Urban traffic road condition data simulation visual rendering method and system

Abstract

The invention relates to the field of real-time visualization of road conditions, in particular to a data simulation visualization rendering method and system for urban traffic road conditions. The method comprises the following steps: extracting a real-time satellite streetscape image based on urban satellite remote sensing scanning, and performing scene pixel-level segmentation to obtain scene texture rendering parameters; scene illumination visual identification is carried out according to the real-time satellite streetscape image, traffic scene background modeling is carried out based on scene texture rendering parameters, and a real-time scene background model is constructed; the method comprises the following steps: acquiring urban-level multi-source traffic monitoring data flow, performing vehicle state sensing, and constructing a multi-dimensional particle feature matrix; road network topological correlation analysis and global traffic network state perception are carried out according to the real-time satellite streetscape images, and a road network state perception model is constructed. According to the invention, a real real-time traffic environment is visualized, scene effects in different traffic states are presented, the current road condition can be rapidly evaluated, and the traffic control decision efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of real-time road condition visualization, and in particular to a data simulation visualization rendering method and system for urban traffic conditions. Background Technology

[0002] Urban traffic management primarily relies on hardware devices such as road sensors, traffic cameras, and traffic light systems for traffic flow monitoring, vehicle dispatching, and road condition prediction. While these traditional methods provide some data support, they often only offer limited monitoring within specific areas and time periods, and struggle to respond in real-time to emergencies or dynamically changing traffic conditions. Furthermore, these traditional methods lack sufficient intelligent support, failing to delve into the potential patterns within massive amounts of traffic data, resulting in insufficient accuracy in traffic prediction and limited decision support capabilities.

[0003] In recent years, with the rapid development of information technology, the Internet of Things, artificial intelligence, big data, and other technologies, data-driven urban traffic simulation and visualization methods have become a research and application hotspot. These technologies have significantly improved the ability to collect, process, and analyze urban traffic data, providing traffic managers with more real-time and accurate decision support. By integrating multi-source data (such as road surveillance videos, traffic sensor data, and social media information) and combining them with advanced traffic flow modeling and simulation analysis algorithms, it is possible to simulate and visualize traffic conditions in different time periods and areas of the city. This not only improves the efficiency of traffic management but also provides citizens with more accurate travel guidance. Although several traffic simulation models and visualization tools have emerged, most methods still have certain limitations. For example, existing simulation models often ignore the randomness and complexity of traffic flow, making it impossible to accurately reflect the impact of sudden events on traffic conditions. In addition, existing visualization rendering methods often lack efficient rendering algorithms when presenting complex urban traffic scenarios, resulting in unrealistic rendering effects that cannot effectively support traffic decision-making. At the same time, traditional traffic simulation systems mostly rely on manual input and static data, lacking real-time performance and automation, and cannot provide timely feedback and adjustments in rapidly changing traffic environments. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention proposes a data simulation visualization rendering method and system for urban traffic conditions, thereby resolving at least one of the aforementioned technical issues.

[0005] To achieve the above objectives, the present invention provides a data simulation visualization rendering method for urban traffic conditions, comprising the following steps: Step S1: Extract real-time satellite street view images based on urban satellite remote sensing scans, perform scene pixel-level segmentation, and obtain scene texture rendering parameters; Step S2: Perform scene lighting visual recognition based on real-time satellite street view images, and model the traffic scene background based on scene texture rendering parameters to construct a real-time scene background model; Step S3: Acquire city-level multi-source traffic monitoring data streams, perform vehicle status perception, and construct a multi-dimensional particle feature matrix; Step S4: Perform road network topology correlation analysis and global traffic network status perception based on real-time satellite street view images, and construct a road network status perception model; Step S5: Perform real-time vehicle flow particle mapping on the road network state perception model based on the multi-dimensional particle feature matrix, and perform real-time background update and overlay based on the real-time scene background model to execute traffic condition visualization rendering.

[0006] This specification provides a data simulation and visualization rendering system for urban traffic conditions, used to execute the data simulation and visualization rendering method for urban traffic conditions as described above, including: The pixel segmentation module is used to extract real-time satellite street view images based on urban satellite remote sensing scans, perform scene pixel-level segmentation, and obtain scene texture rendering parameters. The background modeling module is used to perform scene lighting visual recognition based on real-time satellite street view images and to perform traffic scene background modeling based on scene texture rendering parameters, thereby constructing a real-time scene background model. The particle abstraction module is used to acquire city-level multi-source traffic monitoring data streams, perform vehicle status perception, and construct a multi-dimensional particle feature matrix. The road network status perception module is used to perform road network topology correlation analysis and global traffic network status perception based on real-time satellite street view images, and to build a road network status perception model. The visualization rendering module is used to perform real-time vehicle flow particle mapping on the road network state perception model based on the multi-dimensional particle feature matrix, and to perform real-time background update and overlay based on the real-time scene background model in order to perform traffic condition visualization rendering operations.

[0007] The beneficial effects of this invention are specifically as follows: By performing pixel-level segmentation on satellite images (e.g., using deep learning semantic segmentation technology), each pixel in the image is accurately classified into a specific traffic element category. This refined segmentation can clearly distinguish objects such as roads, vehicles, buildings, pedestrians, and trees, providing high-quality basic data for subsequent analysis. The segmented scene not only helps in identifying and distinguishing elements, but also accurately extracts the texture, material, and spatial location of each element. These rendering parameters can provide detailed input data for subsequent 3D scene modeling, ensuring the realism and visual effect of the traffic scene simulation rendering. By performing illumination recognition on real-time satellite street view images, the light sources and shadows in the images can be intelligently adjusted to restore the realistic ambient lighting effect. This is adaptable to various lighting conditions in dynamic traffic scenes (such as day and night, and different weather conditions), thereby ensuring that traffic images generated at any time have consistent lighting effects. By using scene texture rendering parameters (such as material type, surface reflectivity, etc.) for background modeling of traffic scenes, a real-time and dynamically updated background model can be constructed. This means that the background of traffic scenarios can be automatically updated based on real-time changes (such as lighting, weather, and environmental factors), ensuring the realism and real-time performance of the scenarios. Real-time vehicle status (such as speed, location, and type) is acquired through multi-source traffic monitoring data streams (such as traffic cameras, sensors, and GPS). The fusion of this real-time data stream helps to accurately monitor the status of each vehicle, providing effective input data for traffic flow analysis, thereby improving the accuracy of traffic management and road condition warnings. The status information of each vehicle is extracted and mapped into a particle feature matrix, which can characterize the vehicle in detail from different dimensions (such as time, space, and speed). This multi-dimensional feature matrix can not only describe the instantaneous state of the vehicle but also predict its future dynamic behavior, providing important decision-making basis for traffic simulation and flow control. Based on real-time satellite imagery and traffic data streams, road network topology analysis can accurately describe the structure and traffic flow of urban roads. This analysis can identify key traffic nodes, road bottlenecks, and their connections, thus providing a scientific basis for optimizing traffic management and adjusting traffic light timings.

[0008] By achieving state awareness of the entire urban transportation network, real-time data on traffic flow, vehicle speed, and road congestion in various areas can be obtained. This global awareness capability helps adjust traffic control strategies such as traffic lights and road segment allocation in real time, improving traffic efficiency and reducing congestion. Utilizing a multi-dimensional particle feature matrix to perform real-time vehicle flow particle mapping on the road network state awareness model allows for precise description of the state and behavior of each vehicle on different road segments. Particle mapping technology enables accurate simulation of complex traffic conditions (such as traffic density and traffic accidents), enhancing the expressiveness of traffic simulation models. Based on the dynamic updating of the real-time scene background model, traffic condition visualization not only displays the realistic traffic environment but also allows for adjustments to the traffic scene based on real-time data. By overlaying vehicle flow particles with the background model, the scene effects under different traffic conditions (such as dense traffic and traffic accidents) can be accurately presented, providing immediate decision support for urban traffic managers. The final visualization rendering results clearly show the road conditions under different traffic states, such as road congestion levels, vehicle speed changes, and traffic bottlenecks. Real-time visualization helps traffic managers quickly assess current road conditions, predict future trends, and make decisions or take measures at critical moments, greatly improving traffic flow and safety. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating the steps of a data simulation and visualization rendering method for urban traffic conditions according to the present invention. Figure 2 This is a detailed flowchart illustrating the implementation steps of step S1. Figure 3 This is a detailed flowchart illustrating the implementation steps of step S2; Figure 4 This is a flowchart illustrating the detailed implementation steps of step S3. Detailed Implementation

[0010] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0011] This application provides a data simulation visualization rendering method and system for urban traffic conditions. The execution entities of the urban traffic condition data simulation visualization rendering method and system include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, network upload devices, etc., which can be considered as general computing nodes in this application. The data processing platform includes, but is not limited to, at least one of an audio-visual management system, an information management system, and a cloud-based data management system.

[0012] Please see Figures 1 to 4 This invention provides a data simulation and visualization rendering method for urban traffic conditions, comprising the following steps: Step S1: Extract real-time satellite street view images based on urban satellite remote sensing scans, perform scene pixel-level segmentation, and obtain scene texture rendering parameters; Step S2: Perform scene lighting visual recognition based on real-time satellite street view images, and model the traffic scene background based on scene texture rendering parameters to construct a real-time scene background model; Step S3: Acquire city-level multi-source traffic monitoring data streams, perform vehicle status perception, and construct a multi-dimensional particle feature matrix; Step S4: Perform road network topology correlation analysis and global traffic network status perception based on real-time satellite street view images, and construct a road network status perception model; Step S5: Perform real-time vehicle flow particle mapping on the road network state perception model based on the multi-dimensional particle feature matrix, and perform real-time background update and overlay based on the real-time scene background model to execute traffic condition visualization rendering.

[0013] In the embodiments of the present invention, see Figure 1 This is a flowchart illustrating the steps of a data simulation and visualization rendering method for urban traffic conditions according to the present invention. In this example, the steps of the data simulation and visualization rendering method for urban traffic conditions include: Step S1: Extract real-time satellite street view images based on urban satellite remote sensing scans, perform scene pixel-level segmentation, and obtain scene texture rendering parameters; In this embodiment, a panoramic scan of the target urban area is performed using a high-resolution satellite remote sensing system. WorldView-4 satellite imagery data with a resolution of 0.3 meters is used, covering the visible light 450-800nm ​​and near-infrared 760-900nm frequency bands. The system updates the imagery data every 15 minutes to ensure real-time performance. Immediately after image acquisition, atmospheric and geometric corrections are performed to eliminate atmospheric scattering and the effects of Earth's curvature, controlling the image positioning accuracy to within 2 meters.

[0014] Next, a deep semantic segmentation network based on the DeepLab v3+ architecture is used to perform pixel-level accurate segmentation of street view images. This network is pre-trained on the Cityscapes dataset, which includes 19 semantic categories such as roads, buildings, vehicles, greenery, and sky. The encoder part of the segmentation network uses ResNet-101 as the backbone, and the decoder part uses a dilated spatial pyramid pooling module to effectively expand the receptive field. The input image size is set to 1024×2048 pixels, the batch size is 8, the initial learning rate is 0.007, and a multinomial decay strategy is used. After semantic segmentation, the system generates a category probability map for each pixel, with a confidence threshold set to 0.85.

[0015] Subsequently, texture feature extraction and material analysis were performed on the segmentation results. The system used a Gabor filter bank to extract texture features at different directions and scales. The filter parameters included 8 directions (0° to 157.5°, intervals of 22.5°) and 5 scales (σ values ​​of 1, 2, 4, 8, and 16, respectively). For road surfaces, the reflectivity of materials such as asphalt, concrete, and stone was analyzed in detail, and a material reflectivity database was established. The reflectivity of asphalt pavement was set to 0.15-0.25, and that of concrete pavement to 0.35-0.45. The texture analysis of building facades used the Local Binary Pattern (LBP) algorithm, with a statistical radius of 3 pixels and 24 sampling points. Finally, a scene texture rendering parameter set containing attributes such as material type, roughness parameters, reflectivity, and transparency was generated, providing accurate material basis data for subsequent 3D scene reconstruction.

[0016] Step S2: Perform scene lighting visual recognition based on real-time satellite street view images, and model the traffic scene background based on scene texture rendering parameters to construct a real-time scene background model; In this embodiment, illumination condition analysis and solar position calculation are performed on satellite images from the scene texture rendering parameter set. The shooting time and geographic coordinates are obtained by analyzing the EXIF ​​information of the images, and the solar position is accurately calculated by combining solar altitude angle and azimuth angle algorithms. The formula for calculating the solar altitude angle α considers latitude φ, solar declination δ, and hour angle ω. A solar altitude angle greater than 30° is considered sufficient illumination, while less than 15° is considered weak light. The system also assesses weather conditions and cloud cover by analyzing the histogram distribution and contrast of the images. Under clear weather conditions, the image contrast is typically between 0.6 and 0.9, while under cloudy weather conditions it drops to 0.3-0.5.

[0017] After the lighting analysis is completed, the system uses a physically based rendering (BRDF) model to calculate the lighting response of each material in the scene. For different material types, the system selects the appropriate BRDF model: the Lambertian model for rough surfaces such as asphalt pavement, the Cook-Torrance model for metal surfaces such as vehicles and road signs, and the Oren-Nayar model for rough, diffuse surfaces such as building walls. The BRDF parameters for each material are determined experimentally; the roughness parameter σ for asphalt pavement is set to 0.3-0.5, and the Fresnel reflectance coefficient F0 for metal surfaces is set to 0.04-0.1.

[0018] Based on texture rendering parameters and lighting analysis results, the system constructs a 3D scene geometric model. Terrain undulations are reconstructed using heightmap and normal mapping techniques. The heightmap resolution is set to correspond to 0.5 meters of actual distance per pixel, achieving a height accuracy of 0.1 meters. Buildings are modeled using a procedural approach, automatically generating simplified geometry based on building outlines from satellite imagery. Building heights are estimated using shadow analysis and stereo vision algorithms, with an average error controlled within 2 meters. The road network uses a centerline extraction algorithm to generate road geometry. Road widths are automatically set based on the number of lanes, with a standard single-lane width of 3.5 meters, including a 0.5-meter shoulder width. The final real-time scene background model contains complete geometric information, material properties, and lighting conditions. The total number of model vertices is controlled within 5 million to ensure real-time rendering performance.

[0019] Step S3: Acquire city-level multi-source traffic monitoring data streams, perform vehicle status perception, and construct a multi-dimensional particle feature matrix; In this embodiment, a multi-source data acquisition network is established, integrating data sources such as traffic monitoring cameras, GPS positioning systems, mobile signaling base stations, roadside radar sensors, and electronic police systems. Camera data uses H.264 compression format, with a resolution of 1920×1080 and a frame rate of 25fps, generating approximately 8GB of data per hour per camera. GPS trajectory data is acquired at a frequency of once per second, with a positioning accuracy requirement within 5 meters. The data format includes fields such as timestamp, latitude and longitude, speed, and direction angle. Mobile signaling data is obtained through cooperation with operators, with a sampling interval of 5 minutes, covering over 95% of the urban area, and obtaining approximately 1 million location records per sampling. Roadside sensors include millimeter-wave radar and lidar, with a detection range of 150 meters, an angular coverage of 120°, and a data update frequency of 10Hz.

[0020] In the data preprocessing stage, a Kalman filter algorithm is used to smooth the GPS trajectory, with filtering parameters including process noise variance Q=0.1 and observation noise variance R=5.0. GPS points are projected onto the road network using a map matching algorithm, with a matching radius set to 30 meters. A Hidden Markov Model is then used to calculate the optimal path. Mobile signaling data undergoes spatial interpolation using a Voronoi diagram algorithm, dividing the base station coverage area into non-overlapping polygonal regions. The traffic density of each region is calculated using signaling strength weighting. Camera video is processed using the YOLO v5 object detection algorithm to identify vehicles, achieving an accuracy rate of over 92%, while also extracting vehicle bounding boxes, categories, and confidence scores.

[0021] Traffic flow abstraction processes each detected vehicle as a dynamic particle with multi-dimensional attributes. The particle's position attribute uses the UTM coordinate system with meter-level accuracy; velocity attribute is calculated through continuous inter-frame position difference, including velocity magnitude and direction angle; acceleration attribute is calculated using second-order difference with a smoothing window of 3 seconds; vehicle type attributes include eight categories such as cars, trucks, and buses, each assigned different physical parameters such as length, width, and maximum speed. Particle behavior characteristics include following distance, lane change frequency, and stopping time. Following distance is obtained through statistical analysis of the distance between vehicles in front and behind, with a normal following distance set at 15-30 meters. The system generates approximately 500,000 dynamic particles per second. A clustering algorithm merges particles with similar attributes, ultimately constructing a multi-dimensional particle feature matrix of dimension N×12, where N is the number of particles, and the 12 dimensions include spatial coordinates, motion state, vehicle attributes, and behavioral characteristics.

[0022] Step S4: Perform road network topology correlation analysis and global traffic network status perception based on real-time satellite street view images, and construct a road network status perception model; In this embodiment, the topology of the road network in the real-time scene background model is extracted and analyzed. A skeletonization algorithm is used to extract the road centerline, and a single-pixel-width road skeleton is obtained through distance transformation and thinning operations. Road intersections are identified using the Harris corner detection algorithm, with a detection threshold set to 0.01 and a non-maximum suppression window size of 3×3 pixels. The system abstracts the road network as a graph structure, with intersections as nodes and road segments as edges, totaling approximately 5000 nodes and 8000 edges. Each node includes attributes such as geographic coordinates, connectivity, and traffic light type, while each edge includes attributes such as length, width, number of lanes, and speed limit. Road segment length is calculated using geodesic distance, with an average length of 200 meters, and the number of lanes is estimated by dividing the road width by the standard lane width of 3.5 meters.

[0023] Next, a Graph Convolutional Neural Network (GCN) is used for deep modeling of the road network topology. The GCN network consists of three convolutional layers with 64, 32, and 16 hidden units per layer, respectively. The ReLU activation function is used, and the dropout rate is set to 0.2 to prevent overfitting. Network input features include topological metrics such as degree centrality, betweenness centrality, and compact centrality of nodes, as well as dynamic features such as real-time traffic flow, average speed, and occupancy. Degree centrality is calculated by the number of edges connecting a node, betweenness centrality by the number of shortest paths passing through that node, and compact centrality by the average shortest distance from that node to all other nodes. Dynamic features are updated every minute; traffic flow is expressed as vehicles per hour, average speed as kilometers per hour, and occupancy represents the percentage of time vehicles occupy the road.

[0024] Global traffic network state perception is implemented using a graph attention network (GAT) based on an attention mechanism. The GAT network adaptively learns the importance weights between nodes, with 8 attention heads and an output dimension of 8 for each head. The network aggregates neighbor node information through a multi-head attention mechanism, and the attention weights in the calculation formula are obtained through a linear transformation activated by LeakyReLU, with a negative slope of 0.2. The system uses a sliding window mechanism to maintain historical traffic states, with a window size of 30 minutes and a sliding window every 5 minutes. By comparing the current state with historical patterns, the system can identify abnormal traffic events, with the anomaly detection threshold set to the historical mean plus or minus 2 standard deviations. The final constructed road network state perception model can reflect the real-time traffic operation status of the entire network, achieving a prediction accuracy of over 85% and a response latency controlled within 30 seconds.

[0025] Step S5: Perform real-time vehicle flow particle mapping on the road network state perception model based on the multi-dimensional particle feature matrix, and perform real-time background update and overlay based on the real-time scene background model to execute traffic condition visualization rendering.

[0026] In this embodiment, a spatial hashing algorithm is used to efficiently map dynamic particles in the multidimensional particle feature matrix to the corresponding road segments in the road network state perception model. The grid size of the spatial hash table is set to 50×50 meters to ensure fast spatial query performance. Each particle calculates a hash value based on its coordinate position and maps it to the edge of the corresponding road segment. The mapping accuracy is calculated using the shortest distance from a point to a line segment, with a distance threshold set at 15 meters. For multiple particles on the same road segment, the system calculates statistical indicators including particle density, average speed, and speed variance. Particle density is defined as the number of particles per unit length of road segment, with units of vehicles per kilometer. Average speed is calculated by the arithmetic mean of the speeds of all particles, and speed variance reflects the stability of traffic flow; a variance value greater than 20 indicates a congested state.

[0027] After particle mapping is completed, the system updates the real-time traffic parameters of each road segment in the road network state perception model. Traffic flow is calculated by the number of particles passing through the road segment within a statistical time window, which is set to 5 minutes, and the flow unit is converted to vehicles per hour. Occupancy rate is calculated as the proportion of particle dwell time on the road segment to the total time; an occupancy rate greater than 0.3 indicates road segment congestion. Speed ​​parameters directly use the instantaneous velocity of particles, while the 85th percentile velocity is calculated as a reference for free-flow velocity. The congestion index is calculated by the speed ratio, i.e., actual speed divided by free-flow velocity; a ratio less than 0.5 indicates severe congestion, 0.5-0.7 indicates moderate congestion, and greater than 0.7 indicates smooth flow.

[0028] The visualization rendering phase first maps the updated road network status to visual elements. The system uses a heatmap to represent traffic density, with a color gradient from green to red. Green (RGB: 0,255,0) represents smooth traffic, yellow (RGB: 255,255,0) represents slow traffic, and red (RGB: 255,0,0) represents congestion. Particle rendering uses point sprite technology, rendering each particle as a dot with a radius of 2-8 pixels. The size is adjusted according to vehicle type, and the color is set according to speed. Particle trajectories are displayed by connecting historical positions, with the trajectory length set to the 10 most recent positions, the line width 1-3 pixels, and the transparency decreasing over time.

[0029] The final rendering stage composites dynamic traffic elements with the real-time scene background model. The system employs layered rendering technology, with the background model as the bottom static scene, the traffic flow heatmap as the middle semi-transparent overlay, and dynamic particles as the top real-time elements. Rendering parameters include a viewing height of 500-2000 meters, a viewing pitch angle of 45-90 degrees, and a horizontal viewing angle of 60-120 degrees. The system uses a multi-threaded rendering architecture: the background rendering thread handles static scene updates, the particle rendering thread handles dynamic element rendering, and the main rendering thread handles final compositing. The target frame rate is set at 30fps, automatically reducing particle rendering precision when the system load is too high to ensure a smooth user experience. Rendering output supports multiple resolutions, from 1080p to 4K, with image formats supporting PNG and JPEG, and video output supporting H.264 encoding, ultimately forming a complete real-time visualization rendering system for urban traffic conditions.

[0030] In this embodiment, see Figure 2 The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include: A panoramic satellite remote sensing scan of urban traffic was conducted to extract real-time satellite street view images; Global brightness optimization is performed on real-time satellite street view images to construct a brightness-enhanced street view image; Scene pixel-level segmentation is performed on the brightness-enhanced street scene image to obtain the traffic scene semantic mask; Based on the semantic mask of the traffic scene, we perform element-by-element morphological analysis to obtain the spatial location and geometric shape of each element in the scene. Based on the spatial location and geometry, depth element feature mining is performed to extract scene depth information features; The scene depth information features are textured and rendered to obtain scene texture rendering parameters.

[0031] In this embodiment, the urban area to be monitored is determined by defining the scanning area using latitude and longitude coordinates or GIS vector boundaries. Satellite remote sensing scanning can utilize a combination of multispectral and panchromatic remote sensing sensors to ensure that the image has both high resolution (e.g., 0.5–1 meter spatial resolution) and retains multispectral information for subsequent processing. To ensure real-time performance, a low-Earth orbit satellite constellation with a revisit cycle of minutes (e.g., commercial remote sensing constellations such as PlanetScope or Jilin-1) can be used, and data streams can be acquired in real time via satellite data receiving ground stations or APIs. During the scanning process, the system needs to perform geometric correction (orthorectification, radiometric calibration, etc.) on the original satellite image to eliminate the influence of terrain, satellite attitude, and sensor distortion on the image, ensuring accurate geographical alignment. In addition, cloud detection and shadow removal are required. Threshold-based cloud masking algorithms (e.g., Fmask) can be used to mark areas obscured by clouds, and multi-temporal synthesis can be used to fill in missing information as needed. The final output is a real-time satellite street view image of the urban traffic area. The image data is generally saved in GeoTIFF or other georegistration formats for easy subsequent processing. Satellite street view images are often affected by factors such as atmospheric scattering, lighting conditions, and shooting time, resulting in problems such as insufficient brightness, low contrast, or excessively dark shadows. The goal of global brightness optimization is to improve overall visibility and contrast for more accurate subsequent pixel segmentation. This process can employ histogram equalization (HEM), adaptive histogram equalization (CLAHE), or brightness restoration methods based on Retinex theory. In the experiment, it was assumed that the average brightness of the original image was only 90 (8-bit grayscale range 0-255), while the optimal visual brightness should be increased to around 150, which can be achieved through global gamma correction (Gamma = 1.2–1.5) and multi-scale Retinex enhancement. In multispectral data, brightness compensation can also be performed on shadow areas based on NIR (near-infrared) band information to avoid color distortion caused by brightness enhancement. Brightness enhancement not only adjusts the overall brightness curve but also needs to suppress bright areas to avoid overexposure. This can be achieved by limiting the local brightness enhancement range using enhancement algorithms with limited local contrast. The final brightness-enhanced image is visually clearer, with more obvious details and textures (such as road markings, vehicle outlines, and traffic signal facilities), which is beneficial to improving the accuracy of subsequent semantic segmentation.

[0032] The enhanced street view image is decomposed into different semantic categories, such as roads, lane lines, vehicles, sidewalks, green belts, and buildings, for subsequent morphological analysis. Pixel-level segmentation can be achieved using deep convolutional neural networks (such as DeepLabv3+, U-Net, and SegFormer). Training data can be derived from open-source remote sensing image segmentation datasets (such as SpaceNet and DeepGlobe) and locally labeled traffic scene data, maintaining the same resolution as the original image (0.5–1 meter). Batch size can be set to 8–16, and training epochs can be between 200–500 to ensure model convergence. The segmentation model can be combined with a multi-scale feature extraction module to adapt to targets of different sizes in satellite imagery. During inference, to improve speed, the image can be sliced ​​(e.g., 1024×1024 pixel blocks) and processed in parallel, and finally stitched back into a panoramic mask. To improve robustness under different lighting, seasonal, and climatic conditions, the model can be trained using data augmentation (rotation, flipping, color perturbation, random cropping, etc.). The output semantic mask assigns a corresponding category label to each pixel (e.g., road=1, vehicle=2, building=3, etc.) and saves it in GeoTIFF or PNG format for subsequent morphological analysis. The pixel set and boundary contour of each target are obtained through connected component analysis or contour detection (e.g., findContours in OpenCV). Next, the spatial geometric features of each target are calculated, including the minimum bounding rectangle, aspect ratio, area (number of pixels), perimeter, and orientation angle (obtained through principal component analysis (PCA)). For roads and lane lines, skeletonization is performed to obtain the centerline for constructing road topology; for vehicles, the shape of the bounding ellipse is calculated to infer the vehicle's orientation and size (which can be converted to meters based on pixel resolution, e.g., 1 pixel = 0.5 meters). To ensure spatial accuracy, the pixel coordinates are converted to geographic coordinates (latitude and longitude or projected coordinate system, such as WGS84 / UTM) using previously obtained georegistration information. In the experiment, it was assumed that 1200 vehicle targets were detected, the total road length was approximately 52 kilometers, and the green area coverage was approximately 21%. This spatial and geometric data will provide the basic input for subsequent deep feature mining.

[0033] This process transforms two-dimensional semantic and morphological information into three-dimensional spatial information, providing depth features such as height, volume, and hierarchy for subsequent rendering. Implementation methods can be based on multi-view stereo reconstruction (MVS), optical flow estimation combined with DEM (Digital Elevation Model) constraints, or methods based on depth estimation networks (such as MiDaS and DPT). In experimental environments, if the satellite platform provides dual-view or multi-view imagery, disparity maps can be calculated through semi-global matching, and the disparity can be converted into elevation by combining known camera extrinsic and intrinsic parameters (satellite orbital parameters). For single-view imagery, depth estimation networks can be used to predict the relative depth value of each pixel under semantic mask constraints, and a reasonable height range can be set according to the target category (e.g., 1.5–2 meters for cars, 3–4 meters for buses, and buildings calibrated based on local POI data or LiDAR data). The extracted depth information will be stored in the form of point clouds, raster elevation maps, or 3D meshes, with each pixel corresponding to an (X, Y, Z) coordinate, forming a complete three-dimensional structural framework for urban transportation. To achieve realistic rendering, the texture information of the original image is mapped onto the surface of a 3D model. First, the enhanced street scene image is registered with the 3D geometric model (UV mapping) to ensure that each 3D vertex corresponds to the correct image pixel. For categories such as roads, buildings, and vehicles, different rendering shaders can be selected based on material characteristics. For example, in a PBR (Physically Based Rendering) model, diffuse maps (Albedo), normal maps, metallic maps, and roughness maps are defined. In experiments, the reflectivity of road materials can be set to approximately 0.05–0.1, the roughness of building facades to 0.6–0.8, and the metallicity of vehicle surfaces to 0.3–0.5. To enhance visual immersion, dynamic lighting simulation based on the real sun's position (calculating the sun's altitude and azimuth using acquisition time and latitude / longitude) can be added during the rendering process. Shadow mapping and ambient occlusion are also introduced to enhance the sense of depth. The final output texture rendering parameters include material type, light intensity, texture path, normal direction, etc. These parameters will be directly called by the rendering engine (such as Unity, Unreal Engine or CesiumJS) to achieve high-fidelity 3D visualization of urban traffic conditions.

[0034] In this embodiment, see Figure 3 The diagram below illustrates the detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include: Geospatial information is analyzed based on real-time satellite street view images to extract basic geospatial information; Based on geospatial information, the distribution of building heights, road geometry, and terrain undulations are calculated to obtain the morphological features of the three-dimensional scene. Visual recognition of building facade gloss and road surface material is performed on real-time satellite street view images to obtain the scene material illumination distribution characteristics. Lighting calculations and shadow rendering are performed based on the lighting distribution characteristics of scene materials to obtain scene background rendering parameters; Based on the scene background rendering parameters and scene texture rendering parameters, traffic scene background modeling is performed to construct a three-dimensional virtual scene background image; Based on the morphological features of the 3D scene, a physics engine is integrated into the 3D virtual scene background image to construct a real-time scene background model.

[0035] In this embodiment, coordinate system analysis and projection transformation are performed on the original image to align the image data to a unified geographic coordinate system (such as WGS84 or UTM). High-resolution image feature detection algorithms (such as Canny edge detection and SIFT feature matching) are used to identify ground feature outlines, and deep learning semantic segmentation networks (such as HRNet and SegFormer) are combined for ground feature classification. For building outlines, a contour optimization algorithm based on morphological dilation-erosion can be used to remove artifacts; for road network extraction, a skeletonized road centerline extraction method can be used, combined with topological connectivity detection to generate a road network topology map. In the experiment, it was assumed that the image resolution was 0.5 m, the area size was 10 km², the number of extracted buildings was approximately 4,500, and the total road length was approximately 130 km. The extracted geospatial basic information is stored in vector data format (such as Shapefile or GeoJSON) as an important input for subsequent 3D morphological feature calculations. The two-dimensional geospatial data is then transformed into 3D morphological information. Building heights are obtained by generating a Digital Surface Model (DSM) from multi-view satellite stereo imagery and then performing differential calculations using a Digital Elevation Model (DEM) to obtain the building height (H = DSM – DEM). In the experiment, if the resolution of the satellite dual-view image pair is 0.5 m, a 1 m elevation accuracy can be obtained using the Semi-Global Matching algorithm, thus obtaining a building height distribution map. Road geometry extraction is based on the road centerline obtained in step one, and a road geometry model is obtained through curvature calculation, piecewise fitting, and lane width estimation. Topographic relief changes can be directly analyzed from the DEM to calculate the slope and aspect distribution, which are used to influence subsequent lighting and traffic flow simulations. Example experimental parameters: average building height 15.3 m, tallest building 82 m; average road width 8.5 m, maximum slope area 12°. The final 3D scene morphological features are saved as point clouds (LAS format) or triangular meshes (TIN format) as the geometric basis for constructing the 3D model.

[0036] The building facade and road areas are segmented from the original image (using the semantic masking results from the previous two steps). Building facade glossiness estimation can be performed by analyzing the pixel brightness distribution and polarization information (if a polarization channel is available) of highlight areas to determine the surface reflection type (spectral, diffuse, or mixed). Road surface material identification can be based on texture analysis methods (such as Gray-Level Co-occurrence Matrix (GLCM) and Local Binary Patterns (LBP)) combined with convolutional neural networks (ResNet, EfficientNet) to classify different materials such as asphalt, concrete, and paving bricks. In the experiment, glossiness can be represented by 0–1 floating-point numbers; the average glossiness of building glass curtain walls is approximately 0.85, concrete walls approximately 0.35, and asphalt roads approximately 0.15. Illumination distribution characteristics are obtained through global illumination estimation algorithms (such as spherical harmonic illumination analysis) to describe the brightness and shadow coverage ratio of different areas at a specific time. The results are stored in the form of illumination intensity distribution maps and material classification maps to prepare for subsequent lighting rendering. The distribution of light and shadow is calculated by combining material properties and spatial geometry. Based on the image acquisition time and geographic coordinates, the solar altitude angle and azimuth angle are obtained using astronomical calculation formulas (e.g., acquisition time is local time 14:30, latitude 39.9°, solar altitude angle is approximately 52°, azimuth angle is approximately 218°). Then, using lighting calculation formulas based on the Phong model or PBR (Physically Based Rendering), considering gloss, roughness, and reflectivity parameters, the lighting intensity of each surface is generated. Shadow rendering can employ ray tracing or shadow mapping methods, using 3D building and terrain data to calculate the shadow projection range and soft / hard boundary effects. In the experiment, the rendering resolution was set to 4096×4096 pixels, the shadow softening radius was set to 1.2 m, and the maximum shadow length was approximately 130 m. The output scene background rendering parameters include lighting direction vectors, ambient light intensity, diffuse and specular reflection coefficients, shadow map data, etc., which are used in the background construction stage.

[0037] The previously obtained geometric models, texture maps, and lighting parameters are integrated to generate a complete 3D background. First, the 3D scene morphological features (results from step two) are imported into a rendering engine (such as Unreal Engine, Unity, or CesiumJS), loading the corresponding texture maps (results from step three) and lighting and shadow parameters (results from step four). To ensure the visualization efficiency of city-scale data, tile-based rendering and Level of Detail (LOD) techniques are used, loading high-precision data at the center of the view and low-precision data at distances to reduce GPU load. Material rendering uses PBR shaders, loading corresponding Albedo, Normal, Roughness, and Metallic textures according to different terrain types (buildings, roads, water bodies, vegetation). In experiments, a single 2×2 km scene model has approximately 12 million triangles, and the rendering frame rate can be stably maintained at 60 FPS (RTX 3080 GPU environment). The final generated 3D virtual scene background image can serve as the static base for the traffic simulation environment. By combining the background scene with a physics engine, the system achieves real-time interactivity and dynamic response capabilities. The physics engine can be Bullet, PhysX, or Havok, generating colliders and physical properties (friction coefficient, elasticity coefficient, etc.) based on the 3D scene's morphological features (road slope, building height, terrain undulations). For roads, multi-segment collision meshes can be created to support vehicle dynamics simulation (acceleration, braking, steering); for buildings and overpasses, rigid body models can be created and collision detection ranges can be set; for terrain, a height field can be created to support the calculation of the slope impact on traffic flow. In the experiment, the friction coefficient was set to 0.75 for asphalt roads, 0.85 for concrete roads, and 0.45 for grass; the gravitational acceleration was set to 9.81 m / s². After integration, the virtual scene can not only serve as a visual background for traffic flow but also interact physically in real-time with vehicles, pedestrians, and environmental dynamic events (such as shadow changes and rain / snow effects), thereby achieving highly realistic traffic condition simulation rendering.

[0038] In this embodiment, reference Figure 4 The diagram below illustrates the detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include: Acquire city-level multi-source traffic monitoring data streams, which include camera video streams, vehicle GPS trajectories, mobile phone monitoring, and roadside vehicle sensors; Timestamps are extracted from city-level multi-source traffic monitoring data streams and then standardized to obtain time-series synchronized data streams. Anomaly detection is performed on the time-synchronized data stream, anomaly data is marked and removed to obtain the optimized data stream; Traffic vehicle multi-source identification is performed based on the eliminated and optimized data stream, and dynamic particle abstraction processing is carried out and marked as dynamic particles; The trajectory change rate, velocity fluctuation pattern, lane change frequency, and particle spacing change of the dynamic particles are calculated to construct a multidimensional particle feature matrix.

[0039] In this embodiment, data is collected in real time from different types of urban traffic sensing devices to form a multi-source traffic monitoring data stream covering the entire city. Camera video streams can come from high-definition monitoring equipment (resolution 1080p or higher, frame rate 25–30 fps) deployed by traffic management departments on major roads and intersections, obtained through RTSP or dedicated data transmission protocols; vehicle GPS trajectory data can be collected from taxis, buses, logistics vehicles, etc., with a positioning accuracy of generally 5–10 meters and a sampling frequency of 1–5 seconds; mobile phone monitoring data can be obtained through anonymized operator base station positioning or APP embedded data, suitable for analyzing large-scale travel patterns; roadside vehicle sensors (such as geomagnetic detectors, millimeter-wave radar, and lidar) provide accurate vehicle detection, speed, and vehicle length information. To ensure data availability, a data access platform needs to be established, supporting multiple communication protocols (HTTP, MQTT, WebSocket, etc.), and data buffering and latency compensation mechanisms need to be set up to avoid single-source latency affecting the overall analysis. In the experiment, a typical city-level system can simultaneously access 3,000 video streams, GPS data from over 500,000 vehicles, and real-time information from thousands of roadside sensors, achieving a data throughput of hundreds of MB per second, providing a rich data foundation for subsequent traffic simulations. The acquisition times of multi-source data are often inconsistent; clock errors from different devices and network transmission delays can lead to timestamp discrepancies. Extracting and standardizing the timestamps from each data source ensures that subsequent analysis is based on a unified time reference. The acquisition time of keyframes from video streams can be extracted through video metadata (such as NTP timestamps in the RTSP protocol) or the frame encoding time field; the UTC time output by the positioning module is directly used from GPS data; the actual acquisition time of mobile phone monitoring data can be estimated based on the reception time recorded by the base station or server; and the timestamps of roadside sensors are provided by the local device clock. Timestamp standardization requires converting all times to a unified UTC standard time and calibrating the time source of the acquisition system using a high-precision time synchronization protocol (PTP or NTP), with the maximum allowable error controlled within ±50 milliseconds. In the experiment, if the time difference between the original data sources is 1–2 seconds, after standardization, the data streams from different sources can be aligned to the same millisecond-level timeline to form a “time-synchronized data stream”, laying the foundation for multi-source information fusion.

[0040] Due to sensor errors, network packet loss, and occlusion interference, outliers inevitably appear in multi-source traffic data. Examples include sudden jumps of several kilometers in vehicle GPS coordinates, flickering or disappearance of identified targets in video streams, and negative speed values. Anomaly detection can be divided into rule-based detection and model-based detection. Rule-based detection is based on threshold judgment; for example, points where vehicle speed exceeds 200 km / h or position offset exceeds 500 meters are directly marked as anomalies. Model-based detection can use statistical models (such as the three Sigma principle) or machine learning methods (Isolation Forest, LOF) to identify anomaly patterns. For video detection data, anomalies can also be determined based on temporal continuity, such as abnormal displacement of the same license plate in adjacent frames. Detected anomaly data needs to be marked and removed in subsequent processing or repaired using interpolation methods (linear interpolation, Kalman filtering). In experiments, the proportion of outliers in the original time-series data may be 2–5%, which can be reduced to below 0.1% after cleaning, thus significantly improving the stability of traffic analysis and simulation models. The output result is the "removed and optimized data stream". The same vehicle from different data sources is matched and fused, and then abstracted as a "dynamic particle" for unified object representation in simulation rendering. Multi-source recognition can employ multi-feature matching methods: GPS data provides location trajectory, video recognition provides license plate, vehicle type, and color features, and roadside sensors provide information such as vehicle length and speed. Association judgment is performed through time- and space-based matching windows (e.g., time error ≤ 0.5 seconds, spatial distance ≤ 20 meters), combined with probabilistic data association algorithms (such as JPDA or Hungarian algorithms) to achieve multi-source fusion. After recognition, each vehicle is abstracted as a "dynamic particle" with a unique ID and dynamic attributes such as position, speed, orientation, and acceleration. The benefit of particle abstraction is reduced rendering burden on the simulation engine and facilitates traffic flow calculation. In experiments, a typical city center area may have 50,000–100,000 dynamic particles during the morning rush hour, with an update frequency of 10 Hz to meet real-time simulation requirements.

[0041] The motion behavior of dynamic particles is quantified into multidimensional features, forming an input matrix that can be used for simulation analysis. The trajectory change rate can be obtained by calculating the change in trajectory direction of the particle per unit time (based on the vector angle); the speed fluctuation pattern can be calculated using a sliding window to determine the mean, variance, and spectral characteristics of the speed, identifying the proportion of acceleration, deceleration, and constant speed states; the lane change frequency needs to be combined with road geometry data, counted by the proportion of particle lateral position changes exceeding the lane width; the particle spacing change can be calculated by determining the time-series change rate of the distance between preceding and following vehicles, reflecting the stability of following behavior. In the experiment, a 60-second analysis window can be selected, and each particle can generate a vector containing 10–15 features, such as an average speed of 42 km / h, a standard deviation of 5 km / h, a lane change frequency of 0.8 times / minute, and a spacing change standard deviation of 3.2 meters. The feature vectors of all particles are stacked temporally to form a multidimensional particle feature matrix (dimension N particles × M features × T time), providing direct data input for traffic flow state analysis, predictive model training, and visualization rendering.

[0042] In this embodiment, step S4 includes the following steps: Identify urban road networks based on real-time satellite street view images; Multi-level intersection topology association mining is performed on the urban road network to obtain the topological structure relationship of the road network; Based on the road network topology, a graph convolutional neural network is modeled to construct a road topology convolutional network. Based on real-time satellite street view images, traffic light phases, lane capacity, and traffic control information at intersections are identified to obtain road network traffic characteristics. Based on the traffic characteristics of the road network, a global traffic network state perception model is constructed by performing global traffic network state perception on the road topology convolutional network.

[0043] In this embodiment, satellite imagery is preprocessed, including geometric correction (eliminating the influence of terrain and satellite attitude), radiometric calibration (unifying brightness and color gamut), and dehazing (based on multi-temporal synthesis or deep learning dehazing algorithms) to ensure the input image is clear and aligned with the geographic coordinate system. Road identification can employ deep learning semantic segmentation methods, such as DeepLabv3+, U-Net, and HRNet. Training data is derived from existing remote sensing road extraction datasets (SpaceNet, DeepGlobe) and locally labeled samples, with a spatial resolution maintained at 0.3–0.5 m / pixel. After segmenting the road regions, morphological processing (erosion, dilation, skeletonization) is used to extract road centerlines, and connected component analysis is used to generate road vectorized data. In experiments, for a 100 km² urban image, thousands of road features can be identified, with an average accuracy (IoU) exceeding 0.85. The output of this step is a complete road geometric network, including nodes (intersections) and edges (road segments), providing foundational data for subsequent topology analysis. The road centerline data is segmented into nodes to identify intersection locations (intersections of two or more road centerlines) and a unique ID is assigned to each node. Multi-level topology association mining can be divided into macro-level (main roads, highways), meso-level (urban main roads, secondary roads), and micro-level (local roads, lanes), which can be layered based on attributes such as road width, speed limit information, and functional classification. The relationships between intersections can be identified through spatial buffer analysis to recognize neighboring nodes and adjacent road segments, and directed edges are established (considering directional constraints such as one-way streets and no-left-turn restrictions). In experiments, in a network of 1200 intersection nodes and 3500 road segments, multi-level association analysis generates a weighted directed graph, where weights can represent distance, travel time, or road capacity. The result of this step is a structured road network topology table (adjacency matrix or edge table), laying the foundation for graph convolution modeling.

[0044] Graph Convolutional Neural Networks (GCNs) are used for feature propagation and structural modeling of road networks. The input data consists of the road network topology adjacency matrix and road attribute features (length, number of lanes, speed limit, functional level, etc.). The core idea of ​​GCNs is to perform feature convolution on the graph structure, achieving weighted aggregation of features from neighboring nodes. The model can employ a two- or three-layer GCN structure. The first layer aggregates local neighbor features, and the second layer captures a wider range of network structural features. ReLU or LeakyReLU activation functions can be used, with Adam as the optimizer and a learning rate of 0.001–0.005. In experiments, the model was trained on a network of 3500 road segments, with each node having a feature dimension of 16. After two layers of GCN convolution, a 64-dimensional embedding vector is output, effectively capturing the spatial dependencies of traffic flow in the road network. The completed road topology convolutional network can serve as the core computational model for subsequent traffic state prediction, path planning, and simulation rendering. The intersection area is located in satellite street view imagery (using the intersection node locations generated in step one). Then, deep learning object detection algorithms (YOLOv8, Faster R-CNN) are used to identify the location and shape of traffic lights. Signal phase durations (durations of red, green, and yellow lights) are inferred through color recognition and video frame time series analysis. Lane capacity is calculated by combining road geometry (measured from imagery) and conventional design standards (e.g., maximum capacity per lane of 1800 pcu / h), and corrected based on real-time traffic flow detected by roadside sensors or video. Traffic control information (no entry, one-way, tidal flow lanes) can be obtained through image analysis of signs or by fusing traffic management databases. In the experiment, the traffic light cycle of a typical intersection might be 120 seconds, with 45 seconds of green, 65 seconds of red, and 10 seconds of yellow; the capacity of each of the two lanes in both directions is approximately 7200 pcu / h. Finally, the road network traffic characteristic data output in this step is combined with a road topology convolutional network to model dynamic changes in traffic flow.

[0045] To achieve global traffic network state awareness, static topology information (the GCN model in step three) and dynamic traffic features (the results in step four) must be fused. Methodologically, real-time traffic features (remaining traffic light phase time, current traffic flow, lane utilization, average speed, etc.) can be added to the node feature vectors of the GCN. The model is then trained using a Spatiotemporal Graph Convolutional Network (ST-GCN) or a Graph Attention Network (GAT) to capture both spatial dependencies and temporal dynamic changes. The input is a sequence of node features from the past T time slices (e.g., T=12, each time slice is 5 minutes), and the output is a predicted traffic state for several future time slices (e.g., speed, flow rate, congestion index). In experiments, using data from the past hour to predict the speed of the entire road network for the next 15 minutes, the average absolute error can be controlled within 5 km / h. The final constructed road network state awareness model can directly provide real-time road traffic status to the traffic simulation engine, enabling high-precision urban traffic visualization rendering based on satellite remote sensing data.

[0046] In this embodiment, the specific steps of step S5 are as follows: Traffic density distribution data is obtained by calculating the traffic density distribution based on the multidimensional particle feature matrix. Real-time vehicle flow particle mapping is performed on the road network state perception model based on traffic density distribution data to construct a vehicle traffic simulation model. A digital twin model of traffic conditions is constructed by using a real-time scene background model as the rendering background and updating and overlaying the vehicle traffic simulation model in real time. The performance constraints of the traffic condition digital twin model rendering are evaluated, and adaptive inter-frame rendering intervals are calculated to perform traffic condition visualization rendering tasks.

[0047] In this embodiment, the multidimensional particle feature matrix generated in the previous steps (containing the trajectory change rate, speed fluctuation pattern, lane change frequency, particle spacing change, etc. for each vehicle) is transformed into a spatialized traffic density distribution. First, the road network is divided into fixed-length analysis units (e.g., road segments of 50–100 meters), and the number of particles within each segment in each time slice is counted to obtain the correspondence between the number of vehicles and the road space. To ensure calculation accuracy, the movement of vehicles between road segments needs to be considered, and the density distribution of vehicles across segments is corrected through linear interpolation or trajectory fitting. Traffic density can be calculated using the number of vehicles per unit length (veh / km) or the flow density formula commonly used in traffic engineering (Density = Flow / Speed). In the experiment, the density on main roads during the morning rush hour can typically reach 60–80 vehicles / km, while the density during off-peak hours can drop to 10–20 vehicles / km. The calculation results need to be smoothed in time and space (e.g., Gaussian smoothing or Kalman filtering) to remove "noise" density peaks caused by instantaneous fluctuations. The final output traffic density distribution data is stored in the form of a road network-based spatial raster or vector attribute table, with a time resolution that can be set to 1–5 seconds, supporting subsequent simulated vehicle mapping. The traffic density distribution is combined with a road network state perception model (including road geometry, traffic characteristics, traffic light status, etc.) to generate a simulated vehicle layout that conforms to actual traffic flow characteristics. First, based on the current density value and road capacity of each road segment, the number of vehicle particles to be generated on that segment is calculated. Then, considering factors such as segment length, number of lanes, and speed limits, the particles are distributed evenly or according to congestion patterns on the road. For intersection areas, traffic light phases and waiting queue lengths need to be considered to ensure that the queuing phenomena in the simulation match reality. The dynamic attributes of the particles (speed, acceleration, lane-changing behavior) can be sampled from a preceding multidimensional feature matrix to preserve realistic traffic fluctuation patterns. In the experiment, a 2 km long road segment with three lanes in each direction can generate approximately 300–400 particles during peak hours, with an update frequency set to 10–20 Hz to ensure real-time changes in vehicle position and speed. The final vehicle traffic simulation model is a particle system that can evolve over time, providing a dynamic entity for subsequent digital twin visualization.

[0048] This process merges a 3D static background (a real-time scene background model composed of satellite imagery, terrain data, and building models) with a dynamic vehicle simulation particle system to generate a complete digital twin model of traffic conditions. First, the scene background model is loaded into a rendering engine (such as Unreal Engine, Unity, or CesiumJS), and particle positions are aligned to a real-world geographic coordinate system. Vehicle particles are bound to corresponding 3D vehicle models (cars, buses, trucks, etc.), and vehicle animation is driven by particle attributes (speed, direction, lane). To ensure realism, real-time lighting and shadow effects are added to ensure consistent lighting and shadows between the vehicles and the background scene. Furthermore, behaviors such as queuing, lane changing, and overtaking are simulated based on traffic flow characteristics, and vehicle positions and states are updated in real time. In experiments, a single 4 km² urban background can simultaneously render 5,000–8,000 dynamic vehicle objects, maintaining 60 FPS with a latency of less than 50 ms under an RTX 3080 GPU environment. The result of this step is a real-time interactive digital twin urban traffic scene that highly replicates real-world traffic conditions.

[0049] Digital twin rendering requires a balance between visual quality and performance; therefore, this step necessitates dynamically evaluating rendering performance and adjusting the inter-frame interval. First, rendering performance metrics, including frame rate (FPS), GPU / CPU utilization, and the number of rendering batches (Draw Calls), are collected in real time to detect performance bottlenecks (such as excessive particle count or excessively large textures). Based on the performance evaluation results, adaptive algorithms (such as frame rate stabilization algorithms based on PID control or binary search) are used to dynamically adjust the inter-frame rendering interval. For example, under high load, the rendering interval is adjusted from 16.7 ms (60 FPS) to 33.3 ms (30 FPS) to ensure smooth operation. For objects that do not affect critical visual perception (distant vehicles, background buildings), LOD (Level of Detail) or deferred rendering strategies can be used to reduce computational load. In experiments, when the number of particles increased to over 10,000, the system automatically adjusted the rendering interval to 25 ms, ensuring an overall latency of less than 100 ms. Ultimately, the rendering operation can maintain stable operation under different hardware environments and traffic flow intensities, providing a reliable digital twin display for traffic visualization and decision support.

[0050] In this embodiment, the specific steps for evaluating the rendering performance constraints of the traffic condition digital twin model and calculating the adaptive inter-frame rendering interval to perform the traffic condition visualization rendering operation are as follows: The rendering performance constraints of the traffic condition digital twin model are evaluated to obtain the rendering requirement evaluation value; The rendering requirement assessment values ​​are used to allocate GPU parallel computing resources and generate rendering resource configuration parameters. Multi-level rendering detail adjustments are made to the rendering resource configuration parameters to obtain a rendering precision control strategy; The visual effects are dynamically allocated based on the rendering precision control strategy, and the inter-frame rendering interval is calculated adaptively to construct a real-time traffic condition rendering image sequence to complete the traffic condition visualization rendering task.

[0051] In this embodiment, key performance indicators of the current digital twin model are collected, including the total number of polygons in the scene (Polygon Count), total texture memory usage (Texture Memory), real-time particle count (VehiclesParticles Count), and lighting pass time. To more accurately assess rendering requirements, the GPU stage time (Vertex Shader, Pixel Shader, Post-Processing, etc.) and CPU scene management time of the rendering pipeline also need to be monitored. By setting performance thresholds (e.g., target frame rate ≥ 60 FPS, GPU utilization ≤ 90%), the Rendering Demand Index (RDI) of the current scene is calculated. This value can be obtained using the formula RDI = (Scene Polygon Load Weight × α) + (Texture Memory Usage Weight × β) + (Particle Count Weight × γ) + (Lighting Pass Time Weight × δ). In the experiment, for a test scene of a 4 km² urban area with 8000 particles, the typical RDI value is between 0.65 and 0.85. When the RDI exceeds 0.85, it indicates that rendering resource optimization is needed to prevent frame drops and latency. Combining the RDI value with hardware detection results (GPU model, number of CUDA cores, memory size, bandwidth, etc.), the priority and resource allocation of each rendering task are dynamically calculated. For example, real-time traffic particle animation (vehicle movement, lane changing, collision detection) is assigned to the GPU Compute Shader module, high-resolution texture mapping is handled by dedicated Texture Units, and shadow and lighting calculations are arranged in parallel rasterization channels. The resource allocation strategy is based on Tiled Rendering and Task Batching techniques, dividing the scene into multiple rendering blocks (e.g., 256×256 pixels), which are then processed in parallel by different GPU cores. In experiments, an RTX 3080 GPU (8704 CUDA cores, 10GB memory) can simultaneously process 10,000+ particles and 2GB of textures at 60 FPS. The final output rendering resource configuration parameters include task priority tables, core allocation tables, and video memory caching strategies, which are used to guide subsequent adjustments to rendering details.

[0052] The scene is adjusted in multiple levels of rendering detail based on GPU resource configuration parameters to generate a rendering precision control strategy. These adjustments mainly include Level of Detail (LOD), dynamic texture streaming, shadow quality level, and post-processing effects on / off. Specifically, the model's LOD is set to 3-5 layers. Models at closest view distance retain full high-precision polygons and high-resolution textures, while distant models automatically switch to a low-poly version (reducing the number of faces by 50%-80%) and reduce texture resolution (e.g., from 4K to 512×512). For vehicle models, only simple shape outlines and monochrome textures are retained at distances exceeding 300 meters. Shadow calculations can be switched to half-resolution shadow maps depending on performance. In experiments, by reducing particle detail by 30% and shadow precision by 20% at urban intersections, the average GPU utilization decreased from 92% to 78% while maintaining acceptable visual quality. The final rendering precision control strategy is stored in the form of a parameter table and can be updated in real time.

[0053] The rendering precision control strategy is applied to the actual rendering process to achieve dynamic allocation of visual effects and adaptive adjustment of the inter-frame rendering interval. Dynamic allocation of visual effects refers to dynamically enabling or disabling certain visual effects (such as reflections, depth of field, volumetric fog, motion blur, etc.) based on current performance during the rendering process, prioritizing the visual quality of critical elements (such as foreground vehicles and road details), while degrading the rendering of non-critical elements (such as distant buildings and the sky background). Adaptive inter-frame rendering interval calculation is based on real-time FPS and RDI values, dynamically adjusting the rendering time budget for each frame. For example, when performance is good (FPS > 70), the interval can be locked at 16.7 ms (60 FPS), while under high load (FPS < 50), the interval can be extended to 25-33 ms to balance smoothness and hardware load. The final output is a continuous sequence of real-time traffic rendering images, which can be presented as a video stream or interactive view in the traffic digital twin platform. Experimental results show that in a 4 km² urban traffic simulation scenario, after dynamic allocation and adaptive rendering, the system can maintain a stable frame rate of 55-65 FPS with 12,000 particles and complex lighting enabled, and ensure a latency of less than 100 ms.

[0054] In this embodiment, a data simulation and visualization rendering system for urban traffic conditions is provided, used to execute the data simulation and visualization rendering method for urban traffic conditions as described above, including: The pixel segmentation module is used to extract real-time satellite street view images based on urban satellite remote sensing scans, perform scene pixel-level segmentation, and obtain scene texture rendering parameters. The background modeling module is used to perform scene lighting visual recognition based on real-time satellite street view images and to perform traffic scene background modeling based on scene texture rendering parameters, thereby constructing a real-time scene background model. The particle abstraction module is used to acquire city-level multi-source traffic monitoring data streams, perform vehicle status perception, and construct a multi-dimensional particle feature matrix. The road network status perception module is used to perform road network topology correlation analysis and global traffic network status perception based on real-time satellite street view images, and to build a road network status perception model. The visualization rendering module is used to perform real-time vehicle flow particle mapping on the road network state perception model based on the multi-dimensional particle feature matrix, and to perform real-time background update and overlay based on the real-time scene background model in order to perform traffic condition visualization rendering operations.

[0055] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0056] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein are implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A data simulation visualization rendering method for urban traffic conditions, characterized in that, Includes the following steps: Step S1: Extract real-time satellite street view images based on urban satellite remote sensing scans, perform scene pixel-level segmentation, and obtain scene texture rendering parameters; Step S2: Perform scene lighting visual recognition based on real-time satellite street view images, and model the traffic scene background based on scene texture rendering parameters to construct a real-time scene background model; Step S3: Acquire city-level multi-source traffic monitoring data streams, perform vehicle status perception, and construct a multi-dimensional particle feature matrix; Step S4: Perform road network topology correlation analysis and global traffic network status perception based on real-time satellite street view images, and construct a road network status perception model; Step S5: Perform real-time vehicle flow particle mapping on the road network state perception model based on the multi-dimensional particle feature matrix, and perform real-time background update and overlay based on the real-time scene background model to execute traffic condition visualization rendering.

2. The data simulation and visualization rendering method for urban traffic conditions according to claim 1, characterized in that, The specific steps of step S1 are as follows: A panoramic satellite remote sensing scan of urban traffic was conducted to extract real-time satellite street view images; Global brightness optimization is performed on real-time satellite street view images to construct a brightness-enhanced street view image; Scene pixel-level segmentation is performed on the brightness-enhanced street scene image to obtain the traffic scene semantic mask; Based on the semantic mask of the traffic scene, we perform element-by-element morphological analysis to obtain the spatial location and geometric shape of each element in the scene. Based on the spatial location and geometry, depth element feature mining is performed to extract scene depth information features; The scene depth information features are textured and rendered to obtain scene texture rendering parameters.

3. The data simulation and visualization rendering method for urban traffic conditions according to claim 1, characterized in that, The specific steps of step S2 are as follows: Geospatial information is analyzed based on real-time satellite street view images to extract basic geospatial information; Based on geospatial information, the distribution of building heights, road geometry, and terrain undulations are calculated to obtain the morphological features of the three-dimensional scene. Visual recognition of building facade gloss and road surface material is performed on real-time satellite street view images to obtain the scene material illumination distribution characteristics. Lighting calculations and shadow rendering are performed based on the lighting distribution characteristics of scene materials to obtain scene background rendering parameters; Based on the scene background rendering parameters and scene texture rendering parameters, traffic scene background modeling is performed to construct a three-dimensional virtual scene background image; Based on the morphological features of the 3D scene, a physics engine is integrated into the 3D virtual scene background image to construct a real-time scene background model.

4. The data simulation and visualization rendering method for urban traffic conditions according to claim 1, characterized in that, Step S3 is as follows: Acquire city-level multi-source traffic monitoring data streams, which include camera video streams, vehicle GPS trajectories, mobile phone monitoring, and roadside vehicle sensors; Timestamps are extracted from city-level multi-source traffic monitoring data streams and then standardized to obtain time-series synchronized data streams. Anomaly detection is performed on the time-synchronized data stream, anomaly data is marked and removed to obtain the optimized data stream; Traffic vehicle multi-source identification is performed based on the eliminated and optimized data stream, and dynamic particle abstraction processing is carried out and marked as dynamic particles; The trajectory change rate, velocity fluctuation pattern, lane change frequency, and particle spacing change of the dynamic particles are calculated to construct a multidimensional particle feature matrix.

5. The data simulation and visualization rendering method for urban traffic conditions according to claim 1, characterized in that, The specific steps of step S4 are as follows: Identify urban road networks based on real-time satellite street view images; Multi-level intersection topology association mining is performed on the urban road network to obtain the topological structure relationship of the road network; Based on the road network topology, a graph convolutional neural network is modeled to construct a road topology convolutional network. Based on real-time satellite street view images, traffic light phases, lane capacity, and traffic control information at intersections are identified to obtain road network traffic characteristics. Based on the traffic characteristics of the road network, a global traffic network state perception model is constructed by performing global traffic network state perception on the road topology convolutional network.

6. The data simulation and visualization rendering method for urban traffic conditions according to claim 1, characterized in that, The specific steps of step S5 are as follows: Traffic density distribution data is obtained by calculating the traffic density distribution based on the multidimensional particle feature matrix. Real-time vehicle flow particle mapping is performed on the road network state perception model based on traffic density distribution data to construct a vehicle traffic simulation model. A digital twin model of traffic conditions is constructed by using a real-time scene background model as the rendering background and updating and overlaying the vehicle traffic simulation model in real time. The performance constraints of the traffic condition digital twin model rendering are evaluated, and adaptive inter-frame rendering intervals are calculated to perform traffic condition visualization rendering tasks.

7. The data simulation and visualization rendering method for urban traffic conditions according to claim 6, characterized in that, The specific steps for evaluating the rendering performance constraints of the traffic condition digital twin model and calculating the adaptive inter-frame rendering interval to perform traffic condition visualization rendering operations are as follows: The rendering performance constraints of the traffic condition digital twin model are evaluated to obtain the rendering requirement evaluation value; The rendering requirement assessment values ​​are used to allocate GPU parallel computing resources and generate rendering resource configuration parameters. Multi-level rendering detail adjustments are made to the rendering resource configuration parameters to obtain a rendering precision control strategy; The visual effects are dynamically allocated based on the rendering precision control strategy, and the inter-frame rendering interval is calculated adaptively to construct a real-time traffic condition rendering image sequence to complete the traffic condition visualization rendering task.

8. A data simulation and visualization rendering system for urban traffic conditions, characterized in that, The data simulation visualization rendering method for urban traffic conditions as described in claim 1 includes: The pixel segmentation module is used to extract real-time satellite street view images based on urban satellite remote sensing scans, perform scene pixel-level segmentation, and obtain scene texture rendering parameters. The background modeling module is used to perform scene lighting visual recognition based on real-time satellite street view images and to perform traffic scene background modeling based on scene texture rendering parameters, thereby constructing a real-time scene background model. The particle abstraction module is used to acquire city-level multi-source traffic monitoring data streams, perform vehicle status perception, and construct a multi-dimensional particle feature matrix. The road network status perception module is used to perform road network topology correlation analysis and global traffic network status perception based on real-time satellite street view images, and to build a road network status perception model. The visualization rendering module is used to perform real-time vehicle flow particle mapping on the road network state perception model based on the multi-dimensional particle feature matrix, and to perform real-time background update and overlay based on the real-time scene background model in order to perform traffic condition visualization rendering operations.

Images