A method and system for dynamic modeling of urban environments for embedded training simulators

Abstract

This patent application provides a method and system for dynamic urban environment modeling in an embedded training simulator, comprising: acquiring the current point matrix data template of an initial dynamic urban environment model; obtaining the rendering scene of the current viewpoint based on the current point matrix data template; determining the dynamically updated point matrix data based on the dynamically updated spatial coordinates and viewing direction; fusing the dynamically updated point matrix data and the current point matrix data; performing detailed rendering of the scene of the updated viewpoint again; until the rendering scene of each viewpoint is completed, the final dynamic urban environment model is obtained. This invention can generate the current point matrix data of the current viewpoint in real time based on the spatial coordinates and viewing direction of the current viewpoint, meeting the requirements for viewpoint changes under the hardware limitations of embedded systems; and can determine the dynamic update of each viewpoint through the dynamic changes of spatial coordinates and viewing direction.

Description

Technical Field

[0001] This invention patent application belongs to the field of embedded training simulator technology, specifically relating to a method and system for dynamic modeling of urban environments for embedded training simulators. Background Technology

[0002] With the widespread application of urban environment modeling in training simulators, virtual reality, and autonomous driving, the demand for real-time generation, dynamic updating, and adaptive detail adjustment of 3D urban environment data is increasing. However, the hardware resource limitations of embedded systems pose a significant challenge to the real-time processing of large-scale 3D urban environment raster data. Traditional 3D modeling methods typically rely on high-performance computing resources, making it difficult to achieve high efficiency and real-time performance in embedded environments, resulting in rendering latency and decreased visual quality.

[0003] Existing technologies face the following main problems when dynamically modeling urban environments in embedded training simulators: First, the real-time generation and dynamic updating of raster data requires high computing power and memory bandwidth, while the resources of embedded systems are limited and difficult to meet these requirements; second, traditional raster data rendering methods lack intelligent adjustment of visual saliency and level of detail, resulting in low rendering efficiency and inability to meet the real-time requirements of dynamic scenes; finally, existing dynamic update technologies usually use fixed frequency or simple threshold judgment, lacking the ability to adaptively adjust to scene complexity and viewpoint changes, and cannot optimize resource consumption while ensuring rendering quality.

[0004] Therefore, how to achieve real-time generation, dynamic updating, and adaptive adjustment of detail levels of large-scale 3D urban environment raster data under the hardware limitations of embedded systems has become a technical problem that urgently needs to be solved. Summary of the Invention

[0005] To overcome the shortcomings of the prior art, this patent application proposes a method for dynamic modeling of urban environments for embedded training simulators, comprising:

[0006] Obtain the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint, and determine the current dot matrix data template based on the spatial coordinates and viewing direction;

[0007] A lightweight random sampling algorithm is used to sample the current point data template to obtain the current point data. Based on the current point data, the scene of the current viewpoint of the initial urban environment dynamic model is rendered in detail to obtain the rendering scene of the current viewpoint.

[0008] When the spatial coordinates and viewing direction of the viewpoint are randomly and dynamically updated, the dynamically updated dot matrix data is determined based on the dynamically updated spatial coordinates and viewing direction.

[0009] The dynamically updated dot matrix data and the current dot matrix data are fused together, and the scene of the viewpoint after the initial urban environment dynamic model is updated is rendered in detail again until the rendering scene of each viewpoint is completed, so as to obtain the final urban environment dynamic model and complete the urban environment dynamic modeling.

[0010] The initial urban environment dynamic model is constructed based on a set of dynamic parameters of the target urban environment collected by a multi-source sensor array and through an embedded training simulator.

[0011] Preferably, the step of obtaining the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint, and determining the current dot matrix data template based on the spatial coordinates and viewing direction, includes:

[0012] Obtain the spatial coordinates and viewing direction of the initial dynamic urban environment model in the embedded training simulator at the current viewpoint;

[0013] Based on the spatial coordinates and viewing direction, extract spatiotemporally continuous urban environmental dynamic information from the initial urban environmental dynamic model;

[0014] Based on the aforementioned urban environmental dynamic information, the corresponding Level of Detail (LOD) level is determined;

[0015] From the pre-built LOD level library, find the raster data template that matches the LOD level.

[0016] Preferably, after extracting spatiotemporally continuous urban environmental dynamic information from the initial urban environmental dynamic model based on the spatial coordinates and viewpoint direction, and before determining the corresponding Level of Detail (LOD) level based on the urban environmental dynamic information, the method further includes:

[0017] Based on the dynamic information of the urban environment, determine the magnitude of change of the current viewpoint;

[0018] When the change amplitude is less than a preset change amplitude threshold, a low-frequency update strategy is adopted for the current viewpoint to perform dynamic updates;

[0019] When the change magnitude exceeds a preset change magnitude threshold, a high-frequency update strategy is adopted to dynamically update the current viewpoint.

[0020] Preferably, the step of employing a lightweight random sampling algorithm to sample the current point data template to obtain current point data, and then performing detailed rendering of the current viewpoint scene of the initial urban environment dynamic model based on the current point data to obtain the current viewpoint rendering scene, includes:

[0021] Based on the correlation of the initial urban environment dynamic model in the view frustum space at the current viewpoint, a probability-guided sampling function is established.

[0022] According to the probability-guided sampling function, the current dot matrix data template is sampled to obtain the current dot matrix data. The current dot matrix data is then subjected to visual saliency-driven adaptive filtering to obtain optimized current dot matrix data.

[0023] Extract the visual attention heatmap of the current viewpoint from the optimized current dot matrix data;

[0024] Calculate the heatmap value for each region of interest in the visual attention heatmap;

[0025] For regions of interest whose heatmap values ​​are greater than a preset heatmap threshold, bilateral filtering is used to preserve edge details. For regions of interest whose heatmap values ​​are less than or equal to the heatmap threshold, anisotropic diffusion filtering is used to reduce noise. This completes the detailed rendering of the scene at the current viewpoint, resulting in the rendered scene at the current viewpoint.

[0026] Preferably, the expression for the probability-guided sampling function is as follows:

[0027]

[0028] in, This represents the sampling probability of the i-th dot matrix data in the current dot matrix data template. Distance is the influencing factor. This refers to the i-th dot matrix data in the current dot matrix data template. As the directional influence factor, This refers to the viewpoint of the i-th dot matrix data in the current dot matrix data template. Geometric feature influencing factor Let i be the geometric characteristic function of the i-th dot matrix data in the current dot matrix data template. For color feature influencing factors, The color feature of the i-th dot matrix data in the current dot matrix data template. This is the normalization constant.

[0029] Preferably, the step of fusing the dynamically updated dot matrix data and the current dot matrix data to perform detailed rendering of the scene at the viewpoint after the initial urban environment dynamic model is updated, until the rendering scene of each viewpoint is completed, to obtain the final urban environment dynamic model, thus completing the urban environment dynamic modeling, includes:

[0030] A dynamic Kalman filter is used to calculate the spatial correlation matrix between the dynamically updated dot matrix data and the current dot matrix data. Based on the spatial correlation matrix, an adaptive weighted fusion algorithm is used to generate a hybrid dot matrix dataset.

[0031] Based on the hybrid raster dataset, a hierarchical rendering pipeline is used to perform detailed rendering of the scene at the viewpoint after the initial urban environment dynamic model is updated, resulting in the rendered scene after the viewpoint is updated.

[0032] The process continues until the rendering of each viewpoint is completed, resulting in the final dynamic model of the urban environment, thus completing the dynamic modeling of the urban environment.

[0033] Preferably, the step of using a hierarchical rendering pipeline to perform detailed rendering of the scene at the viewpoint after the initial dynamic urban environment model is updated, based on the hybrid raster dataset, to obtain the rendered scene after the viewpoint update includes:

[0034] Based on the hybrid bit dataset, a hierarchical rendering pipeline is used to divide the scene of the viewpoint after the initial urban environment dynamic model is updated into multiple rendering layers.

[0035] Based on the viewpoint distance and scene complexity after the viewpoint is updated, dynamic lighting and advanced texture mapping techniques are used to perform detailed rendering on each rendering layer to obtain the rendered scene after the viewpoint is updated.

[0036] Calculate the value of the quality evaluation index of the rendered scene after the viewpoint is updated. When the value of the quality evaluation index meets the threshold of the pre-set quality evaluation index, the detailed rendering is completed. Otherwise, adjust the rendering parameters according to the rendered scene after the viewpoint is updated until the calculated value of the quality evaluation index meets the threshold of the pre-set quality evaluation index, and then complete the rendered scene after the viewpoint is updated.

[0037] The quality assessment indicators include one or more of the following: geometric complexity, texture consistency, and lighting smoothness.

[0038] Based on the same inventive concept, this patent application also provides a dynamic urban environment modeling system for an embedded training simulator, including: a current dot matrix data template determination module, a viewpoint rendering module, a dot matrix data dynamic update module, and a data fusion rendering module;

[0039] The current dot matrix data template determination module is used to obtain the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint, and determine the current dot matrix data template based on the spatial coordinates and viewing direction.

[0040] The viewpoint rendering module is used to sample the current dot matrix data template using a lightweight random sampling algorithm to obtain the current dot matrix data, and to perform detailed rendering of the current viewpoint scene of the initial urban environment dynamic model based on the current dot matrix data to obtain the rendering scene of the current viewpoint.

[0041] The dot matrix data dynamic update module is used to determine the dynamically updated dot matrix data based on the dynamically updated spatial coordinates and viewing direction when the spatial coordinates and viewing direction of the viewpoint are randomly and dynamically updated.

[0042] The data fusion rendering module is used to fuse the dynamically updated dot matrix data and the current dot matrix data, and to perform detailed rendering of the scene of the viewpoint after the initial urban environment dynamic model is updated, until the rendering scene of each viewpoint is completed, so as to obtain the final urban environment dynamic model and complete the urban environment dynamic modeling.

[0043] The initial urban environment dynamic model is constructed based on a set of dynamic parameters of the target urban environment collected by a multi-source sensor array and through an embedded training simulator.

[0044] Preferably, the current dot matrix data template determination module includes:

[0045] The viewpoint information acquisition submodule is used to acquire the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint;

[0046] The urban environment dynamic information extraction submodule is used to extract urban environment dynamic information with spatiotemporal continuity from the initial urban environment dynamic model based on the spatial coordinates and viewing direction.

[0047] The LOD level determination submodule is used to determine the corresponding LOD level based on the urban environment dynamic information.

[0048] The dot matrix data template matching submodule is used to search for dot matrix data templates that match the LOD level from a pre-built LOD level library.

[0049] Preferably, the current dot matrix data template determination module further includes: a viewpoint frequency conversion update module, used for:

[0050] Based on the dynamic information of the urban environment, determine the magnitude of change of the current viewpoint;

[0051] When the change amplitude is less than a preset change amplitude threshold, a low-frequency update strategy is adopted for the current viewpoint to perform dynamic updates;

[0052] When the change magnitude exceeds a preset change magnitude threshold, a high-frequency update strategy is adopted to dynamically update the current viewpoint.

[0053] Preferably, the viewpoint rendering module is specifically used for:

[0054] Based on the correlation of the initial urban environment dynamic model in the view frustum space at the current viewpoint, a probability-guided sampling function is established.

[0055] According to the probability-guided sampling function, the current dot matrix data template is sampled to obtain the current dot matrix data. The current dot matrix data is then subjected to visual saliency-driven adaptive filtering to obtain optimized current dot matrix data.

[0056] Extract the visual attention heatmap of the current viewpoint from the optimized current dot matrix data;

[0057] Calculate the heatmap value for each region of interest in the visual attention heatmap;

[0058] For regions of interest whose heatmap values ​​are greater than a preset heatmap threshold, bilateral filtering is used to preserve edge details. For regions of interest whose heatmap values ​​are less than or equal to the heatmap threshold, anisotropic diffusion filtering is used to reduce noise. This completes the detailed rendering of the scene at the current viewpoint, resulting in the rendered scene at the current viewpoint.

[0059] Preferably, the expression for the probability-guided sampling function is as follows:

[0060]

[0061] in, This represents the sampling probability of the i-th dot matrix data in the current dot matrix data template. Distance is the influencing factor. This refers to the i-th dot matrix data in the current dot matrix data template. As the directional influence factor, This refers to the viewpoint of the i-th dot matrix data in the current dot matrix data template. Geometric feature influencing factor Let i be the geometric characteristic function of the i-th dot matrix data in the current dot matrix data template. For color feature influencing factors, The color feature of the i-th dot matrix data in the current dot matrix data template. This is the normalization constant.

[0062] Preferably, the data fusion rendering module includes:

[0063] The hybrid dot matrix dataset generation submodule is used to calculate the spatial correlation matrix between the dynamically updated dot matrix data and the current dot matrix data using a dynamic Kalman filter, and generate a hybrid dot matrix dataset based on the spatial correlation matrix and an adaptive weighted fusion algorithm.

[0064] The new viewpoint rendering scene generation submodule is used to perform detailed rendering of the viewpoint scene after the initial urban environment dynamic model is updated, based on the hybrid bit dataset and using a hierarchical rendering pipeline, to obtain the viewpoint-updated rendering scene.

[0065] The urban environment dynamic model construction submodule is used to complete the rendering scene of each viewpoint, obtain the final urban environment dynamic model, and complete the urban environment dynamic modeling.

[0066] Preferably, the new viewpoint rendering scene generation submodule is specifically used for:

[0067] Based on the hybrid bit dataset, a hierarchical rendering pipeline is used to divide the scene of the viewpoint after the initial urban environment dynamic model is updated into multiple rendering layers.

[0068] Based on the viewpoint distance and scene complexity after the viewpoint is updated, dynamic lighting and advanced texture mapping techniques are used to perform detailed rendering on each rendering layer to obtain the rendered scene after the viewpoint is updated.

[0069] Calculate the value of the quality evaluation index of the rendered scene after the viewpoint is updated. When the value of the quality evaluation index meets the threshold of the pre-set quality evaluation index, the detailed rendering is completed. Otherwise, adjust the rendering parameters according to the rendered scene after the viewpoint is updated until the calculated value of the quality evaluation index meets the threshold of the pre-set quality evaluation index, and then complete the rendered scene after the viewpoint is updated.

[0070] The quality assessment indicators include one or more of the following: geometric complexity, texture consistency, and lighting smoothness.

[0071] Based on the same inventive concept, this patent application also provides an electronic device, comprising: at least one processor and a memory; wherein the memory and the processor are connected via a bus.

[0072] The memory is used to store one or more programs;

[0073] When the one or more programs are executed by the at least one processor, a method for dynamic modeling of urban environments for an embedded training simulator as described above is implemented.

[0074] Based on the same inventive concept, this patent application also provides a readable storage medium on which a computer program is stored, and an executable program is stored thereon. When the executable program is executed, it implements the above-described method for dynamic modeling of urban environment for embedded training simulators.

[0075] Compared with the closest prior art, the beneficial effects of this invention patent application are as follows:

[0076] This invention patent application provides a method and system for dynamic urban environment modeling in an embedded training simulator, comprising: acquiring the spatial coordinates and viewing direction of an initial dynamic urban environment model in the embedded training simulator at the current viewpoint; determining a current point matrix data template based on the spatial coordinates and viewing direction; sampling the current point matrix data template using a lightweight random sampling algorithm to obtain current point matrix data; performing detailed rendering of the scene at the current viewpoint of the initial dynamic urban environment model based on the current point matrix data to obtain a rendered scene at the current viewpoint; determining dynamically updated point matrix data based on the dynamically updated spatial coordinates and viewing direction when the spatial coordinates and viewing direction of the viewpoint undergo random dynamic updates; and performing data processing on the dynamically updated point matrix data and the current point matrix data. The process involves fusing the initial dynamic urban environment model with updated viewpoints and then rendering the scene in detail again until the scene rendering for each viewpoint is complete, resulting in the final dynamic urban environment model and thus completing the dynamic modeling of the urban environment. The initial dynamic urban environment model is constructed using a multi-source sensor array to collect dynamic parameter sets of the target urban environment through an embedded training simulator. This invention can generate real-time point matrix data for the current viewpoint based on its spatial coordinates and viewing direction, meeting the needs of viewpoint changes under the hardware limitations of embedded systems. The dynamic updates of each viewpoint can be determined through the dynamic changes in spatial coordinates and viewing direction. After each dynamic viewpoint update, the scene for that viewpoint is rendered in detail based on the generated point matrix data, achieving adaptive adjustment of the level of detail. Attached Figure Description

[0077] Figure 1 A schematic diagram of a method for dynamic modeling of urban environment for an embedded training simulator provided in this patent application;

[0078] Figure 2 A schematic diagram of a dynamic urban environment modeling system for an embedded training simulator provided in this patent application;

[0079] Figure 3 This is a schematic diagram of the structure of an electronic device provided in this patent application. Detailed Implementation

[0080] The specific embodiments of this patent application will be further described in detail below with reference to the accompanying drawings.

[0081] Example 1:

[0082] This invention patent application provides a method for dynamic modeling of urban environments for embedded training simulators, such as... Figure 1 Shown, including:

[0083] Step 1: Obtain the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint, and determine the current dot matrix data template based on the spatial coordinates and viewing direction;

[0084] Step 2: Using a lightweight random sampling algorithm, sample the current dot matrix data template to obtain the current dot matrix data. Based on the current dot matrix data, perform detailed rendering of the scene at the current viewpoint of the initial urban environment dynamic model to obtain the rendering scene at the current viewpoint.

[0085] Step 3: When the spatial coordinates and viewing direction of the viewpoint are randomly and dynamically updated, determine the dynamically updated dot matrix data based on the dynamically updated spatial coordinates and viewing direction.

[0086] Step 4: Perform data fusion between the dynamically updated dot matrix data and the current dot matrix data, and then perform detailed rendering of the scene of the viewpoint after the initial urban environment dynamic model is updated, until the rendering scene of each viewpoint is completed, to obtain the final urban environment dynamic model and complete the urban environment dynamic modeling.

[0087] The initial urban environment dynamic model is constructed based on a set of dynamic parameters of the target urban environment collected by a multi-source sensor array and through an embedded training simulator.

[0088] For example, the initial urban environment dynamics model integrates dynamic elements such as traffic flow and weather effects that evolve over time.

[0089] In one implementation, step 1 above, which involves obtaining the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint, and determining the current dot matrix data template based on the spatial coordinates and viewing direction, includes:

[0090] Obtain the spatial coordinates and viewing direction of the initial dynamic urban environment model in the embedded training simulator at the current viewpoint;

[0091] Based on the spatial coordinates and viewing direction, extract spatiotemporally continuous urban environmental dynamic information from the initial urban environmental dynamic model;

[0092] Based on the aforementioned urban environmental dynamic information, determine the corresponding Level of Detail (LOD) level.

[0093] From the pre-built LOD level library, find the raster data template that matches the LOD level.

[0094] For example, firstly, the precise spatial coordinates and viewpoint direction of the simulator operator or virtual observation point are captured in real time through the built-in sensors or user interface of the embedded training simulator during the modeling process. The viewpoint direction includes attitude parameters such as yaw, pitch, and roll, thereby locking the current observation focus. Based on the spatial coordinates and changing viewpoint direction, spatiotemporally continuous urban environmental dynamic information is intelligently extracted from the initial urban environmental dynamic model. This urban environmental dynamic information not only includes the instantaneous scene state at the current viewpoint, but also ensures a smooth transition of scene information between adjacent moments through interpolation or prediction algorithms, realistically simulating the dynamic evolution of the urban environment. The urban environmental dynamic information specifically includes comprehensive factors such as dynamic target density, motion complexity, distance from the viewpoint, and relative motion speed. Subsequently, based on the extracted urban environmental dynamic information containing spatiotemporal dynamic characteristics, the optimal level of detail is dynamically evaluated and intelligently determined. This decision-making process goes beyond traditional distance-based LOD selection, creatively incorporating environmental dynamics into the LOD grading factor. For example, even if the target is far away, areas with rapid movement (such as dense traffic) may be assigned a higher LOD level than their static distance to preserve key dynamic details; conversely, distant areas that are static or slow-moving may use a lower LOD to save computational resources. Finally, based on the intelligently determined LOD level, the system accurately matches the corresponding raster data template from a pre-built and optimized LOD level library. This template library is specially designed to store efficient geometric representations of environmental models at different LOD levels, along with necessary texture / material index information. Through this series of tightly integrated steps incorporating dynamic perception and intelligent decision-making, the system can provide the rendering engine with raster data best suited to the current viewpoint (considering spatial location, viewing direction, and scene dynamics) within the limited resource constraints of the embedded platform, significantly improving the real-time performance and immersiveness of the training simulation.

[0095] In one implementation, after extracting spatiotemporally continuous urban environmental dynamic information from the initial urban environmental dynamic model based on the spatial coordinates and viewpoint direction, and before determining the corresponding Level of Detail (LOD) level based on the urban environmental dynamic information, the method further includes:

[0096] Based on the dynamic information of the urban environment, determine the magnitude of change of the current viewpoint;

[0097] When the change amplitude is less than a preset change amplitude threshold, a low-frequency update strategy is adopted for the current viewpoint to perform dynamic updates;

[0098] When the change magnitude exceeds a preset change magnitude threshold, a high-frequency update strategy is adopted to dynamically update the current viewpoint.

[0099] For example, based on extracted, spatiotemporally continuous urban environmental dynamics, particularly the relative relationship between the current viewpoint and its adjacent historical viewpoints, the system calculates and analyzes the magnitude of change at the current viewpoint in real time. The magnitude of change is a multi-dimensional quantitative indicator that comprehensively reflects the intensity of movement across multiple dimensions, including viewpoint spatial displacement, rate of change of viewing direction, and even the frequency of focus switching. Next, the calculated real-time magnitude of change is intelligently compared with a pre-set threshold. This threshold is dynamically configured based on device performance, simulation task requirements, or environmental complexity. A differentiated update strategy is implemented based on the comparison results: when the magnitude of change is determined to be less than the preset threshold, it indicates that the observer during the modeling process is in a relatively static or slowly moving / observing state, such as stationary observation or slow cruising, in which case a low-frequency update strategy is activated. Under this strategy, the system does not completely stop updating; instead, while ensuring the spatiotemporal continuity of the scene, it selectively reduces the sampling frequency of dynamic information, extends the LOD (Level of Detail) evaluation cycle, or reuses some calculated intermediate results, significantly saving computational resources and energy consumption. Conversely, when the change magnitude is determined to be greater than a preset threshold, it indicates that the observer is in a state of rapid movement, frequent turning, or intense manipulation, such as high-speed flight or emergency turns, and a high-frequency update strategy is immediately activated. Under this strategy, the system prioritizes increasing the frequency of dynamic information capture, accelerating the iteration rate of LOD level evaluation, and ensuring the real-time matching and loading of dot matrix data, thereby prioritizing the smoothness of vision under high-speed motion and the real-time presentation of key dynamic details. Ultimately, regardless of the update strategy adopted, the system will execute subsequent steps such as "determining the corresponding LOD level based on the latest valid information" based on the latest valid information. This adaptive update mechanism based on viewpoint motion status enables the system to achieve a dynamic optimal match between computing resources and visual fidelity under the stringent resource constraints of embedded platforms, effectively avoiding extreme situations such as "screen delay / stuttering during intense motion" or "resource waste during static states," providing a smooth, efficient, and highly realistic experience for training simulations.

[0100] In one implementation, step 2 above employs a lightweight random sampling algorithm to sample the current point data template to obtain current point data. Based on the current point data, detailed rendering of the current viewpoint scene of the initial urban environment dynamic model is performed to obtain the current viewpoint rendering scene, including:

[0101] Based on the correlation of the initial urban environment dynamic model in the view frustum space at the current viewpoint, a probability-guided sampling function is established.

[0102] According to the probability-guided sampling function, the current dot matrix data template is sampled to obtain the current dot matrix data. The current dot matrix data is then subjected to visual saliency-driven adaptive filtering to obtain optimized current dot matrix data.

[0103] Extract the visual attention heatmap of the current viewpoint from the optimized current dot matrix data;

[0104] Calculate the heatmap value for each region of interest in the visual attention heatmap;

[0105] For regions of interest whose heatmap values ​​are greater than a preset heatmap threshold, bilateral filtering is used to preserve edge details. For regions of interest whose heatmap values ​​are less than or equal to the heatmap threshold, anisotropic diffusion filtering is used to reduce noise. This completes the detailed rendering of the scene at the current viewpoint, resulting in the rendered scene at the current viewpoint.

[0106] For example, firstly, using the view frustum defined by the spatial coordinates and viewing direction of the current viewpoint, the spatial correlation of the initial urban environment dynamic model within the view frustum is analyzed. Spatial correlation includes object distribution density and geometric complexity gradient, and a probability-guided sampling function is constructed based on this correlation. This probability-guided sampling function is not uniformly random, but dynamically tends to allocate higher sampling probabilities to areas with high spatial correlation, rich potential details, or active movement, thus prioritizing the capture of key data with greater impact on the final perception under a limited number of sampling points. Next, the probability-guided sampling function is applied to perform lightweight random sampling on the current dot matrix data template, efficiently generating the initial current dot matrix dataset. The sampling results are not directly used for rendering, but instead undergo a crucial visual saliency-driven adaptive filtering step: based on the characteristics of the human visual system, including sensitivity to edges, motion, and high-contrast areas, a pre-trained saliency analysis model is designed or invoked to analyze and calculate the initially sampled current dot matrix data, generating a visual attention heatmap characterizing the strength of visual attention in different areas of the scene. This heatmap quantifies the degree to which user attention may be concentrated at the current viewpoint. Based on this heatmap, the system performs intelligent partitioning and adaptive filtering optimization: calculating the aggregated heatmap values, such as average and maximum values, for each semantically or spatially partitioned region of interest in the heatmap. For critical regions of interest with heatmap values ​​higher than a pre-set and configurable heatmap threshold (including near targets, high-speed moving objects, and high-contrast boundaries), the system uses bilateral filtering with strong edge-preserving capabilities. This effectively suppresses noise while accurately preserving crucial geometric edges and texture details, ensuring the clarity and realism of core objects. For non-critical regions with heatmap values ​​equal to or lower than the threshold (including distant backgrounds and large static surfaces), the system uses anisotropic diffusion filtering, which is more computationally efficient and has stronger noise reduction capabilities. This significantly reduces noise interference in these regions and smooths them out with acceptable detail loss, thereby greatly saving the filtering computational resources used for these low-interest regions. Finally, the optimized current pixel data obtained after this visual attention-based partitioning filtering optimization completes the detailed rendering of the current viewpoint scene, outputting a high-quality, low-noise final rendered scene that conforms to human visual perception characteristics. This entire process, from perception-guided sampling to visual saliency-driven intelligent filtering optimization, enables the directional focusing and efficient allocation of embedded rendering resources in key visual areas, significantly improving the subjective visual quality and immersiveness of the output image under the same hardware resources.

[0107] In one implementation, the expression for the probability-guided sampling function is as follows:

[0108]

[0109] in, This represents the sampling probability of the i-th dot matrix data in the current dot matrix data template. Distance is the influencing factor. This refers to the i-th dot matrix data in the current dot matrix data template. As the directional influence factor, This refers to the viewpoint of the i-th dot matrix data in the current dot matrix data template. Geometric feature influencing factor Let i be the geometric characteristic function of the i-th dot matrix data in the current dot matrix data template. For color feature influencing factors, The color feature of the i-th dot matrix data in the current dot matrix data template. This is the normalization constant.

[0110] In one implementation, when the spatial coordinates and viewing direction of the viewpoint are randomly and dynamically updated in step 3 above, the dynamically updated dot matrix data is determined based on the dynamically updated spatial coordinates and viewing direction.

[0111] For example, during the initial training of the urban environment dynamic model, the viewpoint status is continuously monitored. Once a random, non-preset path dynamic update of the viewpoint's spatial coordinates and / or viewing direction is detected, this update event serves as the trigger signal for this step. The system immediately captures and acquires the precise spatial coordinates and viewing direction data after this dynamic update. Subsequently, the system reuses the core processing mechanism of step 1: that is, based on the updated spatial coordinates and viewing direction, it redetermines the corresponding dot matrix data template. This "redetermination" process is essentially the same as step 1, including but not limited to: extracting spatiotemporally continuous environmental information from the initial urban environment dynamic model based on the new viewpoint information, evaluating and determining the most suitable new LOD level accordingly, and finally matching and searching for the dynamically updated dot matrix data template from the pre-built LOD level library. This newly matched template is the basis for generating the dynamically updated dot matrix data. In this way, the system can respond to random dynamic changes in viewpoint in real time, efficiently acquire bitmap data templates that match the new observation perspective, provide necessary input for subsequent sampling and rendering steps, thereby ensuring the spatiotemporal continuity and rendering coherence of the simulated scene during viewpoint switching, avoiding screen breaks or delays, and enhancing the immersive experience of training.

[0112] In one implementation, step 4 above involves fusing the dynamically updated dot matrix data and the current dot matrix data to perform detailed rendering of the viewpoint scene after the initial urban environment dynamic model is updated, until the rendering scene for each viewpoint is completed, thus obtaining the final urban environment dynamic model and completing the urban environment dynamic modeling. This includes:

[0113] A dynamic Kalman filter is used to calculate the spatial correlation matrix between the dynamically updated dot matrix data and the current dot matrix data. Based on the spatial correlation matrix, an adaptive weighted fusion algorithm is used to generate a hybrid dot matrix dataset.

[0114] Based on the hybrid raster dataset, a hierarchical rendering pipeline is used to perform detailed rendering of the scene at the viewpoint after the initial urban environment dynamic model is updated, resulting in the rendered scene after the viewpoint is updated.

[0115] The process continues until the rendering of each viewpoint is completed, resulting in the final dynamic model of the urban environment, thus completing the dynamic modeling of the urban environment.

[0116] For example, a dynamic Kalman filter is used as the core fusion engine. Based on the change vectors of the viewpoint's spatial coordinates and viewing direction, as well as the inherent spatial-temporal attributes of the point data, the dynamic Kalman filter predicts the possible state of the data at the updated viewpoint in real time. Based on this prediction, the filter accurately calculates the spatial correlation matrix between the dynamically updated point data and the current (previous) point data. This spatial correlation matrix deeply quantifies the correspondence between scene elements in 3D space under the old and new viewpoints, the confidence level of relative displacement, and the degree of visibility / occlusion changes caused by the viewpoint change. Using this spatial correlation matrix as the core input, an adaptive weighted fusion algorithm is driven. This algorithm dynamically calculates and assigns optimal fusion weights to each corresponding element or region in the old and new raster data based on the spatial relationship density, motion prediction confidence, and dynamic characteristics of scene elements contained in the spatial correlation matrix. For example, regions with high prediction confidence, strong spatial correlation, and smooth motion are given higher weights in the updated data; regions that may appear due to occlusion or have high prediction uncertainty rely more on the stability of the current data. Finally, based on these dynamically calculated weights, the adaptive weighted fusion algorithm performs weighted fusion of the spatial location, color, normal, and other attributes of overlapping and newly added parts, generating a hybrid raster dataset with significantly improved spatiotemporal continuity, effective noise suppression, and richer details. Subsequently, using an efficient and modular hierarchical rendering pipeline, based on this optimized hybrid raster dataset, the initial urban environment dynamic model is rapidly and with high-quality detail rendering of the scene at the updated viewpoint, outputting the rendered scene after the viewpoint update. This process (detecting viewpoint updates → acquiring new pixel templates → sampling → intelligent fusion of old and new data → rendering new viewpoints) is executed continuously as an iterative unit. Whenever the viewpoint undergoes a new dynamic update due to user interaction or environmental simulation, the system executes a new iteration, iteratively integrating / updating the information of the newly rendered scene into the continuously evolving model representation. This continues until all preset or real-time generated viewpoint sequences have been traversed or covered. Finally, the system integrates the high-precision scene information from all viewpoint rendering outputs to construct a final dynamic urban environment model with high spatiotemporal consistency, complete details, and dynamic realism, completing the entire dynamic urban environment modeling process. This iterative modeling mechanism based on intelligent data fusion and progressive rendering not only effectively overcomes the limitations of single-viewpoint rendering but also, under embedded resource constraints, achieves online, incremental construction and real-time updating of high-fidelity dynamic environment models.

[0117] In one implementation, the step of using a hierarchical rendering pipeline based on the hybrid raster dataset to perform detailed rendering of the scene at the viewpoint after the initial dynamic urban environment model is updated, to obtain the rendered scene after the viewpoint update, includes:

[0118] Based on the hybrid bit dataset, a hierarchical rendering pipeline is used to divide the scene of the viewpoint after the initial urban environment dynamic model is updated into multiple rendering layers.

[0119] Based on the viewpoint distance and scene complexity after the viewpoint is updated, dynamic lighting and advanced texture mapping techniques are used to perform detailed rendering on each rendering layer to obtain the rendered scene after the viewpoint is updated.

[0120] Calculate the value of the quality evaluation index of the rendered scene after the viewpoint is updated. When the value of the quality evaluation index meets the threshold of the pre-set quality evaluation index, the detailed rendering is completed. Otherwise, adjust the rendering parameters according to the rendered scene after the viewpoint is updated until the calculated value of the quality evaluation index meets the threshold of the pre-set quality evaluation index, and then complete the rendered scene after the viewpoint is updated.

[0121] The quality assessment indicators include one or more of the following: geometric complexity, texture consistency, and lighting smoothness.

[0122] For example, based on a hybrid raster dataset, a divide-and-conquer architecture using a hierarchical rendering pipeline intelligently decouples the entire scene to be rendered from the initial dynamic urban environment model at the updated viewpoint into multiple logically or physically independent rendering layers, based on spatial relationships, object types, or material characteristics, such as: foreground dynamic target layer, midground building layer, background environment layer, and effects layer. This layered strategy naturally aligns with the parallel processing potential of embedded platforms, allowing some layers to be rendered asynchronously or in parallel. Next, for each independent rendering layer, the most suitable combination of rendering techniques is dynamically selected and applied, considering the specific spatial distance at the updated viewpoint and the real-time complexity of the scene within that layer: for critical foreground or highly dynamic layers, physically based rendering techniques, dynamic lighting calculations, and high-resolution advanced texture mapping can be finely applied to render details to the fullest extent; for non-critical or low-complexity layers, more efficient simplified lighting and standard texture mapping are used, and even some pre-computed lighting information is reused, significantly saving computational overhead. After each layer rendering is completed, the result is not directly output. Instead, a crucial "perception-optimization" closed loop is executed: one or more core quality evaluation metrics of the rendered scene after the current viewpoint update are calculated in real time. These metrics go beyond traditional frame rate / latency, are deeply related to human visual perception, and include the following:

[0123] The calculated measured values ​​of the indicators are compared with pre-set quality assessment thresholds corresponding to different scene requirements. If all key indicator values ​​meet their corresponding thresholds, the current rendering scene is deemed to meet the quality standards, and the final rendered scene after this round of viewpoint update is output. Conversely, if any key indicator value fails to meet the standards, an adaptive adjustment mechanism for rendering parameters is automatically triggered. This mechanism intelligently decides and dynamically fine-tunes the key rendering parameters affecting that quality dimension based on the specific type and degree of difference of the non-compliant indicator: for example, if texture consistency is insufficient, the texture filtering level of the relevant layer may be increased or the texture sampling rate may be increased; if the lighting smoothness is poor, the shadow map resolution may be increased, soft shadow filtering may be enabled, or ambient light occlusion sampling may be enhanced; if geometric complexity is insufficient, the LOD level or subdivision surface factor of the layer may be temporarily increased within the resource allowance. The system re-executes or incrementally optimizes the rendering process of the affected layer based on the adjusted parameters and performs quality assessment again until all quality assessment indicators meet the preset threshold requirements before outputting the final qualified rendered scene after the viewpoint update. This intelligent mechanism, which integrates layered rendering, perceived quality metrics, and closed-loop self-optimization, ensures that the rendering output after each viewpoint update can accurately achieve the preset visual fidelity target under limited embedded computing power. It avoids wasting resources on repeated calculations caused by over-rendering or insufficient quality, and provides high-quality and highly consistent input for the progressive construction of subsequent models.

[0124] In one implementation, the step of fusing the dynamically updated dot matrix data and the current dot matrix data to perform detailed rendering of the viewpoint scene after the initial urban environment dynamic model is updated, until the rendering scene of each viewpoint is completed, thus obtaining the final urban environment dynamic model, further includes:

[0125] It can receive scene event commands and user input operation commands in real time;

[0126] Based on the scene event command and the operation command, dynamically adjust the spatial coordinates and viewing direction of the current viewpoint in the urban environment dynamic model;

[0127] Based on the adjusted spatial coordinates and viewing direction, the corresponding dynamic scene is extracted and rendered in real time from the urban environment dynamic model.

[0128] The rendered dynamic scene is output to the display terminal of the embedded training simulator for operators to perform training tasks or for viewers to analyze and evaluate.

[0129] The scene event command is generated based on scene events in the target urban environment corresponding to the urban environment dynamic model.

[0130] For example, scenario events include, but are not limited to, historical fire data, historical traffic control data, historical military mission data, and ongoing event data. The urban environment dynamic model is mainly used to perform training tasks and conduct analysis and evaluation based on scenario event instructions.

[0131] When the scene event command is a historical fire data command, historical traffic control data command, or historical military mission data, it executes training requirements in multiple fields such as military, emergency response, and transportation based on the user's input operation command, and presents the training scenario in an immersive way; when the scene event command is generated based on the ongoing event data, it is presented through a dynamic urban environment model to facilitate analysis and evaluation by the viewing personnel.

[0132] Example 2:

[0133] Based on the same inventive concept, this patent application also provides a dynamic urban environment modeling system for embedded training simulators, such as... Figure 2 As shown, it includes: a current dot matrix data template determination module, a viewpoint rendering module, a dot matrix data dynamic update module, and a data fusion rendering module;

[0134] The current dot matrix data template determination module is used to obtain the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint, and determine the current dot matrix data template based on the spatial coordinates and viewing direction.

[0135] The viewpoint rendering module is used to sample the current dot matrix data template using a lightweight random sampling algorithm to obtain the current dot matrix data, and to perform detailed rendering of the current viewpoint scene of the initial urban environment dynamic model based on the current dot matrix data to obtain the rendering scene of the current viewpoint.

[0136] The dot matrix data dynamic update module is used to determine the dynamically updated dot matrix data based on the dynamically updated spatial coordinates and viewing direction when the spatial coordinates and viewing direction of the viewpoint are randomly and dynamically updated.

[0137] The data fusion rendering module is used to fuse the dynamically updated dot matrix data and the current dot matrix data, and to perform detailed rendering of the scene of the viewpoint after the initial urban environment dynamic model is updated, until the rendering scene of each viewpoint is completed, so as to obtain the final urban environment dynamic model and complete the urban environment dynamic modeling.

[0138] The initial urban environment dynamic model is constructed based on a set of dynamic parameters of the target urban environment collected by a multi-source sensor array and through an embedded training simulator.

[0139] Preferably, the current dot matrix data template determination module includes:

[0140] The viewpoint information acquisition submodule is used to acquire the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint;

[0141] The urban environment dynamic information extraction submodule is used to extract urban environment dynamic information with spatiotemporal continuity from the initial urban environment dynamic model based on the spatial coordinates and viewing direction.

[0142] The LOD level determination submodule is used to determine the corresponding LOD level based on the urban environment dynamic information.

[0143] The dot matrix data template matching submodule is used to search for dot matrix data templates that match the LOD level from a pre-built LOD level library.

[0144] Preferably, the current dot matrix data template determination module further includes: a viewpoint frequency conversion update module, used for:

[0145] Based on the dynamic information of the urban environment, determine the magnitude of change of the current viewpoint;

[0146] When the change amplitude is less than a preset change amplitude threshold, a low-frequency update strategy is adopted for the current viewpoint to perform dynamic updates;

[0147] When the change magnitude exceeds a preset change magnitude threshold, a high-frequency update strategy is adopted to dynamically update the current viewpoint.

[0148] Preferably, the viewpoint rendering module is specifically used for:

[0149] Based on the correlation of the initial urban environment dynamic model in the view frustum space at the current viewpoint, a probability-guided sampling function is established.

[0150] According to the probability-guided sampling function, the current dot matrix data template is sampled to obtain the current dot matrix data. The current dot matrix data is then subjected to visual saliency-driven adaptive filtering to obtain optimized current dot matrix data.

[0151] Extract the visual attention heatmap of the current viewpoint from the optimized current dot matrix data;

[0152] Calculate the heatmap value for each region of interest in the visual attention heatmap;

[0153] For regions of interest whose heatmap values ​​are greater than a preset heatmap threshold, bilateral filtering is used to preserve edge details. For regions of interest whose heatmap values ​​are less than or equal to the heatmap threshold, anisotropic diffusion filtering is used to reduce noise. This completes the detailed rendering of the scene at the current viewpoint, resulting in the rendered scene at the current viewpoint.

[0154] Preferably, the expression for the probability-guided sampling function is as follows:

[0155]

[0156] in, This represents the sampling probability of the i-th dot matrix data in the current dot matrix data template. Distance is the influencing factor. This refers to the i-th dot matrix data in the current dot matrix data template. As the directional influence factor, This refers to the viewpoint of the i-th dot matrix data in the current dot matrix data template. Geometric feature influencing factor Let i be the geometric characteristic function of the i-th dot matrix data in the current dot matrix data template. For color feature influencing factors, The color feature of the i-th dot matrix data in the current dot matrix data template. This is the normalization constant.

[0157] Preferably, the data fusion rendering module includes:

[0158] The hybrid dot matrix dataset generation submodule is used to calculate the spatial correlation matrix between the dynamically updated dot matrix data and the current dot matrix data using a dynamic Kalman filter, and generate a hybrid dot matrix dataset based on the spatial correlation matrix and an adaptive weighted fusion algorithm.

[0159] The new viewpoint rendering scene generation submodule is used to perform detailed rendering of the viewpoint scene after the initial urban environment dynamic model is updated, based on the hybrid bit dataset and using a hierarchical rendering pipeline, to obtain the viewpoint-updated rendering scene.

[0160] The urban environment dynamic model construction submodule is used to complete the rendering scene of each viewpoint, obtain the final urban environment dynamic model, and complete the urban environment dynamic modeling.

[0161] Preferably, the new viewpoint rendering scene generation submodule is specifically used for:

[0162] Based on the hybrid bit dataset, a hierarchical rendering pipeline is used to divide the scene of the viewpoint after the initial urban environment dynamic model is updated into multiple rendering layers.

[0163] Based on the viewpoint distance and scene complexity after the viewpoint is updated, dynamic lighting and advanced texture mapping techniques are used to perform detailed rendering on each rendering layer to obtain the rendered scene after the viewpoint is updated.

[0164] Calculate the value of the quality evaluation index of the rendered scene after the viewpoint is updated. When the value of the quality evaluation index meets the threshold of the pre-set quality evaluation index, the detailed rendering is completed. Otherwise, adjust the rendering parameters according to the rendered scene after the viewpoint is updated until the calculated value of the quality evaluation index meets the threshold of the pre-set quality evaluation index, and then complete the rendered scene after the viewpoint is updated.

[0165] The quality assessment indicators include one or more of the following: geometric complexity, texture consistency, and lighting smoothness.

[0166] Example 3

[0167] like Figure 3 As shown, the present invention also provides an electronic device, which may be a computer device, a microcontroller device, a smart mobile device, etc. The electronic device in this embodiment may include a processor, a memory, a transceiver component, etc. The memory, processor, and transceiver component are connected via a bus; the memory can be used to store executable programs, and an exemplary executable program may include instructions; the processor is used to execute the instructions stored in the memory. The memory can also be used to store data, which can be accessed and / or modified when instructions are executed.

[0168] The processor may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and it is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the storage medium to implement the corresponding method flow or corresponding function, so as to implement the steps of the urban environment dynamic modeling method for an embedded training simulator in the above embodiments.

[0169] Example 4

[0170] Based on the same inventive concept, this invention also provides a readable storage medium, specifically an electronic device readable storage medium (Memory). This readable storage medium is a memory device within an electronic device used to store programs and data. It is understood that the storage medium here can include both built-in storage media within the electronic device and extended storage media supported by the electronic device. The storage medium provides storage space, which stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more executable programs (including program code). It should be noted that the storage medium here can be high-speed RAM or non-volatile memory, such as at least one disk storage device. Loading and executing one or more instructions stored in the storage medium by the processor can implement the steps of the urban environment dynamic modeling method for an embedded training simulator described in the above embodiments.

[0171] Those skilled in the art will understand that embodiments of this patent application can be provided as methods, systems, or computer program products. Therefore, this patent application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this patent application can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0172] This patent application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the patent application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0173] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0174] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0175] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this patent application and not to limit its scope of protection. Although the patent application has been described in detail with reference to the above embodiments, those skilled in the art should understand that after reading this patent application, they can still make various changes, modifications or equivalent substitutions to the specific implementation of the application, but these changes, modifications or equivalent substitutions are all within the scope of protection of the claims pending approval.

Claims

1. A method for dynamic modeling of urban environments for embedded training simulators, characterized in that, include: Obtain the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint, and determine the current dot matrix data template based on the spatial coordinates and viewing direction; A lightweight random sampling algorithm is used to sample the current point data template to obtain the current point data. Based on the current point data, the scene of the current viewpoint of the initial urban environment dynamic model is rendered in detail to obtain the rendering scene of the current viewpoint. When the spatial coordinates and viewing direction of the viewpoint are randomly and dynamically updated, the dynamically updated dot matrix data is determined based on the dynamically updated spatial coordinates and viewing direction. The dynamically updated dot matrix data and the current dot matrix data are fused together, and the scene of the viewpoint after the initial urban environment dynamic model is updated is rendered in detail again until the rendering scene of each viewpoint is completed, so as to obtain the final urban environment dynamic model and complete the urban environment dynamic modeling. The initial urban environment dynamic model is constructed based on a set of dynamic parameters of the target urban environment collected by a multi-source sensor array and through an embedded training simulator. The process employs a lightweight random sampling algorithm to sample the current point matrix data template, obtaining current point matrix data. Based on this current point matrix data, detailed rendering of the current viewpoint scene of the initial urban environment dynamic model is performed to obtain the current viewpoint rendering scene, including: Based on the correlation of the initial urban environment dynamic model in the view frustum space at the current viewpoint, a probability-guided sampling function is established. The current dot matrix data is obtained by sampling the current dot matrix data template according to the probability-guided sampling function. The current dot matrix data is analyzed and calculated to generate a visual attention heatmap that represents the strength of visual attention in different areas of the scene. Based on the visual attention heatmap, visual saliency-driven adaptive filtering is performed to obtain optimized current dot matrix data. Extract the visual attention heatmap of the current viewpoint from the optimized current dot matrix data; Calculate the heatmap value for each region of interest in the current viewpoint's visual attention heatmap; For regions of interest whose heatmap values ​​are greater than a preset heatmap threshold, bilateral filtering is used to preserve edge details. For regions of interest whose heatmap values ​​are less than or equal to the heatmap threshold, anisotropic diffusion filtering is used to reduce noise. This completes the detailed rendering of the scene at the current viewpoint, resulting in the rendered scene at the current viewpoint.

2. The method as described in claim 1, characterized in that, The process of obtaining the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint, and determining the current dot matrix data template based on the spatial coordinates and viewing direction, includes: Obtain the spatial coordinates and viewing direction of the initial dynamic urban environment model in the embedded training simulator at the current viewpoint; Based on the spatial coordinates and viewing direction, extract spatiotemporally continuous urban environmental dynamic information from the initial urban environmental dynamic model; Based on the aforementioned urban environmental dynamic information, the corresponding Level of Detail (LOD) level is determined; From the pre-built LOD level library, find the raster data template that matches the LOD level.

3. The method as described in claim 2, characterized in that, After extracting spatiotemporally continuous urban environmental dynamic information from the initial urban environmental dynamic model based on the spatial coordinates and viewpoint direction, and before determining the corresponding Level of Detail (LOD) level based on the urban environmental dynamic information, the method further includes: Based on the dynamic information of the urban environment, determine the magnitude of change of the current viewpoint; When the change amplitude is less than a preset change amplitude threshold, a low-frequency update strategy is adopted for the current viewpoint to perform dynamic updates; When the change magnitude exceeds a preset change magnitude threshold, a high-frequency update strategy is adopted to dynamically update the current viewpoint.

4. The method as described in claim 1, characterized in that, The expression for the probability-guided sampling function is as follows: in, This represents the sampling probability of the i-th dot matrix data in the current dot matrix data template. Distance is the influencing factor. This refers to the i-th dot matrix data in the current dot matrix data template. As the directional influence factor, This refers to the viewpoint of the i-th dot matrix data in the current dot matrix data template. Geometric feature influencing factor Let i be the geometric characteristic function of the i-th dot matrix data in the current dot matrix data template. For color feature influencing factors, The color feature of the i-th dot matrix data in the current dot matrix data template. This is the normalization constant.

5. The method as described in claim 1, characterized in that, The process of fusing the dynamically updated dot matrix data with the current dot matrix data, and then re-rendering the scene of the viewpoint after updating the initial urban environment dynamic model, continues until the rendering scene of each viewpoint is completed, resulting in the final urban environment dynamic model and completing the urban environment dynamic modeling, includes: A dynamic Kalman filter is used to calculate the spatial correlation matrix between the dynamically updated dot matrix data and the current dot matrix data. Based on the spatial correlation matrix, an adaptive weighted fusion algorithm is used to generate a hybrid dot matrix dataset. Based on the hybrid raster dataset, a hierarchical rendering pipeline is used to perform detailed rendering of the scene at the viewpoint after the initial urban environment dynamic model is updated, resulting in the rendered scene after the viewpoint is updated. The process continues until the rendering of each viewpoint is completed, resulting in the final dynamic model of the urban environment, thus completing the dynamic modeling of the urban environment.

6. The method as described in claim 5, characterized in that, Based on the hybrid raster dataset, a hierarchical rendering pipeline is used to perform detailed rendering of the scene at the updated viewpoint after the initial dynamic urban environment model is updated, resulting in the following rendered scene: Based on the hybrid bit dataset, a hierarchical rendering pipeline is used to divide the scene of the viewpoint after the initial urban environment dynamic model is updated into multiple rendering layers. Based on the viewpoint distance and scene complexity after the viewpoint is updated, dynamic lighting and advanced texture mapping techniques are used to perform detailed rendering on each rendering layer to obtain the rendered scene after the viewpoint is updated. Calculate the value of the quality evaluation index of the rendered scene after the viewpoint is updated. When the value of the quality evaluation index meets the threshold of the pre-set quality evaluation index, the detailed rendering is completed. Otherwise, adjust the rendering parameters according to the rendered scene after the viewpoint is updated until the calculated value of the quality evaluation index meets the threshold of the pre-set quality evaluation index, and then complete the rendered scene after the viewpoint is updated. The quality assessment indicators include one or more of the following: geometric complexity, texture consistency, and lighting smoothness.

7. A dynamic urban environment modeling system for an embedded training simulator, characterized in that, include: The module includes: current dot matrix data template determination module, viewpoint rendering module, dot matrix data dynamic update module, and data fusion rendering module. The current dot matrix data template determination module is used to obtain the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint, and determine the current dot matrix data template based on the spatial coordinates and viewing direction. The viewpoint rendering module is used to sample the current dot matrix data template using a lightweight random sampling algorithm to obtain the current dot matrix data, and to perform detailed rendering of the current viewpoint scene of the initial urban environment dynamic model based on the current dot matrix data to obtain the rendering scene of the current viewpoint. The dot matrix data dynamic update module is used to determine the dynamically updated dot matrix data based on the dynamically updated spatial coordinates and viewing direction when the spatial coordinates and viewing direction of the viewpoint are randomly and dynamically updated. The data fusion rendering module is used to fuse the dynamically updated dot matrix data and the current dot matrix data, and to perform detailed rendering of the scene of the viewpoint after the initial urban environment dynamic model is updated, until the rendering scene of each viewpoint is completed, so as to obtain the final urban environment dynamic model and complete the urban environment dynamic modeling. The initial urban environment dynamic model is constructed based on a set of dynamic parameters of the target urban environment collected by a multi-source sensor array and through an embedded training simulator. The viewpoint rendering module is specifically used for: Based on the correlation of the initial urban environment dynamic model in the view frustum space at the current viewpoint, a probability-guided sampling function is established. The current dot matrix data is obtained by sampling the current dot matrix data template according to the probability-guided sampling function. The current dot matrix data is analyzed and calculated to generate a visual attention heatmap that represents the strength of visual attention in different areas of the scene. Based on the visual attention heatmap, visual saliency-driven adaptive filtering is performed to obtain optimized current dot matrix data. Extract the visual attention heatmap of the current viewpoint from the optimized current dot matrix data; Calculate the heatmap value for each region of interest in the current viewpoint's visual attention heatmap; For regions of interest whose heatmap values ​​are greater than a preset heatmap threshold, bilateral filtering is used to preserve edge details. For regions of interest whose heatmap values ​​are less than or equal to the heatmap threshold, anisotropic diffusion filtering is used to reduce noise. This completes the detailed rendering of the scene at the current viewpoint, resulting in the rendered scene at the current viewpoint.

8. The system as described in claim 7, characterized in that, The current dot matrix data template determination module includes: The viewpoint information acquisition submodule is used to acquire the spatial coordinates and viewing direction of the initial urban environment dynamic model in the embedded training simulator at the current viewpoint; The urban environment dynamic information extraction submodule is used to extract urban environment dynamic information with spatiotemporal continuity from the initial urban environment dynamic model based on the spatial coordinates and viewing direction. The LOD level determination submodule is used to determine the corresponding LOD level based on the urban environment dynamic information. The dot matrix data template matching submodule is used to search for dot matrix data templates that match the LOD level from a pre-built LOD level library.

9. The system as described in claim 8, characterized in that, The current dot matrix data template determination module further includes: a viewpoint frequency conversion update module, used for: Based on the dynamic information of the urban environment, determine the magnitude of change of the current viewpoint; When the change amplitude is less than a preset change amplitude threshold, a low-frequency update strategy is adopted for the current viewpoint to perform dynamic updates; When the change magnitude exceeds a preset change magnitude threshold, a high-frequency update strategy is adopted to dynamically update the current viewpoint.

Images