Virtual reality urban environment design system based on AR technology

Abstract

The invention discloses a virtual reality urban environment design system based on an AR technology, and the system comprises seven core units: an urban geographic information multi-dimensional data fusion analysis unit, a hidden Markov dynamic scene deduction simulation unit and the like. Through an improved hidden Markov-attention fusion model, an attention model based on a hyperbolic tangent function and the like, deep fusion analysis and dynamic scene deduction of multi-source city geographic information are realized; adaptive virtual elements are generated by using generative adversarial network variants, AR interaction rules are formulated by means of reinforcement learning, real-time rendering is optimized in combination with a hierarchical rendering algorithm, and stable and accurate operation of the system is guaranteed based on an error correction feedback control model and a time sequence dynamic optimization mechanism. According to the system, the problems of poor data fusion, inaccurate scene deduction and the like of a traditional design system are solved, and efficient, accurate and immersive virtual reality urban environment design is realized.

Description

Technical Field

[0001] The present invention relates to the field of virtual reality urban environment design, and in particular to a virtual reality urban environment design system based on AR technology. Background Art

[0002] With the acceleration of urbanization, the application of virtual reality technology in urban environmental design is becoming increasingly important. AR technology, with its unique blend of virtual and real-world capabilities, offers new insights and directions for urban environmental design. It allows designers and users to more intuitively experience urban environmental design solutions, improving design efficiency and quality. However, existing virtual reality urban environmental design systems still face numerous challenges and struggle to meet the increasingly complex demands of urban design.

[0003] Existing technologies for processing urban geographic information struggle to efficiently integrate heterogeneous data from multiple sources, resulting in poor data correlation. This results in inaccurate and incomplete comprehensive analysis of urban geographic information, hindering the provision of precise data support for design. Traditional models struggle to effectively simulate the random changes and uncertainties in the urban environment, resulting in inaccurate and inaccurate predictions of future urban environments. Furthermore, feature processing fails to highlight key information based on scenario importance and relevance, leading to a lack of understanding of key elements in design.

[0004] When it comes to virtual element generation and interaction design, existing systems produce virtual elements that are poorly adapted to urban environments, lacking a sense of realism and immersion. AR interaction rules are simplistic, failing to provide personalized interactive experiences tailored to specific urban environments and virtual elements. Furthermore, balancing real-time rendering efficiency and quality is difficult, making it difficult to achieve smooth rendering across different devices. Furthermore, the system lacks an effective feedback and calibration mechanism, making it unable to adapt to the dynamic changes in the urban environment and the diverse needs of users. These factors have limited the further development and application of virtual reality urban environment design systems. Summary of the Invention

[0005] In order to overcome the shortcomings and deficiencies of the existing technology, the present invention provides a virtual reality urban environment design system based on AR technology.

[0006] The technical solution adopted by the present invention is a virtual reality urban environment design system based on AR technology, which includes:

[0007] Urban geographic information multi-dimensional data fusion and analysis unit: This unit is connected to the urban environment dynamic data acquisition module, receives multi-source heterogeneous urban geographic information data from the module, including topography, building distribution, and transportation network data, and uses data fusion algorithms to fuse and analyze these data to form a comprehensive urban geographic information data set;

[0008] Hidden Markov dynamic scene simulation unit: This unit is connected to the urban geographic information multi-dimensional data fusion and analysis unit, obtains the comprehensive data set output by it, and simulates the dynamic scenes in the urban environment based on the hidden Markov model to generate dynamic scene data of the future urban environment;

[0009] Attention mechanism feature weight allocation and control unit: This unit is connected to the hidden Markov dynamic scene deduction and simulation unit, extracts features from the dynamic scene data obtained by deduction and simulation, uses the attention mechanism to assign corresponding weights to different features, and dynamically adjusts the weights according to the importance and relevance of the scene;

[0010] Virtual element intelligent generation and adaptation unit: This unit is connected to the attention mechanism feature weight allocation and control unit. Based on the weighted feature information, it intelligently generates various virtual elements suitable for the virtual reality city environment, including virtual buildings and virtual landscapes, and adapts these virtual elements to integrate them with the urban geographical environment and dynamic scenes.

[0011] AR interaction rule strategy formulation unit: This unit is connected to the virtual element intelligent generation and adaptation unit. It formulates AR interaction rules and strategies based on the generated virtual elements and the characteristics of the urban environment, including the interaction methods and interaction conditions between users and virtual elements.

[0012] Real-time rendering optimization processing unit: This unit is connected to the AR interaction rule strategy formulation unit, receives the formulated interaction rules and virtual element data, and uses real-time rendering algorithms to render the virtual reality city environment and perform optimization processing;

[0013] System integrated feedback calibration unit: This unit is connected to the real-time rendering optimization processing unit to perform integrated management of the operation of the entire system, receive user feedback information and system operation data, and calibrate and adjust each unit in the system.

[0014] Furthermore, the hidden Markov dynamic scene deduction simulation unit adopts an improved hidden Markov-attention fusion model when performing dynamic scene deduction simulation. The model formula is:

[0015]

[0016] Among them, P(S t+1 ∣S t , F t ) represents the state S at time t t and feature F t Next, the state S at time t+1 t+1 The probability of occurrence, P(S t+1 ∣S t) is the state transition probability in the traditional hidden Markov model, which represents the state S at time t t Transition to state S at time t+1 t+1 The probability of F t is the eigenvector at time t, denoted as F t =(F t，1 , F t，2 ,…,F t，n ), where the characteristics are the building density in the urban environment, the population flow speed, the virtual reality urban environment design parameters, Att(F t，i ) is the attention mechanism for feature F t，i The calculated attention weight reflects the importance of the feature in the current scene deduction, α i It is an adjustment coefficient used to balance the influence of the state transition probability and attention weight of the traditional hidden Markov model.

[0017] Furthermore, the attention mechanism feature weight allocation control unit adopts an attention model based on the hyperbolic tangent function when calculating the attention weight. The model formula is:

[0018]

[0019] Among them, W is a learnable weight matrix used to adjust the feature F t，i A linear transformation is performed, and its dimension is determined according to the dimension of the feature vector and the model design. b is a bias vector, which is a learnable parameter used to adjust the result of the linear transformation. Tanh is a hyperbolic tangent function, which maps the linear transformation result of the input to the interval (-1, 1), so that the attention weight has certain nonlinear characteristics.

[0020] Furthermore, when generating virtual elements, the virtual element intelligent generation and adaptation unit combines the output results of the improved hidden Markov-attention fusion model and the attention model based on the hyperbolic tangent function. The unit first screens out key feature data from the comprehensive data set output by the urban geographic information multidimensional data fusion and analysis unit based on the feature weight distribution control unit of the attention mechanism. Then, using these key feature data and the dynamic scene information deduced by the hidden Markov dynamic scene deduction and simulation unit, a variant model of the generative adversarial network is used to generate virtual elements. The training objective functions of the generator G and the discriminator D of the variant model are as follows:

[0021]

[0022] Among them, x is the data distribution p from the real urban environment data The samples of (x) include real building models, landscape images, and z are obtained from the noise distribution p zThe random noise vector sampled from (z), F w It is a feature vector weighted by the attention mechanism, which contains the key feature information in the design of virtual reality urban environment, including architectural style, landscape type, G(z, F w ) is the generator based on random noise z and weighted features F w The generated virtual element, D(x) is the discriminator's judgment result on the real sample x, D(G(z, F w )) is the virtual element G(z, F) generated by the discriminator w )’s judgment result.

[0023] Furthermore, when formulating interaction rules and strategies, the AR interaction rule strategy formulation unit considers the dynamic scene deduced by the hidden Markov dynamic scene deduction simulation unit and the feature weights assigned by the attention mechanism feature weight distribution control unit. The unit constructs a state-action-reward based reinforcement learning model, in which the state S is composed of the dynamic scene information of the urban environment and the feature information of the virtual element, the action A is the interaction method between the user and the virtual element, and the reward R is set according to the interaction effect and the scene goal. The value function update formula of the reinforcement learning model is:

[0024]

[0025] Among them, Q(S t , A t ) indicates that in state S t Next, take action A t The value estimate of , α is the learning rate, which controls the step size of each update, γ is the discount factor, which is used to balance the importance of immediate rewards and future rewards, R t+1 In state S t Take action A t The immediate reward after S t+1 Is to perform action A t Then transfer to the next state.

[0026] Furthermore, when performing rendering optimization, the real-time rendering optimization processing unit combines the outputs of the hidden Markov dynamic scene deduction simulation unit and the attention mechanism feature weight allocation control unit. The unit adopts a hierarchical rendering algorithm based on feature importance to divide the elements in the virtual reality city environment into different levels according to their feature importance. For elements with higher feature importance, a high-precision rendering algorithm is used for rendering, and for elements with lower feature importance, a low-precision rendering algorithm is used for fast rendering. The hierarchical rule of the algorithm is expressed by the following formula:

[0027]

[0028] Among them, Li Indicates the rendering level of element i, H represents the high-precision rendering level, L represents the low-precision rendering level, Att(F i ) is the feature F of element i by the attention mechanism i The calculated attention weight,θ, is a pre-set threshold used to divide the rendering levels.

[0029] Furthermore, the system integrated feedback calibration unit adopts a feedback control model based on error correction when performing system calibration and adjustment. The model calculates the output error of each unit based on user feedback information and system operation data. For the hidden Markov dynamic scene deduction simulation unit, the error calculation formula is:

[0030]

[0031] Among them, E HMM is the error of the hidden Markov dynamic scene simulation unit, P sim (S i ) is the state S obtained by the unit simulation i The probability of occurrence, P real (S i ) is the actual observed state S i The probability of occurrence, m is the number of states, and for the attention mechanism feature weight distribution control unit, the error calculation formula is:

[0032]

[0033] Among them, E Att is the error of the attention mechanism feature weight distribution control unit, Att sim (F j ) is the characteristic F calculated by the unit j The attention weight, Att real (F j ) is the feature F obtained based on actual evaluation j The attention weight is , and n is the number of features.

[0034] Furthermore, when performing data fusion analysis, the urban geographic information multidimensional data fusion analysis unit combines the information of the hidden Markov dynamic scene deduction simulation unit and the attention mechanism feature weight distribution control unit. The unit adopts a multi-source data fusion model based on the Bayesian network to fuse urban geographic information data from different sources. The joint probability distribution formula of the Bayesian network is:

[0035]

[0036] Among them, X1, X2, ..., X kIt is the various variables in urban geographic information data, including terrain height, building area, Pa(X i ) is the variable X i The parent node set of X i A set of variables that have a direct causal relationship.

[0037] Furthermore, during the overall operation of the system, a dynamic optimization mechanism based on time series is adopted. This mechanism combines the output of the hidden Markov dynamic scene deduction simulation unit and the attention mechanism feature weight distribution control unit to dynamically adjust each unit in the system. At each time step t, the system adjusts the current state S according to the current state S. t and feature F t , calculate the adjustment parameters of each unit, and for the hidden Markov dynamic scene simulation unit, adjust its state transition probability matrix A t The formula is:

[0038] A t+1 =A t +β·ΔA t

[0039] Among them, A t+1 is the state transition probability matrix for the next time step, A t is the state transition probability matrix of the current time step, β is the adjustment coefficient, which controls the amplitude of the adjustment, ΔA t It is the adjustment amount of the state transition probability matrix calculated based on the current state and features;

[0040] For the attention mechanism feature weight distribution control unit, adjust its weight matrix W t The formula is:

[0041] W t+1 =W t +γ·ΔW t

[0042] Among them, W t+1 is the weight matrix for the next time step, W t is the weight matrix of the current time step, γ is the adjustment coefficient, ΔW t It is the adjustment amount of the weight matrix calculated based on the current state and features.

[0043] A virtual reality urban environment design system based on AR technology, the system operation includes:

[0044] Multi-source geographic data access and integration: Urban geographic information data from different data sources, including satellite remote sensing image data, ground surveying and mapping data, and urban planning data, are accessed and initially integrated to form a relevant raw data set that includes basic information on the city's topography, landforms, and building distribution.

[0045] Preliminary deduction of hidden Markov scenarios: Using the hidden Markov model to process the initially integrated raw data set, taking into account various uncertainties and random changes in the urban environment, preliminary deduction of the dynamic scenarios of the urban environment is carried out to generate initial data for future dynamic scenarios of the urban environment;

[0046] Attention feature screening and extraction: The attention mechanism is used to screen and extract features from the initial dynamic scene data obtained through preliminary deduction. Different features are assigned corresponding weights based on the importance and relevance of the scene, highlighting the key feature information and forming a feature-weighted key data set.

[0047] Virtual element generation and adaptation: Based on a weighted feature set of key data, a variant model of a generative adversarial network is used to intelligently generate various virtual elements suitable for the VR city environment, including virtual buildings and virtual landscapes. These generated virtual elements are then adapted to blend in with the city's geographical environment and dynamic scenes in terms of color, shape, and size.

[0048] Dynamic formulation of AR interaction rules: Based on the generated and adapted virtual elements and urban environment characteristics, a state-action-reward-based reinforcement learning model is constructed. Through continuous training with user interactions, AR interaction rules and strategies are dynamically formulated, including the interaction methods and conditions between users and virtual elements.

[0049] Real-time rendering optimization: Combining the results of Hidden Markov dynamic scene deduction and attention mechanism feature weight allocation, a layered rendering algorithm based on feature importance is used to render the virtual reality city environment, and multi-threaded rendering technology and hardware acceleration technology are used for optimization.

[0050] System feedback calibration loop adjustment: collect user feedback information and system operation data, calculate the output error of each unit, and use the feedback control model based on error correction to adjust the parameters of each unit in the system, including adjusting the state transition probability matrix of the hidden Markov model and the weight matrix of the attention mechanism. Then, re-enter the next round of operation process, forming a cyclic adjustment process.

[0051] Beneficial Effects: This invention proposes a virtual reality urban environment design system based on AR technology. Regarding data processing, the system's urban geographic information multidimensional data fusion and analysis unit is capable of deeply integrating multi-source, heterogeneous urban geographic information data, providing an accurate and closely related comprehensive data set for subsequent processes. The hidden Markov dynamic scene simulation unit, combined with an improved hidden Markov-attention fusion model, fully considers multiple uncertainties and accurately deduces dynamic urban environment scenarios, such as predicting building construction trends and population mobility changes. The attention mechanism feature weight allocation and control unit, using an attention model based on the hyperbolic tangent function, flexibly adjusts feature weights, highlights key features, and provides targeted data for virtual element generation. The virtual element intelligent generation and adaptation unit combines virtual elements generated by multiple models, highly integrating them with the urban environment and enhancing the realism of the virtual scene. The AR interaction rule strategy formulation unit uses a reinforcement learning model to formulate interaction rules, providing personalized interactive experiences based on different scenarios and element features. The real-time rendering optimization processing unit's layered rendering algorithm, combined with multi-threading and hardware acceleration technology, effectively improves rendering efficiency and quality, ensuring smooth rendering on different devices. The system's integrated feedback calibration unit continuously adjusts each unit's parameters through an error-correction feedback control model to ensure system stability and accuracy. The Bayesian network multi-source data fusion model of the urban geographic information multidimensional data fusion and analysis unit combines dynamic scenarios and feature weight information to make the fused data more realistic. A time-series-based dynamic optimization mechanism adapts to changes in the urban environment in real time and continuously optimizes system performance. The entire system, from data processing, scenario simulation, element generation, to interactive experience and system optimization, comprehensively improves the quality and effectiveness of virtual reality urban environment design. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 This is a diagram of the system unit composition of the present invention;

[0053] Figure 2 This is a flow chart of the operating steps of the system of the present invention. DETAILED DESCRIPTION

[0054] It should be noted that, unless there is a conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The application is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0055] like Figure 1 As shown, a virtual reality urban environment design system based on AR technology is characterized by including:

[0056] Urban geographic information multi-dimensional data fusion and analysis unit: This unit is connected to the urban environment dynamic data acquisition module, and receives multi-source heterogeneous urban geographic information data from the module, including topography, building distribution, transportation network and other data, and uses complex data fusion algorithms to deeply fuse and analyze these data to form a comprehensive urban geographic information data set with high relevance and accuracy.

[0057] Specifically, this unit is primarily responsible for receiving multi-source, heterogeneous data from the Urban Environment Dynamic Data Acquisition Module. This data covers multiple aspects of urban geographic information, including topographic data, with centimeter-level accuracy, detailing the city's elevation; building distribution data, including location, height, floor area, and other information, accurate to the specific latitude and longitude coordinates and number of floors; and transportation network data, such as road width, number of lanes, and traffic flow. This unit utilizes complex data fusion algorithms to integrate this data from various formats and sources. For example, fusion algorithms based on feature matching and data association are employed to ensure the relevance and accuracy of the different data.

[0058] The significance of this unit lies in providing comprehensive and accurate foundational data for subsequent urban environmental design. The integration of multi-source, heterogeneous data eliminates redundancy and conflict, enabling designers to obtain more complete and authentic urban geographic information. This helps to improve the scientific and rational nature of urban environmental design and avoid design errors caused by inaccurate or incomplete data.

[0059] For example, when designing a new urban area, this unit can integrate satellite remote sensing imagery, ground-based mapping data, and data from urban planning databases. By analyzing this integrated data, designers can gain a clear understanding of the area's topography, existing building distribution, and transportation network, enabling them to rationally plan new building layouts, road alignments, and the location of public facilities. For example, topographic data can help avoid building major structures in low-lying, flood-prone areas, while traffic flow data can optimize road design and the layout of transportation hubs.

[0060] Hidden Markov dynamic scenario deduction and simulation unit: This unit is connected to the urban geographic information multi-dimensional data fusion and analysis unit, obtains its output comprehensive data set, and deduces and simulates the dynamic scenes in the urban environment based on the hidden Markov model, taking into account various uncertain factors and random changes, and generates a series of possible future urban environment dynamic scenario data.

[0061] Specifically, this unit is connected to the Urban Geographic Information Multidimensional Data Fusion and Analysis Unit, receiving its output comprehensive data sets. Dynamic scenario simulations are performed based on a hidden Markov model. In this model, state transition probability is a key parameter that describes the likelihood of an urban environment shifting from one state to another. For example, the probability of a certain area in a city being converted from commercial to residential, or the probability of traffic flow on a certain road shifting from peak to trough, can be determined with high accuracy through analysis of historical data and training with machine learning algorithms.

[0062] By simulating dynamic urban environments, designers can predict future development trends and potential problems. This helps factor in various factors during the design phase, leading to more forward-looking and adaptable urban design solutions. For example, if population growth in a particular area is predicted to lead to traffic congestion, additional transportation facilities or optimized routes can be planned in advance.

[0063] For example, let's say you're designing a city's commercial district. This unit could simulate its development over the next few years based on historical data and current urban development trends. It might predict that with the growth of surrounding residential areas, foot traffic in the commercial district will increase significantly, leading to greater commercial activity. Based on this prediction, the designer can appropriately increase parking spaces, public rest areas, and commercial facilities in the design to meet future demand.

[0064] Attention mechanism feature weight allocation and control unit: This unit is connected to the hidden Markov dynamic scene deduction and simulation unit, extracts features from the dynamic scene data obtained by the deduction and simulation, uses the attention mechanism to assign corresponding weights to different features, and dynamically adjusts the weights according to the importance and relevance of the scene to highlight key feature information.

[0065] Specifically, this unit extracts features from the dynamic scene data obtained by the Hidden Markov Dynamic Scene Simulation Unit. Features include various factors in the urban environment, such as the appearance of buildings, green coverage, noise levels, etc. The attention mechanism highlights key features by calculating the importance weight of each feature. The weight calculation is based on the relevance and importance of the feature to the current scene. For example, when designing a virtual reality environment for a tourist attraction, the weight of the landscape's aesthetics and uniqueness may be relatively high, while some minor infrastructure features may have lower weights.

[0066] By properly assigning feature weights, this unit can help designers focus on key factors in the urban environment, improving design efficiency and quality. It can also help prevent excessive distractions from secondary information during the design process, allowing design solutions to highlight key points and meet design goals.

[0067] For example, when designing a city's cultural plaza, this unit can assign weights to different features based on the attention mechanism. Features related to the plaza's cultural atmosphere, such as cultural sculptures and displays of historical relics, are given higher weights, while features of ordinary commercial shops surrounding the plaza are given lower weights. This allows designers to prioritize the integration of cultural elements during the design process, creating a plaza with a rich cultural atmosphere.

[0068] Virtual element intelligent generation and adaptation unit: It is connected to the attention mechanism feature weight distribution control unit. Based on the feature information after weight distribution, it intelligently generates various virtual elements suitable for the virtual reality city environment, such as virtual buildings, virtual landscapes, etc., and adapts and adjusts these virtual elements to make them integrate with the urban geographical environment and dynamic scenes.

[0069] Specifically, this unit is connected to the attention mechanism feature weight allocation control unit and uses a variant model of the Generative Adversarial Network (GAN) to generate virtual elements based on the feature information after weight allocation. The generated virtual elements include virtual buildings, virtual landscapes, etc. During the generation process, the model adjusts the attributes of the virtual elements based on the input feature information, such as the style, color, and shape of the building, and the plant species and layout of the landscape. For example, when generating a virtual building with a European style, the model will design the building's appearance based on the characteristics of European architecture, such as spires, arches, and reliefs.

[0070] This unit can quickly and intelligently generate virtual elements suitable for VR urban environments, integrating them with the city's geographic context and dynamic scenes. This significantly improves the efficiency of urban environment design and reduces the workload of designers in manually creating virtual elements. Furthermore, the generated virtual elements exhibit a high degree of realism and adaptability, enhancing the user's sense of immersion.

[0071] For example, when designing a virtual reality city park, the unit can generate various virtual landscape elements, such as lakes, hills, flowers, and trees, based on the park's location, surrounding environment, and design objectives. These virtual elements will be adapted to the park's actual terrain and style, making the virtual park more realistic and beautiful. For example, if the park is located in a mountainous area, the generated landscape elements will be more like a natural forest, with more trees and undulating terrain.

[0072] AR interaction rule strategy formulation unit: This unit is connected to the virtual element intelligent generation and adaptation unit. Based on the generated virtual elements and the characteristics of the urban environment, it formulates AR interaction rules and strategies, including the interaction methods and interaction conditions between users and virtual elements, to achieve a natural and smooth AR interaction experience.

[0073] Specifically, this unit is connected to the virtual element intelligent generation and adaptation unit. Based on the generated virtual elements and the characteristics of the urban environment, a state-action-reward reinforcement learning model is constructed to formulate AR interaction rules and strategies. The state includes the dynamic scene information of the urban environment and the characteristic information of the virtual elements. The action is the way the user interacts with the virtual elements, such as clicking, touching, and moving. The reward is set based on the interaction effect and the scene goal. For example, in a virtual city shopping scene, if a user clicks on a virtual store and successfully purchases an item, the system will provide a certain reward to encourage more user interaction.

[0074] Reasonable AR interaction rules and strategies can enhance the user experience with the VR city environment, increasing their sense of engagement and immersion. By continuously optimizing interaction rules through reinforcement learning models, we can provide personalized interaction experiences based on different user behaviors and scenario changes, meeting users' diverse needs.

[0075] For example, in a VR city tour guide system, this unit can create different interaction rules. When a user approaches a historical building, the system automatically displays a historical introduction and related images. If the user observes the building in detail and asks questions, the system provides more in-depth answers based on the user's questions. This personalized interaction allows users to better understand the city's history and culture, making the tour more interesting and educational.

[0076] Real-time rendering optimization processing unit: This unit is connected to the AR interaction rule strategy formulation unit, receives the formulated interaction rules and virtual element data, uses advanced real-time rendering algorithms to render the virtual reality city environment, and performs optimization processing to improve rendering efficiency and quality, ensuring smooth rendering effects on different devices.

[0077] Specifically, this unit connects to the AR interaction rule strategy formulation unit and receives the formulated interaction rules and virtual element data. A layered rendering algorithm based on feature importance is used for rendering optimization. Elements in the VR city environment are divided into different layers based on their feature importance. High-importance elements are rendered using a high-precision rendering algorithm, such as one with more texture detail and lighting effects. Low-importance elements are rendered using a low-precision rendering algorithm for faster rendering. Furthermore, multi-threaded rendering technology and hardware acceleration are combined to improve rendering efficiency.

[0078] The real-time rendering optimization processing unit improves rendering efficiency while maintaining rendering quality, ensuring smooth rendering across different devices. This is particularly true when processing large-scale VR city environment data, effectively reducing rendering time and resource consumption, allowing users to experience VR city environments more smoothly.

[0079] For example, in a large-scale VR city plan display, key elements such as major buildings and landmarks are rendered with high precision, showcasing a realistic appearance and detail. Meanwhile, more distant secondary buildings and background elements are rendered with low precision. This ensures the visual quality of key elements while increasing overall rendering speed, preventing users from experiencing lag when viewing the city plan.

[0080] System integrated feedback calibration unit: This unit is connected to the real-time rendering optimization processing unit to perform integrated management of the operation of the entire system, receive user feedback information and system operation data, and calibrate and adjust each unit in the system to ensure system stability and accuracy.

[0081] Specifically, this unit is connected to the real-time rendering optimization processing unit to provide integrated management of the entire system's operations. It receives user feedback and system operation data and calculates the output errors of each unit. For example, for the Hidden Markov Dynamic Scenario Simulation Unit, the error is calculated by comparing the deduced results with the observed urban environment. For the Attention Mechanism Feature Weight Allocation Control Unit, the error is calculated by comparing the calculated feature weights with the actual evaluated weights. Based on this error information, the parameters of each unit in the system are adjusted.

[0082] The system's integrated feedback calibration unit ensures stability and accuracy. By continuously receiving feedback and adjusting parameters, the system adapts to the dynamic changes in the urban environment and the diverse needs of users. If errors or incompatibility with new conditions occur, timely calibration and optimization can be performed to improve system performance and reliability.

[0083] For example, in a long-running virtual reality urban environment design system, as the city develops and changes, the actual urban environment may deviate from the system's simulated scenarios. The system's integrated feedback calibration unit can adjust the state transition probability matrix of the Hidden Markov Model dynamic scenario simulation unit based on user feedback and actual data, making the simulated results more consistent with reality. Simultaneously, the weight matrix of the attention mechanism feature weight allocation control unit is adjusted to better highlight key features in the current urban environment.

[0084] Preferably, the hidden Markov dynamic scene deduction simulation unit adopts an improved hidden Markov-attention fusion model when performing dynamic scene deduction simulation. The model formula is:

[0085]

[0086] Among them, P(S t+1 ∣St , F t ) represents the state S at time t t and feature F t Next, the state S at time t+1 t+1 The probability of occurrence. P(S t+1 ∣S t ) is the state transition probability in the traditional hidden Markov model, which represents the state S at time t t Transition to state S at time t+1 t+1 The probability of F t is the eigenvector at time t, which can be expressed as F t =(F t，1 , F t，2 ,…,F t，n ), where the features can be the building density, population flow speed and other virtual reality urban environment design parameters in the urban environment. t，i ) is the attention mechanism for feature F t，i The calculated attention weight reflects the importance of the feature in the current scene deduction. i It is an adjustment coefficient used to balance the influence of the state transition probability and attention weight of the traditional hidden Markov model.

[0087] This unit uses this model in combination with the comprehensive data set output by the urban geographic information multidimensional data fusion and analysis unit, taking into account the importance of different features, to more accurately deduce dynamic scenarios in the urban environment, such as predicting future building construction trends in a certain area of ​​the city, changes in population mobility, and other scenarios.

[0088] Preferably, the attention mechanism feature weight allocation control unit adopts an attention model based on the hyperbolic tangent function when calculating the attention weight. The model formula is:

[0089]

[0090] Among them, W is a learnable weight matrix used to adjust the feature F t，i A linear transformation is performed, whose dimension is determined by the dimension of the feature vector and the model design. b is a bias vector, also a learnable parameter, used to adjust the result of the linear transformation. tanh is the hyperbolic tangent function, which maps the linear transformation result of the input to the interval (-1, 1), giving the attention weights a certain nonlinearity.

[0091] This unit uses this model to process the features in the dynamic scene data output by the Hidden Markov Dynamic Scene Simulation Unit, assigning each feature a corresponding attention weight. In VR urban environment design, different features have varying degrees of importance to the scene. For example, when designing a VR environment around a city transportation hub, traffic flow features may be given a relatively high weight, while less important landscape features may be given a lower weight. This model allows for more flexible control of feature weights, highlighting key feature information and providing more targeted feature data for the subsequent intelligent generation and adaptation unit of virtual elements.

[0092] Preferably, the virtual element intelligent generation and adaptation unit combines the output results of an improved hidden Markov-attention fusion model and an attention model based on a hyperbolic tangent function when generating virtual elements. This unit first screens key feature data from the comprehensive dataset output by the urban geographic information multidimensional data fusion and analysis unit based on the feature weights assigned by the attention mechanism feature weight allocation control unit. Then, using this key feature data and the dynamic scene information deduced by the hidden Markov dynamic scene deduction and simulation unit, a variant model of a generative adversarial network (GAN) is used to generate virtual elements.

[0093] The training objective functions of the generator G and discriminator D of this variant model are as follows:

[0094]

[0095] Among them, x is the data distribution p from the real urban environment data (x) samples, such as real building models, landscape images, etc. z is obtained from the noise distribution p z (z) is the random noise vector sampled from the image. w It is a feature vector weighted by the attention mechanism, which contains key feature information in the design of virtual reality urban environment, such as architectural style, landscape type, etc. G(z, F w ) is the generator based on random noise z and weighted features F w The generated virtual element. D(x) is the discriminator's judgment result on the real sample x, D(G(z, F w )) is the virtual element G(z, F) generated by the discriminator w By training this variant GAN model, the intelligent virtual element generation and adaptation unit can generate virtual elements that better match the characteristics of the VR urban environment and the requirements of dynamic scenes. It also further adapts and adjusts the generated virtual elements to make them blend in with the urban geographical environment and dynamic scenes in terms of color, shape, and size.

[0096] Preferably, the AR interaction rule strategy formulation unit considers the dynamic scene deduced by the hidden Markov dynamic scene deduction simulation unit and the feature weights assigned by the attention mechanism feature weight allocation control unit when formulating interaction rules and strategies. This unit constructs a state-action-reward based reinforcement learning model, where the state S consists of the dynamic scene information of the urban environment and the feature information of the virtual elements, the action A is the interaction method between the user and the virtual element, and the reward R is set according to the interaction effect and the scene goal. The value function update formula of the reinforcement learning model is:

[0097]

[0098] Among them, Q(S t , A t ) indicates that in state S t Next, take action A t The value of is estimated. α is the learning rate, which controls the step size of each update. γ is the discount factor, which is used to balance the importance of immediate rewards and future rewards. t+1 In state S t Take action A t The immediate reward after S t+1 Is to perform action A t Then transfer to the next state.

[0099] Through continuous training through interaction with users, the AR interaction rule strategy formulation unit can formulate optimal AR interaction rules and strategies based on different urban environment dynamic scenes and virtual element characteristics. For example, in a city's commercial district, the interaction rules between users and virtual stores may be different from those in leisure areas such as parks.

[0100] Preferably, the real-time rendering optimization processing unit combines the outputs of the Hidden Markov Dynamic Scene Simulation Unit and the Attention Mechanism Feature Weight Allocation Control Unit when performing rendering optimization. This unit uses a hierarchical rendering algorithm based on feature importance to divide elements in the virtual reality city environment into different levels according to their feature importance.

[0101] For elements with high feature importance, a high-precision rendering algorithm is used to render them to ensure their details and quality; for elements with low feature importance, a low-precision rendering algorithm is used for fast rendering to improve rendering efficiency. The layering rule of this algorithm can be expressed by the following formula:

[0102]

[0103] Among them, L i Indicates the rendering level of element i, H represents the high-precision rendering level, and L represents the low-precision rendering level. i) is the feature F of element i by the attention mechanism i The calculated attention weight. θ is a pre-set threshold used to divide the rendering levels.

[0104] The real-time rendering optimization processing unit renders virtual elements according to this layering rule, and combines multi-threaded rendering technology and hardware acceleration technology to further improve rendering efficiency and quality, ensuring smooth rendering effects on different devices. In particular, when processing large-scale virtual reality urban environment data, it can effectively reduce rendering time and resource consumption.

[0105] Preferably, the system integrated feedback calibration unit adopts a feedback control model based on error correction when performing system calibration and adjustment. The model calculates the output error of each unit based on user feedback information and system operation data.

[0106] For the Hidden Markov Dynamic Scenario Simulation Unit, the error calculation formula is:

[0107]

[0108] Among them, E HMM is the error of the hidden Markov dynamic scene simulation unit. sim (S i ) is the state S obtained by the unit simulation i The probability of occurrence. real (S i ) is the actual observed state S i The probability of occurrence. m is the number of states.

[0109] For the attention mechanism feature weight distribution control unit, the error calculation formula is:

[0110]

[0111] Among them, E Att is the error of the attention mechanism feature weight distribution control unit. sim (F j ) is the feature f calculated by the unit j The attention weight of Att real (F j ) is the feature F obtained based on actual evaluation j The attention weights are n. n is the number of features.

[0112] The system's integrated feedback calibration unit adjusts the parameters of each unit based on this error information, such as the state transition probability matrix of the hidden Markov dynamic scene deduction simulation unit and the weight matrix of the attention mechanism feature weight distribution control unit, to ensure the stability and accuracy of the system, so that the system can better adapt to different urban environments and user needs.

[0113] Preferably, the urban geographic information multidimensional data fusion and analysis unit combines information from the hidden Markov dynamic scene deduction simulation unit and the attention mechanism feature weight allocation control unit when performing data fusion analysis. This unit adopts a multi-source data fusion model based on a Bayesian network to fuse urban geographic information data from different sources. The joint probability distribution formula of the Bayesian network is:

[0114]

[0115] Among them, X1, X2, ..., X k It is each variable in the urban geographic information data, such as terrain height, building area, etc. Pa(X i ) is the variable X i The parent node set of X i A set of variables that have a direct causal relationship.

[0116] During the fusion process, this unit adjusts the Bayesian network's conditional probability table based on the dynamic scene information provided by the Hidden Markov Dynamic Scene Simulation Unit and the feature weight information provided by the Attention Mechanism Feature Weight Allocation Control Unit, ensuring that the fused data better reflects the actual conditions and dynamic changes in the urban environment. For example, when predicting the possibility of new construction in a certain area of ​​the city, the weight of the building-related data in that area is increased, thereby more accurately parsing and integrating the geographic information data of that area.

[0117] Preferably, the system adopts a time series-based dynamic optimization mechanism during its overall operation. This mechanism combines the outputs of the hidden Markov dynamic scene deduction simulation unit and the attention mechanism feature weight allocation control unit to dynamically adjust each unit in the system.

[0118] At each time step t, the system is based on the current state S t and feature F t , calculate the adjustment parameters of each unit. For the hidden Markov dynamic scene simulation unit, adjust its state transition probability matrix A t The formula is:

[0119] A t+1 =A t +β·ΔA t

[0120] Among them, A t+1 Is the state transition probability matrix for the next time step. A t Is the state transition probability matrix of the current time step. β is the adjustment coefficient, which controls the amplitude of the adjustment. ΔA t It is the adjustment amount of the state transition probability matrix calculated based on the current state and features.

[0121] For the attention mechanism feature weight distribution control unit, adjust its weight matrix W t The formula is:

[0122] W t+1 =W t +γ·ΔW t

[0123] Among them, W t+1 is the weight matrix for the next time step. W t is the weight matrix of the current time step. γ is the adjustment coefficient. ΔW t It is the adjustment amount of the weight matrix calculated based on the current state and features.

[0124] Through this dynamic optimization mechanism based on time series, the system can adapt to the dynamic changes of the urban environment in real time, continuously optimize its own performance, and improve the quality and effect of virtual reality urban environment design.

[0125] like Figure 2 The operation of a virtual reality urban environment design system based on AR technology includes the following steps:

[0126] Multi-source geographic data access and integration steps: Urban geographic information data from different data sources, such as satellite remote sensing image data, ground surveying and mapping data, urban planning data, etc., are accessed and preliminarily integrated to form a set of original data with certain relevance. This set contains basic information on the city's topography, landforms, building distribution, and other aspects.

[0127] Preliminary deduction steps of hidden Markov scenarios: Use the hidden Markov model to process the initial integrated raw data set, consider various uncertainties and random changes in the urban environment, conduct preliminary deductions of the dynamic scenarios of the urban environment, and generate a series of initial data for possible future dynamic scenarios of the urban environment.

[0128] Attention feature screening and extraction steps: Use the attention mechanism to screen and extract features of the initial data of the dynamic scene obtained by preliminary deduction, assign corresponding weights to different features according to the importance and relevance of the scene, highlight the key feature information, and form a key data set after feature weighting.

[0129] Virtual element generation, adaptation and adjustment steps: Based on the feature-weighted key data set, the variant model of the generative adversarial network is used to intelligently generate various virtual elements suitable for the virtual reality urban environment, such as virtual buildings, virtual landscapes, etc., and adapt the generated virtual elements to make them blend with the urban geographical environment and dynamic scenes in terms of color, shape, size, etc.

[0130] Steps for dynamically formulating AR interaction rules: Based on the generated and adapted virtual elements and urban environment characteristics, a state-action-reward-based reinforcement learning model is constructed. Through continuous training with users, AR interaction rules and strategies are dynamically formulated, including the interaction methods and interaction conditions between users and virtual elements.

[0131] Real-time rendering optimization processing execution steps: Combining the results of hidden Markov dynamic scene deduction and attention mechanism feature weight allocation, a layered rendering algorithm based on feature importance is used to render the virtual reality city environment, and multi-threaded rendering technology and hardware acceleration technology are used for optimization processing to improve rendering efficiency and quality, ensuring smooth rendering effects on different devices.

[0132] System feedback calibration loop adjustment steps: collect user feedback information and system operation data, calculate the output error of each unit, and use the feedback control model based on error correction to adjust the parameters of each unit in the system, such as adjusting the state transition probability matrix of the hidden Markov model, the weight matrix of the attention mechanism, etc., and then re-enter the next round of operation process to form a cyclic adjustment process to ensure the stability and accuracy of the system.

[0133] This system demonstrates significant advantages in data processing, scenario simulation, and interactive experience, effectively overcoming many of the drawbacks of existing technologies. At the data fusion level, existing systems often face challenges integrating heterogeneous data from multiple sources and poor data correlation, leading to missing or inconsistent basic design data. However, this system, through a multi-dimensional data fusion and analysis unit for urban geographic information, employs complex fusion algorithms to deeply integrate multi-source data such as topography, building distribution, and more, achieving centimeter-level accuracy. This ensures data integrity and accuracy, providing solid data support for design.

[0134] In terms of dynamic scene simulation and feature processing, the models used in traditional systems struggle to simulate the random changes in urban environments and fail to highlight key design elements. This system leverages a Hidden Markov Dynamic Scene Simulation Unit and an improved Hidden Markov-Attention Fusion Model to accurately predict future urban development trends. It also utilizes an attention mechanism feature weighting control unit, combined with an attention model based on the hyperbolic tangent function, to assign weights to different features based on scene importance, preventing designers from being distracted by secondary information and significantly improving the scientific and forward-looking nature of the design.

[0135] In terms of virtual element generation, interaction, and system optimization, existing technologies generate virtual elements with poor environmental adaptability, simple interaction rules, and a difficult balance between rendering efficiency and quality. Furthermore, there is a lack of an effective feedback calibration mechanism. This system's intelligent virtual element generation and adaptation unit generates highly realistic virtual elements by generating adversarial network variants and combining scene features. The AR interaction rule strategy formulation unit builds personalized interaction rules based on reinforcement learning. The real-time rendering optimization processing unit utilizes a layered rendering algorithm and hardware acceleration technology to balance rendering efficiency and quality. The system's integrated feedback calibration unit uses an error correction feedback control model and a time series dynamic optimization mechanism to adjust system parameters in real time, ensuring stable system operation and comprehensively improving the efficiency and user experience of virtual reality urban environment design.

[0136] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "disposed," "installed," "connected," "connected," and "fixed" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; they may refer to mechanical connections or electrical connections; they may refer to direct connections or indirect connections through an intermediate medium; and they may refer to internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0137] While embodiments of the present invention have been shown and described, it will be understood by those skilled in the art that various equivalent changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A virtual reality urban environment design system based on AR technology, characterized by: The system includes: Urban geographic information multi-dimensional data fusion and analysis unit: This unit integrates multi-source heterogeneous urban geographic information data and uses data fusion algorithms to fuse and analyze the data to form a comprehensive urban geographic information data set; Hidden Markov dynamic scene simulation unit: This unit is connected to the urban geographic information multi-dimensional data fusion and analysis unit, obtains the comprehensive data set output by it, and simulates the dynamic scenes in the urban environment based on the hidden Markov model to generate dynamic scene data of the future urban environment; Attention mechanism feature weight allocation and control unit: This unit extracts features from dynamic scene data obtained through simulation, uses the attention mechanism to assign corresponding weights to different features, and dynamically adjusts the weights based on the importance and relevance of the scene; Virtual element intelligent generation and adaptation unit: This unit intelligently generates various virtual elements suitable for the virtual reality city environment based on the feature information after weight distribution, and adapts and adjusts the virtual elements; AR interaction rule strategy formulation unit: This unit is connected to the virtual element intelligent generation and adaptation unit. It formulates AR interaction rules and strategies based on the generated virtual elements and the characteristics of the urban environment, including the interaction methods and conditions between users and virtual elements. Real-time rendering optimization processing unit: This unit is connected to the AR interaction rule strategy formulation unit, receives the formulated interaction rules and virtual element data, and uses real-time rendering algorithms to render the virtual reality city environment and perform optimization processing; System integrated feedback calibration unit: This unit is connected to the real-time rendering optimization processing unit to perform integrated management of the operation of the entire system, receive user feedback information and system operation data, and calibrate and adjust each unit in the system.

2. The virtual reality urban environment design system based on AR technology according to claim 1 is characterized in that: The hidden Markov dynamic scene deduction simulation unit adopts an improved hidden Markov-attention fusion model when performing dynamic scene deduction simulation. The model formula is: Among them, P(S t+1 ∣S t , F t ) represents the state S at time t t and feature F t Next, the state S at time t+1 t+1 The probability of occurrence, P(S t+1 ∣S t ) is the state transition probability in the traditional hidden Markov model, which represents the state S at time t t Transition to state S at time t+1 t+1 The probability of F t is the eigenvector at time t, denoted as F t =(F t，1 , F t，2 ,…,F t，n ), where the characteristics are the building density in the urban environment, the population flow speed, the virtual reality urban environment design parameters, Att(F t，i ) is the attention mechanism for feature F t，i The calculated attention weight reflects the importance of the feature in the current scene deduction, α i It is an adjustment coefficient used to balance the influence of the state transition probability and attention weight of the traditional hidden Markov model.

3. The virtual reality urban environment design system based on AR technology according to claim 1 is characterized in that: The attention mechanism feature weight allocation control unit adopts an attention model based on the hyperbolic tangent function when calculating the attention weight. The model formula is: Among them, W is a learnable weight matrix used to adjust the feature F t，i A linear transformation is performed, and its dimension is determined according to the dimension of the feature vector and the model design. b is a bias vector, which is a learnable parameter used to adjust the result of the linear transformation. Tanh is a hyperbolic tangent function, which maps the linear transformation result of the input to the interval (-1, 1), so that the attention weight has certain nonlinear characteristics.

4. The virtual reality urban environment design system based on AR technology according to claim 1 is characterized in that: When generating virtual elements, the virtual element intelligent generation and adaptation unit combines the output results of the improved hidden Markov-attention fusion model and the attention model based on the hyperbolic tangent function. The unit first screens out key feature data from the comprehensive data set output by the urban geographic information multidimensional data fusion and analysis unit based on the feature weights assigned by the attention mechanism feature weight distribution control unit. Then, using these key feature data and the dynamic scene information deduced by the hidden Markov dynamic scene deduction and simulation unit, a variant model of the generative adversarial network is used to generate virtual elements. The training objective functions of the generator G and discriminator D of the variant model are as follows: Among them, x is the data distribution p from the real urban environment data The samples of (x) include real building models, landscape images, and z are obtained from the noise distribution p z The random noise vector sampled from (z), F w It is a feature vector weighted by the attention mechanism, which contains the key feature information in the design of virtual reality urban environment, including architectural style, landscape type, G(z, F w ) is the generator based on random noise z and weighted features F w The generated virtual element, D(x) is the discriminant result of the real sample x, D(G(z, F w )) is the virtual element G(z, F) generated by the discriminator w )’s judgment result.

5. The virtual reality urban environment design system based on AR technology according to claim 1 is characterized in that: When formulating interaction rules and strategies, the AR interaction rule strategy formulation unit considers the dynamic scene deduced by the hidden Markov dynamic scene deduction simulation unit and the feature weights assigned by the attention mechanism feature weight distribution control unit. The unit constructs a state-action-reward based reinforcement learning model, in which the state S is composed of the dynamic scene information of the urban environment and the feature information of the virtual elements, the action A is the interaction method between the user and the virtual element, and the reward R is set according to the interaction effect and the scene goal. The value function update formula of the reinforcement learning model is: Among them, Q(S t , A t ) indicates that in state S t Next take action A t The value estimate of , α is the learning rate, which controls the step size of each update, γ is the discount factor, which is used to balance the importance of immediate rewards and future rewards, R t+1 In state S t Take action A t The immediate reward after S t+1 Is to perform action A t Then transfer to the next state.

6. The virtual reality urban environment design system based on AR technology according to claim 1 is characterized in that: When performing rendering optimization, the real-time rendering optimization processing unit combines the outputs of the hidden Markov dynamic scene deduction simulation unit and the attention mechanism feature weight allocation control unit. The unit adopts a hierarchical rendering algorithm based on feature importance to divide the elements in the virtual reality city environment into different levels according to their feature importance. For elements with higher feature importance, a high-precision rendering algorithm is used for rendering, and for elements with lower feature importance, a low-precision rendering algorithm is used for fast rendering. The hierarchical rule of the algorithm is expressed by the following formula: Among them, L i Indicates the rendering level of element i, H represents the high-precision rendering level, L represents the low-precision rendering level, Att(F i ) is the feature F of element i by the attention mechanism i The calculated attention weight,θ, is a pre-set threshold used to divide the rendering levels.

7. The virtual reality urban environment design system based on AR technology according to claim 1 is characterized in that: The system integrated feedback calibration unit adopts a feedback control model based on error correction when performing system calibration and adjustment. The model calculates the output error of each unit based on user feedback information and system operation data. For the hidden Markov dynamic scene deduction simulation unit, the error calculation formula is: Among them, E HMM is the error of the hidden Markov dynamic scene simulation unit, P sim (S i ) is the state S obtained by the unit simulation i The probability of occurrence, P real (S i ) is the actual observed state S i The probability of occurrence, m is the number of states, and for the attention mechanism feature weight distribution control unit, the error calculation formula is: Among them, E Att is the error of the attention mechanism feature weight distribution control unit, Att sim (F j ) is the characteristic F calculated by the unit j The attention weight, Att real (F j ) is the feature F obtained based on actual evaluation j The attention weight is , and n is the number of features.

8. The virtual reality urban environment design system based on AR technology according to claim 1 is characterized in that: When performing data fusion analysis, the urban geographic information multidimensional data fusion analysis unit combines information from the hidden Markov dynamic scene deduction simulation unit and the attention mechanism feature weight distribution control unit. The unit adopts a multi-source data fusion model based on the Bayesian network to fuse urban geographic information data from different sources. The joint probability distribution formula of the Bayesian network is: Among them, X1, X2, ..., X k It is the various variables in urban geographic information data, including terrain height, building area, Pa(X i ) is the variable X i The parent node set of X i A set of variables that have a direct causal relationship.

9. The virtual reality urban environment design system based on AR technology according to claim 1 is characterized in that: During the overall operation of the system, a dynamic optimization mechanism based on time series is adopted. This mechanism combines the output of the hidden Markov dynamic scene deduction simulation unit and the attention mechanism feature weight distribution control unit to dynamically adjust each unit in the system. At each time step t, the system adjusts the current state S according to the current state S. t and feature F t , calculate the adjustment parameters of each unit, and for the hidden Markov dynamic scene simulation unit, adjust its state transition probability matrix A t The formula is: A t+1 =A t +β·ΔA t Among them, A t+1 is the state transition probability matrix for the next time step, A t is the state transition probability matrix of the current time step, β is the adjustment coefficient, which controls the amplitude of the adjustment, ΔA t It is the adjustment amount of the state transition probability matrix calculated based on the current state and features; For the attention mechanism feature weight distribution control unit, adjust its weight matrix W t The formula is: IN t+1 =In t +γ ΔW t Among them, W t+1 is the weight matrix for the next time step, W t is the weight matrix of the current time step, γ is the adjustment coefficient, ΔW t It is the adjustment amount of the weight matrix calculated based on the current state and features.

10. A virtual reality urban environment design system based on AR technology according to any one of claims 1 to 9, characterized in that: The system operation includes: Multi-source geographic data access and integration: Urban geographic information data from different data sources, including satellite remote sensing image data, ground surveying and mapping data, and urban planning data, are accessed and initially integrated to form a relevant raw data set that includes basic information on the city's topography, landforms, and building distribution. Preliminary deduction of hidden Markov scenarios: Using the hidden Markov model to process the initially integrated raw data set, taking into account various uncertainties and random changes in the urban environment, preliminary deduction of the dynamic scenarios of the urban environment is carried out to generate initial data for future dynamic scenarios of the urban environment; Attention feature screening and extraction: The attention mechanism is used to screen and extract features from the initial dynamic scene data obtained through preliminary deduction. Different features are assigned corresponding weights based on the importance and relevance of the scene, highlighting the key feature information and forming a feature-weighted key data set. Virtual element generation and adaptation: Based on a weighted feature set of key data, a variant model of a generative adversarial network is used to intelligently generate various virtual elements suitable for the VR city environment, including virtual buildings and virtual landscapes. These generated virtual elements are then adapted to blend in with the city's geographical environment and dynamic scenes in terms of color, shape, and size. Dynamic formulation of AR interaction rules: Based on the generated and adapted virtual elements and urban environment characteristics, a state-action-reward-based reinforcement learning model is constructed. Through continuous training with user interactions, AR interaction rules and strategies are dynamically formulated, including the interaction methods and conditions between users and virtual elements. Real-time rendering optimization: Combining the results of Hidden Markov dynamic scene deduction and attention mechanism feature weight allocation, a layered rendering algorithm based on feature importance is used to render the virtual reality city environment, and multi-threaded rendering technology and hardware acceleration technology are used for optimization. System feedback calibration loop adjustment: collect user feedback information and system operation data, calculate the output error of each unit, and use the feedback control model based on error correction to adjust the parameters of each unit in the system, including adjusting the state transition probability matrix of the hidden Markov model and the weight matrix of the attention mechanism. Then, re-enter the next round of operation process, forming a cyclic adjustment process.

Images