image source in texture coordinates Texture values ​​obtained from sampling at that location, Indicates the image source index. Indicates the total number of image sources. Indicates the first An image source in texture coordinates The fusion weight at the point, Represents the regularization parameter. Represents texture In texture coordinates gradient at, Represents the square of the Euclidean norm;

[0087] The optimal fused texture in texture coordinates is obtained by minimizing the energy function using gradient descent. The RGB vector value at that location.

[0088] The specific steps of using gradient descent to minimize the energy function include:

[0089] 1. Initialize the texture image by taking a weighted average of the corresponding color values ​​of all input texture images at each pixel location, which will serve as the starting point for the texture fusion.

[0090] 2. Calculate the gradient of the energy function E(T) with respect to the texture T. Calculate the gradients of the data term and the smoothing term separately, and add these two gradients together to obtain the total gradient of the energy function at the current texture. The data term is: The smoothing term is: .

[0091] 3. Update the texture T in the opposite direction of the gradient. For each texture coordinate x, subtract the product of the learning rate and the total gradient from the current texture value to obtain the new texture value T_new(x). After the update, it is necessary to ensure that all texture values ​​are within a reasonable range. For example, RGB color values ​​should be limited to 0 to 255. Values ​​outside this range should be truncated. The learning rate is specifically set as follows: the initial learning rate is set to 0.1, and it decays to 0.8 times the original rate every 1000 iterations.

[0092] 4. Repeat the gradient calculation and texture update steps until the convergence condition is met. The convergence condition is set as follows: the magnitude of the gradient vector is less than a preset threshold or the maximum number of iterations is reached. The gradient magnitude threshold is set to 1×10^-5. The maximum number of iterations is set in layers, specifically: the maximum number of iterations for low resolution is set to 1000 times, as low resolution textures have fewer details and converge quickly; the maximum number of iterations for medium resolution is set to 3000 times, as medium detail complexity requires sufficient iterations to ensure fusion quality; and the maximum number of iterations for high resolution is set to 5000 times, as high resolution textures have richer details and require more iterations to reach the optimal state.

[0093] 5. When the algorithm converges, the final texture T is the optimal fused texture in the sense of minimizing the energy function. It ensures a smooth transition between textures while maintaining the texture details of each image source.

[0094] In the energy function, the specific formula for the fusion weight is:

[0095]

[0096] in, Indicates image source exist gradient magnitude at that point This represents the gradient control parameter, typically set to 0.1-0.3, used to adjust the intensity of the impact on sharpness. This represents the angle control parameter, typically set to 0.5-1.0, used to adjust the intensity of the geometric consistency effect. Indicates image source The angle between the shooting angle and the surface normal vector of the 3D model.

[0097] Step S3 specifically includes the following steps:

[0098] Step S3.1: Automatically retrieve relevant text data of the target scenic area from the geographic information service platform using web crawler technology;

[0099] Step S3.2: Perform structured parsing on the collected text data, automatically identify and extract geographical entities including scenic spot names, building signs and street names in the text data;

[0100] Step S3.3: Convert the identified geographic entities into latitude and longitude coordinates to achieve a one-to-one mapping between text data and spatial coordinates;

[0101] Step S3.4: Merge and bind the text data with spatial coordinates with the corresponding positions in the real-world 3D model to generate cultural tourism thematic data.

[0102] Step S4 specifically includes the following steps:

[0103] Step S4.1: Establish a multi-level detail model and dynamically adjust the model according to the view distance;

[0104] Step S4.2: Utilize hardware-accelerated rendering technology to improve rendering efficiency through instantiation rendering, texture compression, and shader optimization;

[0105] The multi-level detail model includes a first level, a second level, and a third level. The first level is applicable to scenes with a viewpoint distance greater than 200 meters, the second level is applicable to scenes with a viewpoint distance of 30-200 meters, and the third level is applicable to scenes with a viewpoint distance less than 30 meters.

[0106] Among them, the number of model facets in the first level is ≤1000 triangular facets, and the texture resolution is ≤512×512 pixels. It is mainly used for the overall overview of the scenic area and rapid navigation and positioning.

[0107] The second-level model has ≤10,000 triangular faces and a texture resolution ≤1024×1024 pixels, and is mainly used for building complex observation and regional planning analysis.

[0108] The third-level model has ≤50,000 triangular faces and a texture resolution ≤2048×2048 pixels, and is mainly used for viewing individual buildings and analyzing details.

[0109] The instantiation rendering includes using instantiation to draw repeated objects to reduce CPU overhead;

[0110] Texture compression includes using texture compression algorithms to reduce video memory usage;

[0111] Shader optimization includes developing efficient shader programs to improve rendering performance.

[0112] In step S5, the calculation of pedestrian density specifically includes:

[0113] The scenic area is divided into multiple zones based on the type of building structure.

[0114] Based on real-time video surveillance data of each area of ​​the scenic area, the number of tourists in each area is output using the YOLOv8 model.

[0115] The population density of each area is calculated based on the number of tourists and the area of ​​each region.

[0116] In step S5, the traffic prediction uses a spatiotemporal graph convolutional network model, which specifically includes the following steps:

[0117] Model the scenic area as a graph structure ,in, This represents a set of nodes, where each node represents a scenic area. Let represent the set of edges, which represents the spatial connectivity between regions. The feature vector of each node at each time step includes: the current time step's pedestrian flow, region area, region carrying capacity, region type code, and weather influence factor.

[0118] Specifically, the area type coding includes: scenic spot = 1, restaurant = 0.7, toilet = 0.5; the weather influence factor includes: sunny day = 1, rainy day = 0.6, extreme weather = 0.3; the node set specifically includes each scenic spot, rest area, entrance and exit, and other building entities, and each node has characteristics such as historical traffic flow, area area, type (such as scenic spot, restaurant, toilet, etc.); the edge set is the connectivity (such as path connection) or distance relationship between areas, which can be represented by an adjacency matrix A. If two areas are adjacent or directly connected, the corresponding edge is 1, otherwise it is 0.

[0119] The node feature vectors from past time steps are input into the spatiotemporal graph convolutional network model, and the output is the predicted pedestrian flow for future time steps.

[0120] The access preference index for each area is calculated based on historical visitor traffic data. The specific formula is as follows:

[0121]

[0122] in, Indicates the region Access preference index, area At a historical moment The actual number of people, This indicates the length of the historical time window, taken as 24 hours. Indicates time Maximum pedestrian traffic in all areas Indicates the current moment. This represents the time index within the historical time window. Indicates the region Variance of pedestrian flow over 24 hours This represents the maximum variance of pedestrian flow across all regions. The constant representing the division by zero is taken as 10^-6;

[0123] The output of the spatiotemporal graph convolutional network model is combined with the access preference index to obtain the final traffic prediction result, specifically using the following formula:

[0124]

[0125] in, This represents the final flow prediction result at time t+1. This represents the original predicted output vector of the spatiotemporal graph convolutional network model at time t+1. This represents the normalized access preference index vector for each region. This represents the carrying capacity vector for each region. It represents the Hadamah accumulation. and These represent the corresponding weights of the spatiotemporal graph convolutional network model output and the access preference index, respectively.

[0126] The Spatiotemporal Graph Convolutional Network (STGCN) consists of spatiotemporal convolutional blocks, each containing two temporal convolutional layers and one spatial graph convolutional layer. The temporal gated convolutional layers are used to capture temporal dynamics, and the spatial graph convolutional layers are used to capture spatial dependencies.

[0127] The training of a spatiotemporal graph convolutional network specifically includes the following stages:

[0128] 1. Data preprocessing stage: Normalize historical traffic data; organize the data into samples, each sample including a traffic sequence within a time window as input, and the traffic at the next time step as the prediction target.

[0129] 2. Training phase: Input training data into the model and calculate the predicted value through forward propagation; use mean squared error to calculate the difference between the predicted value and the true value; update the model parameters through backpropagation and gradient descent algorithms until the model converges or reaches the predetermined number of training rounds.

[0130] Based on experimental verification, the weights are set as follows: A weight of 0.75 is assigned to STGCN for its spatiotemporal dynamic prediction, giving it a relatively high weight. The value is set to 0.25 to preserve the long-term pattern of preference.

[0131] like Figure 2 As shown, the real-scene 3D management system for karst landform scenic areas is implemented based on any one of the aforementioned real-scene 3D modeling methods for karst landform scenic areas, including:

[0132] The multi-source data acquisition module is used to collect aerial multi-angle images, point cloud data, and detailed texture data of the scenic area;

[0133] The hierarchical fusion processing module is used to perform three-dimensional mesh reconstruction on the point cloud data and execute multi-resolution geometric decomposition and texture fusion algorithms to generate a real-world three-dimensional model by minimizing the energy function.

[0134] The cultural and tourism themed data generation module is used to crawl text data using web crawlers, extract place name entities through a named entity recognition model, and bind them with three-dimensional spatial coordinates to generate cultural and tourism themed data.

[0135] The risk assessment and early warning module is used to access video surveillance data and historical passenger flow data, predict passenger flow in each area by constructing a spatiotemporal graph convolutional network model, and estimate passenger flow density and issue early warnings of exceeding limits by combining the calculated access preference index.

[0136] The 3D integration and management module is used to integrate real-scene 3D models, cultural tourism thematic data, and real-time visitor flow data, and present them in a 3D visualization platform.

[0137] The data storage and management module is used to store and manage all data and provides a unified data access interface.

[0138] The system also includes a mobile terminal access module, which provides tourists with real-time heat maps of visitor flow in scenic areas, personalized route recommendations, and early warning information reception functions.

[0139] like Figure 3As shown, this visualization page comprehensively utilizes scenic area tilt model data, refined model data, 360-degree panoramic data, and cultural tourism thematic data. Through terrain-level, city-level, and component-level data scene fusion and lightweight processing, it achieves a full-area, multi-level, three-dimensional display of the target scenic area (such as a typical karst landform scenic area). It presents the surrounding terrain and internal layout of the scenic area in an integrated air-space-ground manner, and meticulously restores the architectural structural features, architectural forms, and exterior details of the target scenic area at a 1:1 scale. By accessing tourist flow data, it can provide spatiotemporal correlation services, offering analysis of tourist flow in different areas of the scenic area, as well as traffic rankings for accommodation / dining / shopping / entertainment. This facilitates scenic area managers in understanding the scenic area's hotspots and operational trends, accurately analyzing tourist preferences, and providing a more scientific basis and data support for subsequent tourist guidance, resource allocation, and investment attraction.

[0140] like Figure 4 As shown, the scenic area's 3D real-scene mini-program utilizes 3D real-scene technology and lightweight processing technology to port the core functions of the web-based real-scene restoration module to the mini-program, achieving cross-terminal interconnection. This allows tourists to enjoy a comprehensive, multi-angle, 3D real-scene browsing experience on their mobile devices from high-altitude, low-altitude, near-ground, and ground-level horizontal perspectives, enhancing the cultural, interactive, and fun aspects. The mini-program embeds real-scene 3D scenes of various key scenic spots and configures path navigation, point marking, multimedia integration, and panoramic integration within these scenes, allowing tourists to browse and navigate in 3D space anytime, anywhere. This facilitates tourists' efficient access to scenic area cultural information and enhances their tourism experience.

[0141] The examples described herein are merely preferred embodiments of the invention and are not intended to limit the concept and scope of the invention. Any modifications and improvements made by those skilled in the art to the technical solutions of the invention without departing from the design concept of the invention should fall within the protection scope of the invention.

Claims

1. A method for real scene three-dimensional modeling for karst landform scenic spots, characterized in that, The method comprises the following steps: Step S1, obtaining target scenic spot image data by unmanned aerial vehicle oblique photography, obtaining point cloud data by three-dimensional laser scanning, and obtaining detail texture data by close-range photogrammetry; Step S2, constructing a hierarchical fusion algorithm to fuse the multi-source data and generate a real scene three-dimensional model; Step S3, generating travel theme data by using network crawler and geographic coding technology; Step S4, realizing cross-terminal visualization of the real scene three-dimensional model by using adaptive LOD and GPU rendering acceleration; Step S5, constructing a risk assessment model based on video monitoring and tourist flow data to realize crowd density calculation, flow prediction and risk warning; Step S6, integrating the real scene three-dimensional model, travel theme data and real-time tourist flow data to realize three-dimensional display and management of the whole scenic spot.

2. The method of claim 1, wherein, The step S2 specifically comprises the following steps: Step S2.1, surface reconstruction is performed on the point cloud data obtained by three-dimensional laser scanning to generate a three-dimensional mesh model; Step S2.2, resolution level decomposition is performed on the three-dimensional mesh model; Step S2.3, for each resolution level, texture fusion is performed on the data of unmanned aerial vehicle photography and close-range photography to obtain a real scene three-dimensional model.

3. The method of claim 2, wherein, In the step S2.3, the texture fusion is based on an energy minimization method, specifically comprising the following steps: An energy function is constructed, and the specific formula is: ； in, Represents texture The energy function, This indicates the blended texture in texture coordinates. RGB vector values ​​at that location Indicates from the first An image source in texture coordinates Texture values ​​obtained from sampling at that location, Indicates the image source index. Indicates the total number of image sources. Indicates the first An image source in texture coordinates The fusion weight at the point, Represents the regularization parameter. Represents texture In texture coordinates gradient at, Represents the square of the Euclidean norm; The optimal fused texture at the texture coordinates is obtained by solving the minimization of an energy function using a gradient descent method to get the RGB vector values.

4. The method of claim 3, wherein, In the energy function, the specific formula of the fusion weight is: ； wherein, denotes the image source At the gradient magnitude at the location, denotes a gradient control parameter for adjusting the strength of the sharpness influence, denotes an angle control parameter for adjusting the strength of the geometric consistency influence, denotes the image source the angle between the camera view direction and the surface normal of the three-dimensional model.

5. The method of claim 4, wherein, The step S3 specifically comprises the following steps: Step S3.1, automatically capturing target scenic spot related text data from a geographic information service platform by using network crawler technology; Step S3.2, structurally analyzing the collected text data, automatically identifying and extracting geographic entities including scenic spot names, building identifiers and street names in the text data; Step S3.3, converting the identified geographic entities into latitude and longitude coordinates to realize one-to-one mapping of the text data and spatial coordinates; Step S3.4, fusing and binding the text data with spatial coordinates with the corresponding positions in the real scene three-dimensional model to generate travel theme data.

6. The method of claim 5, wherein, The step S4 specifically comprises the following steps: Step S4.1, establishing a multi-level detail model and dynamically adjusting the model according to the view distance; Step S4.2, using hardware acceleration rendering technology to improve rendering efficiency through instance rendering, texture compression and shader optimization; The multi-level detail model comprises a first level, a second level and a third level, wherein the first level is suitable for scenes with a view distance greater than 200 meters, the second level is suitable for scenes with a view distance of 30-200 meters, and the third level is suitable for scenes with a view distance less than 30 meters.

7. The method of claim 6, wherein, In the step S5, the crowd density calculation specifically comprises: Dividing the scenic spot into multiple regions according to building entity types; Based on the real-time video monitoring data of each region of the scenic spot, using a YOLOv8 model to output the number of tourists in each region; Based on the number of tourists and the area of each region, calculating the crowd density of each region.

8. The method of claim 7, wherein, In the step S5, the flow prediction adopts a spatiotemporal graph convolution network model, specifically comprising the following steps: Modeling scenic spots as graph structure wherein, represents a set of nodes, each node representing a scenic area, represents a set of edges, representing spatial connectivity between areas, wherein the feature vector of each node at each time step includes: current time step human flow, area size, area carrying capacity, area type encoding and weather influence factor; Input the node feature vector of the past time step into the spatio-temporal graph convolution network model, and output the predicted passenger flow at the future time step; Based on the historical passenger flow data, calculate the access preference index of each area, and the specific formula is: ； wherein, the access preference index of a region, the region the actual flow of people at a historical time, denotes the length of the historical time window, taken as 24 hours, denotes the time, the maximum flow of people in all regions, denotes the current time, denotes the time index within the historical time window, the region the variance of the flow of people in 24 hours, denotes the maximum value of the flow variance of all regions, denotes a constant to prevent division by zero;​​ Combine the output of the spatio-temporal graph convolution network model with the access preference index to obtain the final flow prediction result, and the specific formula is: ； wherein, denotes the final flow prediction result at t+1, denotes the original prediction output vector of the spatio-temporal graph convolution network model at t+1, denotes the normalized access preference index vector of each region, denotes the carrying capacity vector of each region, denotes the Hadamard product, and denote the corresponding weights of the spatio-temporal graph convolution network model output and the access preference index, respectively.

9. A real scene three-dimensional management system for karst landform scenic spots, which is realized based on the real scene three-dimensional modeling method for karst landform scenic spots according to any one of claims 1-8, characterized in that, It comprises: A multi-source data acquisition module is used to acquire aerial multi-angle images, point cloud data and detail texture data of the scenic spot; A hierarchical fusion processing module is used to perform three-dimensional grid reconstruction on the point cloud data, and execute multi-resolution geometric decomposition and texture fusion algorithm, and generate a real scene three-dimensional model by minimizing an energy function; A travel theme data generation module is used to use a network crawler to capture text data, extract place name entities through a named entity recognition model, and bind with three-dimensional space coordinates to generate travel theme data; A risk assessment and early warning module is used to access video monitoring data and historical passenger flow data, predict passenger flow in each area through a spatio-temporal graph convolution network model, and estimate passenger flow density and give early warning of over-limit risk in combination with the calculated access preference index; A three-dimensional integration and management module is used to integrate real scene three-dimensional model, travel theme data and real-time passenger flow data, and present in a three-dimensional visualization platform; A data storage and management module is used to store and manage all data, and provide a unified data access interface.

10. The system of claim 9, wherein, The system further comprises a mobile terminal access module, which is used to provide real-time scenic area passenger flow heat map, personalized route recommendation and early warning information receiving function for tourists.

Images