Urban three-dimensional model rendering method and system

Abstract

The invention discloses a city three-dimensional model rendering method, which comprises the following steps of: establishing a multi-level detail model system to create a plurality of three-dimensional models with different levels of detail for each building object in a city scene so as to form a continuous hierarchical structure from high precision to low precision; spatial distance, view angle factors and object importance are comprehensively considered, effective distance measurement from a view point to a target object is calculated, a continuous LOD switching function is established according to the effective distance measurement, and the distance is mapped into continuous detail level values; a smooth interpolation algorithm based on time is applied, progressive switching is carried out between adjacent LOD layers, and visual jump during switching is eliminated; visual jump of LOD switching in urban three-dimensional model rendering is remarkably reduced, the stability of rendering performance is improved, and smooth and continuous visual experience is achieved.

Description

Technical Field

[0001] The present invention relates to the technical fields of computer graphics and three-dimensional rendering, and particularly to a method and system for rendering urban three-dimensional models. Background Art

[0002] In order to achieve smooth real-time rendering with limited hardware resources, the multi-level of detail (LOD) technology is widely applied to the rendering optimization of large-scale three-dimensional scenes; traditional LOD technology creates geometric models with different precision levels for the same object and dynamically selects an appropriate level of detail for rendering according to the viewing distance, thereby significantly reducing the rendering load while ensuring the visual effect.

[0003] However, the existing LOD switching methods generally have the following technical defects: using the simple Euclidean distance as the sole criterion for LOD selection, failing to fully consider the influence of viewing angle, object importance, screen projection area, etc. on visual perception, resulting in a too rough LOD selection strategy and unable to achieve precise detail control; adopting a direct switching method between adjacent LOD levels, which will cause obvious geometric jumps and texture changes at the moment of switching, seriously affecting the user's visual experience, especially when the camera moves or zooms quickly, this visual discontinuity is more obvious; in addition, the existing technology lacks an effective adaptive optimization mechanism and cannot dynamically adjust the LOD switching strategy according to device performance and scene complexity, resulting in a serious drop in frame rate on devices with limited performance, while on devices with sufficient performance, the hardware resources cannot be fully utilized. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a method and system for rendering urban three-dimensional models, which realizes a visually continuous and performance-stable LOD switching effect by establishing an effective distance metric model that comprehensively considers spatial distance, viewing angle, and importance weight, and combining a time-based smooth interpolation algorithm.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] In the first aspect, the present invention provides a method for rendering urban three-dimensional models, and the method includes the following steps:

[0007] Step S1, establishing a multi-level detail model system, creating three-dimensional models with multiple different levels of detail for each building object in the urban scene, and forming a continuous hierarchical structure from high precision to low precision.

[0008] The at least three different levels of detail are at least three.

[0009] In step S2, considering spatial distance, viewing angle factors, and object importance comprehensively, an effective distance metric from the viewpoint to the target object is calculated. Based on the effective distance metric, a continuous LOD switching function is established to map the distance to a continuous level of detail value.

[0010] , where is the effective distance of the target object, is the Euclidean distance from the observation point to the object center, is the angle between the line of sight direction and the main axis of the object, is the importance weight determined based on object semantics and geometric features.

[0011] ; where is the LOD switching function, and are the highest and lowest level of detail indices respectively, and are the preset distance threshold boundaries;

[0012] In step S3, an interpolation algorithm based on time smoothing is applied to perform progressive switching between adjacent LOD levels to eliminate visual jumps during switching.

[0013] ; where is the currently actually rendered LOD level value, is the level before switching, is the target level, is a smoothing interpolation function to ensure the continuity of the first derivative, , is a normalized transition time parameter, and visual smooth switching is achieved by controlling the transition duration.

[0014] Furthermore, the effective distance metric is adaptively corrected for the screen projection area to consider the influence of the actual visible size of the object on the screen on LOD selection: , where is the total screen pixel area of the current rendering target, is the projected pixel area of the object on the screen plane; this correction factor ensures that smaller projected objects use lower levels of detail and larger projected objects use higher levels of detail, thus optimizing the allocation efficiency of rendering resources.

[0015] Furthermore, the object importance weight is calculated through a weighted combination of multi-dimensional factors to implement a differentiated LOD strategy for different types of buildings:

[0016] , where is the semantic importance weight, and different values are preset according to the functional types of buildings (such as landmark buildings, commercial buildings, residential buildings), with landmark buildings having the highest weight; is the geometric complexity weight, which is calculated based on the number of polygons and the complexity of geometric features of the building model. The higher the complexity, the greater the weight; is the user attention weight, which is statistically obtained based on historical user interaction data (such as clicks, dwell time, zoom operations). The higher the attention of a building, the higher the weight; The constraint condition is: , ensuring the normalization of the weights, is the constraint condition coefficient, which can be dynamically adjusted according to the application scenario.

[0017] Furthermore, the specific calculation of the smoothing interpolation function is:

[0018] , where and are the tangent vector parameters of the starting point and the ending point respectively, , , are the basis functions of the smoothing interpolation function respectively:

[0019] , , .

[0020] By adjusting these parameters, the shape of the transition curve can be controlled, making the LOD switching process show a change law that is more in line with the human eye perception characteristics.

[0021] Furthermore, real-time rendering is implemented based on the GPU parallel computing architecture. The specific processing flow of LOD is as follows:

[0022] Step S21, divide the entire city model into regular grid cells according to the spatial position. Each cell independently performs LOD calculation and rendering processing, where The value of is dynamically determined according to the scene complexity and GPU performance, and the preset value is 50 meters.

[0023] Step S22, parallelly calculate the LOD decision matrix for each grid cell. The decision matrix stores the optimal LOD level of each object in the cell .

[0024] Among them, represents the grid coordinates, is the candidate integer LOD level.

[0025] Step S23, implement batch geometry subdivision and vertex data blending in the GPU shader, and perform LOD interpolation calculation output for multiple vertices through parallel processing .

[0026] , where is the boolean index mask matrix of the level, is the vertex attribute array corresponding to the LOD level, is the blending weight calculated based on the time parameter , and all weights satisfy the normalization condition .

[0027] Furthermore, continuously monitor the rendering performance metrics and define the performance ratio : , where is the preset target frame rate, which is 60fps or 30fps, is the actual rendering frame rate at the current moment.

[0028] Dynamically adjust the distance thresholds of all LOD levels based on the performance ratio to achieve adaptive control of the rendering load:

[0029] ;

[0030] where and are the old and new values of the th threshold respectively, , is the adjustment intensity parameter to ensure smooth threshold changes and avoid sudden LOD level jumps.

[0031] Smooth the adjusted thresholds to prevent visual instability caused by frequent fluctuations, and use a moving average filter to smooth the thresholds of multiple consecutive frames.

[0032] Furthermore, prepare the required level of detail in advance by predicting the user's viewing path, and reduce the calculation delay during switching based on the future time LOD requirements predicted by the camera movement.

[0033] Specifically, it includes the following steps:

[0034] Step S31, establish a prediction model for camera movement, and predict the future position based on the current position, speed, and acceleration information:

[0035] , where are the position vector, speed vector, and acceleration vector of the camera in three-dimensional space respectively, The prediction time window is set to 0.1 - 0.5 seconds.

[0036] Step S32: Calculate the LOD requirements of all objects within the future frustum based on the predicted camera position:

[0037] ; where is the position of each object in the scene, and candidate objects within the frustum are quickly determined through spatial indexing.

[0038] Step S33: Asynchronously preload in a background thread the LOD level data required for prediction but not currently activated, including geometric meshes and texture resources, so that they can be immediately used when actually needed, significantly reducing the loading delay and stuttering during LOD switching.

[0039] Furthermore, a corresponding texture resolution level is established for each LOD level k , maintaining the matching of geometric complexity and texture accuracy:

[0040] ; where is the base texture resolution used for the highest level of detail (such as 4096×4096), , is the texture scaling factor; controlling the attenuation rate of texture resolution with the LOD level.

[0041] During the LOD switching process, progressive blending of textures is synchronized to avoid visual discontinuity caused by texture changes: ; where is the texture data at the level, is the normalized texture coordinate.

[0042] According to the anisotropic filtering algorithm, texture aliasing during distant viewing is reduced:

[0043] ; where ([[]] , ) is the sampling step vector calculated along the main gradient direction of the texture, and N is the number of anisotropic samples, which is dynamically adjusted according to the viewing angle and distance to ensure clear texture effects at various viewing angles.

[0044] Based on the same inventive concept, in a second aspect, the present invention provides a three-dimensional urban model rendering system for executing the method of the first aspect. The system includes: a model management module, a distance metric calculation module, an LOD decision module, a smooth interpolation module, and a rendering output module.

[0045] The model management module is used to construct and maintain multi-level detail models for each building object in the urban scene. The model management module includes a model preprocessing unit, a LOD level generation unit, and a model indexing unit. Among them, the model preprocessing unit is responsible for geometric optimization and topological simplification of the original 3D model. The LOD level generation unit automatically generates at least 3 model versions with different levels of detail based on the principle of decreasing geometric complexity. The model indexing unit establishes the mapping relationship between the model identifier and its respective LOD levels.

[0046] The distance metric calculation module is used to calculate the effective distance metric from the observation point to the target object in real time.

[0047] The LOD decision module is used to determine the target level of detail for each object based on the effective distance metric.

[0048] The smoothing interpolation module is used to achieve smooth temporal transitions between adjacent LOD levels.

[0049] The rendering output module is used to convert the processed LOD data into the final visual output. This module includes a geometric data blending unit, a shader scheduling unit, and a frame buffer management unit. Among them, the geometric data blending unit synthesizes the final vertex and index data according to the interpolation weight. The shader scheduling unit coordinates the allocation of GPU computing resources. The frame buffer management unit controls the output and display of the rendering result.

[0050] Furthermore, the system also includes a user interaction analysis module, which includes an interaction behavior recording unit, an attention degree statistics unit, and a personalized weight adjustment unit.

[0051] The interaction behavior recording unit tracks the operation behaviors of the user, such as clicks, zooms, stays, etc. The attention degree statistics unit analyzes the user's attention patterns to different regions and buildings. The personalized weight adjustment unit dynamically adjusts the object importance weights according to the user's preferences to implement a personalized LOD optimization strategy.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] By introducing technical means such as effective distance metrics and smoothing interpolation algorithms, the present invention significantly reduces the visual jumps during LOD switching in the rendering of urban 3D models, improves the stability of rendering performance, and achieves a smooth and continuous visual experience. Compared with traditional methods, the present invention can reduce the visual discontinuity caused by LOD switching by more than 80%, and reduce the rendering frame rate fluctuation by more than 60%. Description of the Drawings

[0054] Figure 1 It is a flowchart of a method for rendering an urban 3D model of the present invention;

[0055] Figure 2 Schematic diagram of the composition of a 3D model rendering system for cities according to the present invention. Specific implementation manner

[0056] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be described clearly and completely below. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0057] Embodiment 1

[0058] As Figure 1 shown, it is a flowchart of a method for rendering a 3D model of a city according to the present invention, and the method includes the following steps:

[0059] Step S1, establish a multi-level detail model system, create multiple 3D models with different levels of detail for each building object in the urban scene, and form a continuous hierarchical structure from high-precision to low-precision.

[0060] It should be noted that the multiple different levels of detail are at least 3; the multi-level detail model system created by the present invention has hierarchical geometric simplification characteristics, and the transition between each level is smooth and natural, effectively ensuring visual consistency; for a standard office building, 5 LOD levels can be designed in this embodiment: LOD0 (original high-precision model, about 50,000 polygons) includes complete building facade details, window frames, balcony railings, etc.; LOD1 (about 20,000 polygons) retains the main facade structure but simplifies window details; LOD2 (about 8,000 polygons) combines similar texture areas and simplifies protruding structures; LOD3 (about 3,000 polygons) only retains the building body outline and main dividing lines; LOD4 (about 1,000 polygons or less) is represented as a simple geometric body, only retaining the building volume and basic shape. This multi-level model design can provide a visually coherent building performance at different viewing distances.

[0061] Multi - level detail design usually includes 5 - 7 LOD levels. This setting can provide a smoother transition effect when the viewing distance changes. For example, for a landmark building in a city (such as the town hall), 7 fine - grained LOD levels can be designed, including: LOD0 (full details, 100,000+ polygons), LOD1 (80,000 polygons), LOD2 (50,000 polygons), LOD3 (25,000 polygons), LOD4 (10,000 polygons), LOD5 (4,000 polygons), and LOD6 (1,500 polygons). The geometric complexity between each level usually follows a decreasing ratio of about 1 / 2 to 1 / 3. This decreasing pattern has been verified by a large number of experiments and can achieve the best balance between visual quality and performance.

[0062] In step S2, considering spatial distance, viewing angle factors, and object importance comprehensively, calculate the effective distance metric from the viewpoint to the target object, and establish a continuous LOD switching function based on the effective distance metric, mapping the distance to a continuous level - of - detail value.

[0063] By calculating the effective distance metric, the present invention can accurately capture the detail requirements perceived by the human eye for buildings observed at different positions and angles. For example, when an observer views a museum building closely from the front, the building may need to use LOD1 or LOD0 levels to display its fine facade details; while when the same observer views the same building at the same distance from a 45 - degree side angle, due to perspective effects and viewing angle factors, the system may select the LOD2 - level model. This intelligent selection not only ensures visual quality but also improves rendering efficiency. For different types of buildings, the system presets different basic importance weights, such as landmark buildings (weight value 1.2), commercial centers (weight value 1.0), and ordinary residences (weight value 0.8).

[0064] , where is the effective distance of the target object, is the Euclidean distance from the observation point to the center of the object, is the angle between the line - of - sight direction and the main axis of the object, is the importance weight determined based on the object's semantics and geometric features. Taking an actual urban scene as an example, when an observer is at ground - level position and observes a skyscraper at a distance of 500 meters, if the viewing angle is from the front ( = 0°), then = 1; if observing at a 30 - degree elevation angle ( = 30°), then , making the effective distance calculation be meters.

[0065] For different types of buildings, the importance weights also vary: landmark buildings (such as the city landmark tower) can be set to 1.5, indicating that a higher-precision model is required at the same distance; ordinary commercial buildings can be set to 1.0; while simple supporting facility buildings can be set to 0.8, indicating that a lower-precision model can be used, ensuring that important buildings always maintain sufficient visual detail.

[0066] ; among them, is the LOD switching function, and are the highest and lowest detail level indices respectively, and are the preset distance threshold boundaries.

[0067] The setting of the distance threshold boundary has a decisive impact on the LOD switching effect. Taking the building complex in the urban core area as an example, a typical configuration can be: = 5 (highest detail level), = 0 (lowest detail level), = 100 meters (close distance threshold), = 1000 meters (far distance threshold); when the effective distance between the observer and the building is less than 100 meters, the system will use the highest-precision model (LOD5); when the distance is greater than 1000 meters, the lowest-precision model (LOD0) will be used; between 100 and 1000 meters, the system will smoothly transition to use intermediate-precision models, and the precision decreases linearly with the increase of the distance. Verified by experiments, this configuration can achieve a good balance between visual quality and rendering performance in the modern urban scale (about 5 - 10 square kilometers core area).

[0068] For cities or special areas of different scales, these parameters can be adjusted appropriately. For example, in a dense historical block, can be set to 50 meters to provide more refined close-distance details.

[0069] Step S3, apply a time-based smooth interpolation algorithm to perform progressive switching between adjacent LOD levels, eliminating visual jumps during switching.

[0070] In traditional systems, LOD level switching often results in obvious "jumping" or "flickering" effects, seriously affecting the immersion; the smooth interpolation algorithm of the present invention adopts a time-based progressive transition mechanism, effectively eliminating this visual defect.

[0071] Specifically, for a building that switches from LOD2 to LOD3, this algorithm does not immediately replace the model. Instead, within a time window of approximately 250 - 500 milliseconds, it gradually mixes the vertex data of the two models. This mixing process visually presents as a smooth "dissolving" effect. For users, the details of the building seem to change naturally rather than suddenly jump. The smooth transition technology can reduce the perceptibility of LOD switching by approximately 95%, significantly enhancing visual continuity and the immersive experience.

[0072] ; where is the current actual rendered LOD level value, is the level before switching, is the target level, is a smooth interpolation function that ensures the continuity of the first derivative, , is the normalized transition time parameter, which realizes a smooth visual switch by controlling the transition duration.

[0073] In a typical urban browsing scenario, and tend to be integer values (such as 3 and 4), but is a continuous floating - point number, representing the mixing state between two discrete LOD models. For example, when t = 0.5, might be 3.5, indicating the intermediate state between LOD3 and LOD4.

[0074] When setting the transition duration, for normal viewing speeds, a transition duration of 300 milliseconds (i.e., it takes 300 milliseconds from t = 0 to t = 1) usually provides the best smooth experience; for a fast - moving camera (such as in the fast - flight mode), it can be shortened to 150 milliseconds to maintain responsiveness; for slow viewing or close - up shots, it can be extended to 500 milliseconds to obtain a more refined transition effect.

[0075] Perform an adaptive correction on the screen projection area of the effective distance metric to consider the influence of the actual visible size of the object on the screen on LOD selection: , where is the total screen pixel area of the current rendering target, is the projected pixel area of the object on the screen plane; this correction factor ensures that smaller projected objects use a lower level of detail, and larger projected objects use a higher level of detail, thereby optimizing the allocation efficiency of rendering resources.

[0076] Building sizes vary significantly - from small pavilions (a few meters wide) to super-large skyscrapers (hundreds of meters high). Using only distance as the basis for LOD selection can lead to insufficient details for small buildings or waste of resources for large buildings. Take a practical example: on a monitor with a standard resolution of 1920×1080 (about 2.07 million pixels), a large building that occupies 5% of the screen area (about 104,000 pixels) and a small building that occupies 0.5% of the screen area (about 10,400 pixels), even if the actual distances are the same, different LOD levels should be adopted; according to the formula, the correction factor for the large building is while the correction factor for the small building is . This indicates that smaller buildings can use models with lower precision (equivalent to the distance increasing by 14.15 times), while larger buildings need to maintain higher precision (equivalent to the distance only increasing by 4.47 times). This adjustment significantly improves the rendering efficiency while maintaining good visual quality.

[0077] The importance weight of the object described is calculated through the weighted combination of multi-dimensional factors to achieve a differentiated LOD strategy for different types of buildings:

[0078] , where is the semantic importance weight, with different preset values according to the functional type of the building (such as landmark buildings, commercial buildings, residential buildings), and landmark buildings have the highest weight; is the geometric complexity weight, calculated through the number of polygons and the complexity of geometric features of the building model, and the higher the complexity, the greater the weight; is the user attention weight, statistically obtained based on historical user interaction data (such as clicks, dwell time, zoom operations), and the higher the attention of the building, the higher the weight; the constraint condition is: to ensure the normalization of the weights, is the constraint condition coefficient, which can be dynamically adjusted according to the application scenario.

[0079] The semantic importance weight can be classified according to the building function: landmark / historic building = 1.5, government / cultural facility = 1.3, commercial center = 1.1, ordinary office building = 1.0, residential area = 0.8, auxiliary facility = 0.6. The geometric complexity weight is automatically calculated according to the model characteristics: a Gothic building with a finely carved facade is about 1.4, a modern building with a complex facade is about 1.2, a standard glass curtain wall building is about 1.0, and a building with a simple geometric shape is about 0.8. The user attention weight It is dynamically adjusted based on interaction data: The shopping mall that users often visit may rise from the default value of 1.0 to 1.3; The building in the far suburbs that users have never interacted with may drop to 0.7. The coefficient The typical settings are: (Semantic importance dominates), (Geometric complexity is secondary), (User attention assists in adjustment).

[0080] Smoothing interpolation function The specific calculation is:

[0081] , where and are the tangent vector parameters of the starting point and the ending point respectively, , , are the basis functions of the smoothing interpolation function respectively:

[0082] , , .

[0083] By adjusting these parameters, the shape of the transition curve can be controlled, making the LOD switching process present a change law that is more in line with the human eye perception characteristics.

[0084] The parameters and control the tangent direction and magnitude of the curve at the starting point and the ending point, and have a significant impact on the transition rate. In the standard configuration, these two parameters are usually set to: , , This set of parameter values makes the transition start slowly (reducing the initial jump feeling), accelerate in the middle (improving responsiveness), and decelerate again at the end (smoothing the ending).

[0085] Experiments show that this "slow-fast-slow" transition curve reduces about 40% of the visual discomfort compared with the linear transition. For different viewing conditions, the system can also dynamically adjust these parameters: For example, for a fast-moving camera, can be reduced to 0.5 to obtain a faster initial response; For LOD switching with large detail changes (such as from LOD4 to LOD1), can be increased to 0.7 to extend the ending transition time and further reduce visual mutations.

[0086] Real-time rendering is implemented based on the GPU parallel computing architecture. The processing flow of LOD is specifically as follows:

[0087] Step S21, divide the entire city model into Regular grid cells, each cell independently performs LOD calculation and rendering processing, where The value is dynamically determined according to the scene complexity and GPU performance, and the preset value is 50 meters.

[0088] In a typical urban center area (building density is about 150 - 200 buildings per square kilometer), the grid size is usually set to 50 meters × 50 meters. This size can ensure an appropriate number of buildings in each grid (an average of 5 - 8 buildings) and can cooperate efficiently with the frustum culling algorithm. In the suburbs with a lower building density, the grid size can be automatically expanded to 100 meters × 100 meters to reduce management overhead; while in the ultra-high-density commercial center area, it can be reduced to 30 meters × 30 meters to improve culling efficiency.

[0089] Step S22, parallelly calculate the LOD decision matrix for each grid cell The decision matrix stores the optimal LOD level of each object within the cell .

[0090] Among them, represents the grid coordinates, is the candidate integer LOD level.

[0091] For a typical 50 meters × 50 meters grid cell, which contains about 5 - 8 buildings, the system will complete the LOD calculation of all buildings in a single calculation schedule. The decision matrix is usually stored in the GPU memory in a two-dimensional texture format, where $(i,j)$ corresponds to the grid coordinates, and the texture value stores the optimal LOD level of the building at that position; this GPU acceleration strategy enables the LOD decision calculation to be completed in less than 1 millisecond even in a large urban scene containing 10,000+ buildings; on high-end GPUs (such as NVIDIA RTX 3080 or higher), the system can increase the grid precision to 10 meters × 10 meters (about a 25-fold increase in the number of grids) and still keep the LOD decision calculation time per frame below 3 milliseconds.

[0092] Step S23, implement batch geometry subdivision and vertex data blending in the GPU shader, and achieve the output of LOD interpolation calculation for multiple vertices through parallel processing .

[0093] , among them, is the Boolean index mask matrix at the level, is the vertex attribute array corresponding to the LOD level, is the blending weight calculated based on the time parameter , and all weights satisfy the normalization condition 。

[0094] The system uses a specially optimized Geometry Shader or Tessellation Shader for efficient processing. For the transition of a standard office building from LOD2 (8,000 polygons) to LOD1 (20,000 polygons), instead of directly switching the model, the system dynamically calculates the blending weights according to the time parameter t within a 300-millisecond transition window , for example, when t = 0.5 , 。

[0095] To optimize performance, this embodiment adopts a batch processing technology to simultaneously process the blending calculations of up to 256 vertices, making full use of the SIMD (Single Instruction Multiple Data) architecture advantage of the GPU; during processing, the mask matrix ensures that only relevant vertices participate in the calculations, avoiding invalid blending operations. On typical hardware configurations (such as NVIDIA RTX 2070 or equivalent performance), it is possible to simultaneously process the LOD blending of approximately 50 - 100 buildings, while the additional computational overhead only increases the vertex processing time by about 10 - 15%.

[0096] Continuously monitor the rendering performance metrics and define the performance ratio : , where is the preset target frame rate, which is 60fps or 30fps is the actual rendering frame rate at the current moment

[0097] The monitoring uses the sliding window average method to update the performance evaluation every 100 milliseconds (about 6 frames, at 60fps) to avoid excessive adjustment caused by single-frame fluctuations. The target frame rate can be preset according to different application scenarios and device capabilities: for high-performance desktop applications, it is usually set to 60fps to provide a smooth experience; for mobile devices or web applications, it can be set to 30fps to balance performance and power consumption. The system also provides an adaptive target frame rate function: for example, on a battery-powered laptop, when the battery capacity is detected to be less than 20%, the system can automatically reduce the target frame rate to 30fps to extend the usage time; when connected to an external power supply, it will then return to 60fps. Dynamically adjust the distance thresholds of all LOD levels based on the performance ratio to achieve adaptive control of the rendering load:

[0098] ;

[0099] where and are respectively the The old and new values of a threshold To adjust the intensity parameter to ensure a smooth change in the threshold and avoid sudden LOD level jumps.

[0100] Taking a mid-range hardware configuration (such as NVIDIA GTX 1660) as an example: When the user enters a densely built-up area and the frame rate drops from 60fps to 48fps, the performance ratio P(t) = 60 / 48 = 1.25, which is higher than the 1.1 threshold. The system will apply the first rule: This means that all LOD distance thresholds increase by 2.5%, resulting in more buildings using lower-detail models, thus reducing the rendering load; this minor adjustment is hardly noticeable for a single building, but in an urban scene with thousands of buildings, the cumulative effect can increase the frame rate by approximately 10 - 15%.

[0101] Smooth the adjusted threshold to prevent visual instability caused by frequent fluctuations. Use a moving average filter to smooth the thresholds of consecutive multiple frames.

[0102] Prepare the required level of detail in advance by predicting the user's viewing path. Based on the future time LOD requirements predicted from the camera movement, reduce the computational latency during switching.

[0103] Specifically, it includes the following steps:

[0104] Step S31, establish a prediction model for camera movement, and predict the future position based on the current position, speed, and acceleration information:

[0105] , where are the position vector, speed vector, and acceleration vector of the camera in three-dimensional space respectively, is the prediction time window, set to 0.1 - 0.5 seconds.

[0106] Step S32, calculate the LOD requirements for all objects within the future frustum based on the predicted camera position:

[0107] ; where is the position of each object in the scene, and the candidate objects within the frustum are quickly determined through spatial indexing.

[0108] Step S33, asynchronously preload in the background thread the LOD level data required for prediction but not currently activated, including geometric meshes and texture resources, which can be used immediately when actually needed, significantly reducing the loading latency and stuttering during LOD switching.

[0109] Establish a corresponding texture resolution level for each LOD level k , maintaining the matching of geometric complexity and texture accuracy:

[0110] , where is the base texture resolution used for the highest level of detail (such as 4096×4096), , is the texture scaling factor; it controls the attenuation rate of the texture resolution with the LOD level.

[0111] During the LOD switching process, perform progressive blending of textures synchronously to avoid visual discontinuities caused by texture changes: ; where is the texture data at the level, is the normalized texture coordinate.

[0112] Reduce texture aliasing during distant viewing according to the anisotropic filtering algorithm:

[0113] ; where ( , ) is the sampling step vector calculated along the main gradient direction of the texture, and N is the number of anisotropic samples, which is dynamically adjusted according to the viewing angle and distance to ensure clear texture effects at various viewing angles.

[0114] Take a city planning visualization project as an example to show the technical effects of the present invention in practical applications. This project requires real-time rendering of a city core area of about 5 square kilometers on an ordinary office computer, including about 1,200 buildings of different types.

[0115] The initial data of the project comes from the city BIM system, including detailed 3D building models, with a total of about 23 million polygons. The system applies multi-level detail (LOD) processing to each building: LOD5 (highest detail): retain the original details, about 20,000 polygons per building on average; LOD4: simplify window frames and small decorations, about 8,000 polygons / building; LOD3: merge similar surfaces and retain the main contours, about 3,000 polygons / building; LOD2: simplify to basic volumes, about 1,000 polygons / building; LOD1: minimalist representation, about 300 polygons / building; LOD0: boxed contours, about 50 polygons / building.

[0116] The system sets the base LOD thresholds: D1 = 100 meters (close range), Dn = 1000 meters (distant range). Special buildings are given differential treatment: Town Hall (landmark building): semantic importance weight ωs = 1.5, geometric complexity weight ωg = 1.3; Commercial Center (user focus area): user attention weight ωu gradually increases from the initial 1.0 to 1.3; Standard Residential: the base weights are all 1.0; Suburban Industrial Buildings: the importance weight is reduced to 0.8.

[0117] When the observer views from the downtown streets (ground perspective), the system intelligently applies perspective factors: for high-rise buildings viewed frontally, a higher LOD is used; while for buildings at the same distance viewed at a 45° angle, the LOD is automatically reduced by one level, saving approximately 30% of rendering resources.

[0118] Test scenario: The observer approaches the town hall from 300 meters away, triggering the switch from LOD3 to LOD4.

[0119] In the traditional method, the LOD switches instantaneously, resulting in an obvious jumping effect, with a geometric change of approximately 40 pixels. In the method of the present invention, a smooth transition of 300 milliseconds is applied, setting the interpolation parameters v0 = 0.8 and v1 = 0.6, presenting a transition curve with slow-in and fast-out, and the visual experience is coherent without jumps.

[0120] Example of performance adaptation: Using an office computer (i5-11400 processor, RTX 3060 graphics card, 16GB of memory), the rendering frame rate of the starting scene (suburban perspective) is stable at 60fps. When entering the building-dense area, the frame rate drops to 46fps. The system detects that the performance ratio P(t) = 1.30, and the thresholds D1 and Dn are temporarily reduced by 12%. Distant buildings are automatically switched to a lower LOD, and the frame rate recovers to 58fps within 2 seconds, and the user has no obvious sense of lag.

[0121] When the user moves along the main road at a stable speed (equivalent to 30km / h): The system predicts the camera position 0.3 seconds later, calculates in advance the 6 buildings that will appear in the field of view, and asynchronously loads the high-detail LOD resources of these buildings in the background. When the buildings actually enter the field of view, the high-detail models are already ready, reducing the loading delay by 95%.

[0122] Test data shows that the method for rendering urban 3D models of the present invention not only provides excellent visual quality but also realizes more efficient utilization of hardware resources, making real-time rendering of large-scale urban scenes possible on ordinary hardware.

[0123] Embodiment 2

[0124] As Figure 2 shown, the present invention provides a schematic diagram of the composition of an urban 3D model rendering system, which includes: a model management module, a distance measurement calculation module, an LOD decision module, a smooth interpolation module, and a rendering output module.

[0125] The 3D model rendering system of this city adopts a modular design architecture. Each functional module is connected through an efficient data flow pipeline to ensure the real-time and consistency of data transmission. The system processing flow follows the basic link of "model management → distance calculation → LOD decision → smooth interpolation → rendering output". Each module can be independently upgraded and optimized, with good scalability. In terms of hardware implementation, the system supports heterogeneous computing platforms: the CPU is responsible for global management and high-level decision-making, the GPU is responsible for parallel data processing and real-time rendering, and a dedicated hardware accelerator (such as a ray tracing unit) can be integrated when necessary to enhance specific functions. The system adopts a pipeline parallel architecture and can achieve a stable rendering rate of more than 60 frames per second under standard configurations, and supports up to 120 frames per second or higher on high-end devices to meet the high refresh rate display requirements of VR / AR, etc.

[0126] The model management module is used to construct and maintain multi-level detail models of each building object in the urban scene. The model management module includes a model preprocessing unit, an LOD level generation unit, and a model indexing unit. Among them, the model preprocessing unit is responsible for geometric optimization and topological simplification of the original 3D model. The LOD level generation unit automatically generates at least 3 model versions with different levels of detail based on the principle of decreasing geometric complexity. The model indexing unit establishes the mapping relationship between the model identifier and its respective LOD levels.

[0127] The model management module is the basic data provider of the system, responsible for efficiently managing the multi-level geometry and texture data of thousands of buildings in a large-scale urban scene. The model preprocessing unit uses professional mesh optimization algorithms to process the building models exported from the original CAD or BIM, including mesh repair (such as repairing non-manifold edges, incorrect normal directions, overlapping vertices, etc.), UV coordinate optimization, and geometric topology merging. For example, for a standard office building, the preprocessing can optimize the originally imported 500,000-polygon model to about 50,000 polygons while keeping the key geometric features unchanged, and the texture mapping efficiency is increased by about 30%.

[0128] The LOD level generation unit is the core of this module and is responsible for automatically creating multi-level detailed models for each building. The system adopts an improved Progressive Mesh Simplification algorithm, which prioritizes maintaining visual features and intelligently selects simplification operations based on the Edge Collapse Cost Function. For models with rich details such as cultural relic buildings, the system uses a feature protection strategy to retain feature elements such as corner lines, contours, and decorative patterns during the simplification process. For example, for a historical building with complex carving details, the system automatically identifies and marks its carving area as "high protection priority". When generating LOD, even at medium and low levels (such as LOD2 and LOD3), the key contours of these areas can be retained, while a larger proportion of simplification is performed on ordinary walls and structural areas. The system usually generates 5 - 7 LOD levels for each building, and the polygon count between adjacent levels generally follows a reduction ratio of 1:2 to 1:3.

[0129] The model indexing unit uses a hierarchical hash table structure to efficiently manage all model assets. Each building is indexed by a unique identifier (UUID). The first-level index table maps the UUID to the set of all LOD models of the building, and the second-level index table organizes the specific geometric and material data according to the LOD level. The system supports multi-threaded concurrent access, and the response time for reading any LOD level is controlled within 1 millisecond. For ultra-large-scale scenarios (such as a complete city model containing 10,000+ buildings), the system implements a block-based Streaming Loading mechanism, which dynamically loads and unloads model data in different regions according to the viewing position, ensuring that the memory occupancy is stably within a controllable range (usually 2 - 4GB), while maintaining a smooth viewing experience.

[0130] The distance metric calculation module is used to calculate the effective distance metric from the viewing point to the target object in real time.

[0131] The importance weighting unit is responsible for calculating and applying the object importance weight. This unit comprehensively evaluates the semantic type, geometric complexity, and user attention of the building to generate a normalized importance weight. The system has a preset building type weight table: landmark building = 1.5, commercial center = 1.2, cultural facility = 1.3, office building = 1.0, ordinary residence = 0.8, auxiliary building = 0.6. The geometric complexity is calculated by analyzing the surface change rate and feature edges of the original model, and the complexity ranges from 0.7 to 1.4; the user attention is obtained by statistically analyzing historical interaction data, such as the frequency and duration of user clicks, zooms, or stays near the building.

[0132] The LOD decision module is used to determine the target level of detail for each object based on an effective distance metric. This module consists of three key functional units: a continuous mapping unit, a performance adaptation unit, and a decision optimization unit. To handle urban environments of different scales, the system implements an environment adaptation strategy: in dense urban areas, the threshold distance is automatically shortened; in open areas, the threshold distance is enlarged.

[0133] The performance adaptation unit is responsible for dynamically adjusting the LOD threshold parameters to ensure that the system can maintain the target frame rate on different hardware platforms. This unit uses a PID controller (Proportional-Integral-Derivative controller) model to monitor frame rate fluctuations and calculate the adjustment coefficient applied to the LOD threshold. When the frame rate is lower than the target value (insufficient performance), the controller generates a negative feedback signal to lower all LOD thresholds, enabling more buildings to use lower-detail models; conversely, when there is excess performance, the thresholds are increased to provide a better visual effect. The controller parameters are optimized to ensure that the system's response to performance fluctuations is both sensitive and stable - it can complete 90% of the adjustment within 0.5 seconds while avoiding visual jitter caused by over-adjustment. This unit also implements a scene complexity prediction mechanism that pre-adjusts the LOD parameters before the camera enters a high-complexity area (such as a building-dense area) to prevent a steep drop in the frame rate.

[0134] The decision optimization unit globally optimizes the preliminary LOD decision results to ensure efficient allocation of system resources. This unit takes into account factors such as the distribution of visual importance, GPU memory capacity limitations, and the number of objects on the same screen to implement "Budgeted LOD Optimization". For example, when there are too many buildings on the same screen, the system will prioritize ensuring the level of detail of the central area of sight and important buildings while appropriately reducing the detail of peripheral buildings; when the memory is approaching its upper limit, it will automatically reduce the texture resolution of distant buildings to free up resources. This unit also implements Temporal Coherence Optimization to avoid frequent LOD fluctuations by introducing a change threshold and a hysteresis mechanism - the switch is only actually executed when the calculated LOD value changes by more than ±0.5 and persists for at least 5 frames.

[0135] The smooth interpolation module is used to achieve a time-based smooth transition between adjacent LOD levels.

[0136] The smooth interpolation module is a key component for the system to eliminate the visual jump problem in traditional LOD techniques. This module consists of three core functional units: a time control unit, an interpolation function unit, and a resource coordination unit.

[0137] The time control unit is responsible for managing the time parameters of LOD transitions, determining the start time and duration of the transitions. When the LOD decision module determines that a building needs to change its LOD level, this unit will evaluate the visual impact of the change and set the corresponding transition duration.

[0138] For example, for large LOD changes (such as directly from LOD5 to LOD2), the system will set a longer transition time (400 - 500 milliseconds) to reduce visual abruptness; for small changes (such as from LOD3 to LOD4), a shorter transition time (200 - 300 milliseconds) is used to maintain timely response. This unit also implements a line-of-sight based priority strategy - buildings in the central area of the line of sight receive higher priority for processing, ensuring that the transition is completed first in the user's gaze area, while controlling the number of buildings undergoing parallel transitions, usually limited within the range of 50 - 100, to avoid overloading GPU resources.

[0139] The interpolation function unit implements high-quality smooth transition curves. The system optimizes control parameters for transitions between different LODs: when transitioning from high to low detail, a "slow start, fast end" curve ( is adopted to make details gradually disappear; when transitioning from low to high, a "fast start, slow end" curve ( is used to make details emerge naturally. This asymmetric design conforms to the human eye's perception characteristics of detail changes and greatly improves visual comfort. This unit also implements adaptive curve parameters for special viewing conditions (such as a fast-moving camera), using a steeper transition curve in rapidly changing scenes to maintain timely response.

[0140] The resource coordination unit is responsible for coordinating geometry, material, and texture resources during the transition to ensure synchronous and consistent transition effects. This unit implements a "resource readiness check" mechanism to confirm that all necessary LOD resources (geometry data, textures, etc.) are ready before starting the transition, avoiding visual artifacts caused by resource loading delays. To handle a large number of parallel transitions, the unit adopts a resource pool design to pre-allocate and manage the GPU memory and computing resources required for the transition, preventing memory fragmentation and resource contention. During the transition, the system synchronizes the blending of geometric shapes and texture details to avoid common problems such as "geometry has changed but texture lags behind". For transitions with a large LOD span (such as from LOD5 to LOD1), the system implements a "multi-step transition" strategy, decomposing the large-span change into multiple intermediate transitions, each step using independent time control and interpolation functions to further enhance visual coherence.

[0141] The rendering output module is used to convert the processed LOD data into the final visual output. This module includes a geometry data blending unit, a shader scheduling unit, and a frame buffer management unit. Among them, the geometry data blending unit synthesizes the final vertex and index data according to the interpolation weights, the shader scheduling unit coordinates the allocation of GPU computing resources, and the frame buffer management unit controls the output and display of the rendering results.

[0142] The rendering output module is the ultimate executor of the system, responsible for converting the processed LOD data into high-quality visual output. The geometry data blending unit is the core processing engine of this module. It receives the blending weight parameters from the smooth interpolation module and performs efficient vertex data blending calculations on the GPU. This unit uses a specially optimized Geometry Shader to achieve efficient vertex attribute interpolation, including position, normal, texture coordinates, and tangent data. For buildings in transition, the system adopts a multi-channel rendering strategy optimized for cascaded blending: first, perform pre-computations at the vertex level, and then use vertex index remapping technology to efficiently execute the blending rendering. This unit also implements a lightweight rendering path for mobile devices. By simplifying the blending algorithm and optimizing the memory layout, while maintaining the visual effect, it reduces the computational overhead by approximately 40%, enabling this technology to be applicable to various hardware platforms.

[0143] The shader scheduling unit is responsible for efficiently managing the allocation and scheduling of GPU computing resources. This unit implements the "Layered Rendering Queue" mechanism, which groups the buildings in the urban scene according to material types, LOD states (stable or in transition), and rendering priorities, minimizing the shader switching and state change overhead. For large-scale scenes, the system adopts a "Deferred Shading" rendering pipeline, which uniformly processes the geometry data in the G-Buffer stage, and then efficiently applies complex lighting and shadow algorithms in the lighting stage. For mobile platforms, the system automatically switches to the "Deferred Lighting" or "Forward+" rendering path to optimize the use of memory bandwidth. This unit also implements a dynamic LOD shader variant management system, which dynamically selects the most appropriate shader precision level (such as high-precision normal maps, simplified environment reflections, etc.) according to the current scene complexity and available performance.

[0144] The frame buffer management unit controls the processing and output of the final rendering results. The unit implements an advanced post-processing chain, including anti-aliasing (using SMAA technology), ambient occlusion (using GTAO algorithm), depth of field and motion blur, etc., to provide movie-level visual quality for urban scenes. The system supports high dynamic range (HDR) rendering process, uses 16-bit floating point precision to store and process lighting data, and then converts it into a color range that can be presented by the display device through an advanced tone mapping algorithm (ToneMapping). For different display devices, the unit provides adaptive output configuration: for high-resolution displays (such as 4K displays), the system can use dynamic resolution rendering technology to dynamically adjust the actual rendering resolution according to the complexity of the scene and available performance, and then use high-quality upsampling algorithms (such as DLSS or FSR) to generate the final output image to achieve the best balance between performance and quality; the unit also supports a programmable rendering special effect plug-in system, allowing developers to add custom post-processing effects to meet the visual needs of specific application scenarios.

[0145] The system also includes a user interaction analysis module, which includes an interaction behavior recording unit, an attention statistics unit and a personalized weight adjustment unit.

[0146] The user interaction analysis module is an intelligent and personalized enhancement component of the system. By analyzing user behavior, the module automatically adjusts the rendering strategy to provide a visual experience that is more in line with user habits and preferences; the interactive behavior recording unit is responsible for capturing and storing various types of interaction data between users and three-dimensional urban scenes. This unit implements a multi-dimensional interaction tracking system, recording various behavioral data including explicit interactions (such as clicking, selecting, and zooming specific buildings), implicit interactions (viewpoint dwell time, observation angle change mode), and contextual interactions (such as search queries, information requests). The system adopts a lightweight event capture architecture to record mouse movement trajectories and dwell points in a desktop environment, track gesture operation modes on touch devices, and analyze head orientation and gaze points in a VR environment. To protect user privacy, all behavioral data are anonymized locally and are only used to optimize the rendering experience of the current user. By default, they are not uploaded to the server (optional configuration allows cloud personalization, but users must explicitly authorize it).

[0147] The attention statistics unit conducts intelligent analysis on the collected interaction data to extract user attention patterns and the distribution of interest points. This unit adopts a hybrid analysis model, combining techniques such as Heatmap Statistics, Viewpoint Clustering, and Time-series Analysis to identify the types, regions, and features of buildings that users frequently focus on. The system constructs a "User Attention Profile" to record insights such as: the average stay time of users on historical buildings is 2.5 times that of modern buildings; users have magnified the facade details of a specific commercial area multiple times; users tend to use landmark buildings as reference points when navigating. These statistical results are continuously updated using Exponentially Weighted Moving Average (EWMA), with more recent behaviors carrying higher weights to ensure that the attention model can adjust according to changes in user interests. In a multi-user environment, the system supports storing and applying different attention models differentiated by user ID, allowing multiple users to obtain personalized visual experiences on the same system.

[0148] The personalized weight adjustment unit converts user attention data into specific LOD weight adjustment strategies and automatically modifies the importance weights of different buildings according to user attention statistics. For example, when it is detected that a user frequently focuses on a specific commercial area, the system increases the value of the buildings in this area from the default 1.0 to 1.2 - 1.5, enabling these buildings to present a higher level of detail at the same viewing distance. The weight adjustment follows the principle of smooth gradual change to avoid discomfort caused by sudden visual changes. This unit also implements a context-aware adjustment strategy, such as identifying whether the user is in "exploration mode" (frequently changing perspectives) or "observation mode" (studying a specific area for a long time), and adjusting the LOD strategy accordingly - giving priority to ensuring fluency in exploration mode and providing detail clarity in observation mode. The system supports manual intervention and reset functions, allowing users to explicitly adjust or reset personalized parameters to ensure that users always maintain control over the rendering experience.

[0149] The specific implementation manners described above have further elaborated on the purpose, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific implementation manners of the present invention and are not used to limit the protection scope of the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for rendering a three-dimensional model of a city, characterized in that, The method includes the following steps: Step S1, establish a multi-level detail model system, create three-dimensional models with multiple different levels of detail for each building object in the urban scene, and form a continuous hierarchical structure from high-precision to low-precision; The at least three different levels of detail are at least three; Step S2, comprehensively consider spatial distance, viewing angle factors, and object importance, calculate the effective distance metric from the viewpoint to the target object, establish a continuous LOD switching function according to the effective distance metric, and map the distance to a continuous level of detail value; , where is the effective distance of the target object, is the Euclidean distance from the observation point to the center of the object, is the angle between the line-of-sight direction and the main axis of the object, is the importance weight determined based on the semantic and geometric features of the object; ; wherein, is the LOD switching function, and are the highest and lowest level-of-detail indices respectively, and are the preset distance threshold boundaries; Step S3, apply a time-based smooth interpolation algorithm to perform progressive switching between adjacent LOD levels to eliminate visual jumps during switching; ; wherein, is the current actual rendered LOD level value, is the level before switching, is the target level, is a smooth interpolation function to ensure the continuity of the first derivative, , is a normalized transition time parameter, and visually smooth switching is achieved by controlling the transition duration.

2. The method according to claim 1, characterized in that, Perform an adaptive correction on the effective distance metric to account for the influence of the actual visible size of the object on the screen on LOD selection: , where is the total screen pixel area of the current rendering target, is the projected pixel area of the object on the screen plane.

3. The method according to claim 2, characterized in that, The importance weight of the object Calculated through the weighted combination of multi-dimensional factors to implement differentiated LOD strategies for different types of buildings: , where is the semantic importance weight, and different values are preset according to the functional type of the building. The landmark building has the highest weight; is the geometric complexity weight, which is calculated based on the number of polygons and the complexity of geometric features of the building model. The higher the complexity, the greater the weight; is the user attention weight, which is statistically obtained based on historical user interaction data. The higher the attention of the building, the higher the weight; The constraint condition is: , to ensure the normalization of the weight, is the constraint condition coefficient, which can be dynamically adjusted according to the application scenario.

4. The method according to claim 3, characterized in that Smoothing interpolation function The specific calculation is as follows: , where and are the tangent vector parameters of the starting point and the ending point respectively, , , are the basis functions of the smoothing interpolation function : ， ， 。 5. The method according to claim 4, wherein Real-time rendering is implemented based on the GPU parallel computing architecture. The processing flow of LOD is specifically as follows: Step S21 divides the entire city model into regular grid cells, and each cell independently performs LOD calculation and rendering processing, where the value is dynamically determined according to the scene complexity and GPU performance, and the preset value is 50 meters; Step S22, calculate the LOD decision matrix for each grid cell in parallel , where the decision matrix stores the optimal LOD level of each object within the cell ; Among them, represents grid coordinates, is the candidate integer LOD level; Step S23, implement batch geometric subdivision and vertex data blending in the GPU shader, and perform LOD interpolation calculation and output for multiple vertices through parallel processing ; , where is the Boolean index mask matrix of the th level, is the vertex attribute array corresponding to the LOD level, is the blending weight calculated based on the time parameter , and all weights satisfy the normalization condition .

6. The method according to claim 5, characterized in that, Continuously monitor the rendering performance metrics and define the performance ratio : , where is the preset target frame rate, which is 60fps or 30fps, is the actual rendering frame rate at the current moment; Dynamically adjust the distance thresholds of all LOD levels based on the performance ratio to achieve adaptive control of the rendering load: ； Wherein, and are respectively the old value and the new value of the th threshold, , is the adjustment intensity parameter to ensure a smooth change in the threshold and avoid sudden LOD level jumps.

7. The method according to claim 6, characterized in that, Prepare the required level of detail in advance by predicting the user's viewing path, and reduce the computational latency during switching based on the future time LOD requirements predicted by the camera movement.

8. The method according to claim 7, characterized in that Establish corresponding texture resolution levels for each LOD level k , maintaining the matching of geometric complexity and texture accuracy: , where is the base texture resolution used for the highest level of detail, , is the texture scaling factor; Synchronously perform progressive blending of textures during LOD switching to avoid visual discontinuities caused by texture changes: ; among which, is the texture data of the th level, is the normalized texture coordinate.

9. A three-dimensional city model rendering system for performing the method according to any one of claims 1-8, characterized in that, The system includes: a model management module, a distance metric calculation module, an LOD decision module, a smooth interpolation module, and a rendering output module; The model management module is used to construct and maintain the multi-level detail models of each building object in the urban scene. The model management module includes a model preprocessing unit, an LOD level generation unit, and a model indexing unit. Among them, the model preprocessing unit is responsible for geometric optimization and topology simplification of the original three-dimensional model. The LOD level generation unit automatically generates at least three model versions with different levels of detail based on the principle of decreasing geometric complexity. The model indexing unit establishes the mapping relationship between the model identifier and its respective LOD levels; The distance metric calculation module is used to calculate the effective distance metric from the observation point to the target object in real time; The LOD decision module is used to determine the target level of detail of each object based on the effective distance metric; The smooth interpolation module is used to achieve a time-based smooth transition between adjacent LOD levels; The rendering output module is used to convert the processed LOD data into the final visual output. The module includes a geometric data mixing unit, a shader scheduling unit, and a frame buffer management unit. Among them, the geometric data mixing unit synthesizes the final vertex and index data according to the interpolation weight. The shader scheduling unit coordinates the allocation of GPU computing resources. The frame buffer management unit controls the output and display of the rendering result.

10. The system according to claim 9, wherein The system further includes a user interaction analysis module, and the user interaction analysis module includes an interaction behavior recording unit, an attention degree statistics unit, and a personalized weight adjustment unit; The interaction behavior recording unit tracks the user's operation behaviors such as clicks, zooms, and stays; The attention degree statistics unit analyzes the user's attention patterns to different regions and buildings; the personalized weight adjustment unit dynamically adjusts the object importance weights according to the user's preferences to implement a personalized LOD optimization strategy.

Images