A historical scene adaptive LOD generation system adaptive to VR and AR

Abstract

The present application relates to the field of virtual reality / augmented reality technology, and particularly refers to a historical scene adaptive LOD generation system adapted to VR and AR. The system comprises a historical scene semantic asset library module, a multi-dimensional LOD hierarchical generation module, an adaptive scheduling engine module, a power consumption perception rendering optimization module, and an edge-end collaborative low-carbon rendering module. The system takes the LOD generation with cultural feature fidelity constraint and the adaptive LOD scheduling driven by energy consumption budget as the core invention points, wherein LOD0 is the highest precision level and LOD4 is the lowest precision level. The accurate grading of cultural feature fidelity guarantees the professionalism of historical scene display: by incorporating the historical value of cultural relics into the LOD grading basis, a cultural feature fidelity link of "feature recognition-vertex mapping-simplification constraint" is established, realizing the unity of high-precision presentation of core cultural relics and reasonable simplification of secondary areas, avoiding the damage of traditional LOD "one-size-fits-all" simplification to cultural information, and being particularly suitable for professional historical scene VR and AR guide of museums, archaeological sites, etc.

Description

Technical Field

[0001] This invention relates to the field of virtual reality / augmented reality technology, specifically to an adaptive LOD generation system for historical scenes that is compatible with VR and AR. Background Technology

[0002] With the widespread application of VR and AR technologies in digital cultural heritage and historical scene restoration, high-precision 3D reconstruction of historical scenes has become a core means of digitizing cultural heritage. However, historical scenes such as ancient buildings and cultural relic complexes generally have characteristics such as high model polygon count, large amount of texture data, and wide scene spatial span. Directly performing full-precision rendering on VR and AR terminals will lead to problems such as device computing power overload, frame rate drop, severe overheating, and drastically shortened battery life, which seriously affect the user's immersive experience.

[0003] Level of Detail (LOD) technology is the mainstream solution to the above problems. It generates versions of the 3D model with different levels of detail, dynamically switching between them at runtime based on view distance and perspective, sacrificing accuracy in non-critical areas for improved rendering performance. Existing LOD technologies mainly suffer from the following drawbacks:

[0004] The traditional LOD (Level of Detail) classification is based solely on geometric complexity and viewing distance, failing to differentiate the historical value levels of the artifacts themselves, the core components of ancient buildings, and ordinary environmental decorations. This easily leads to unreasonable phenomena such as national treasure-level artifacts being downscaled while background decorations consume a large amount of computing power. Furthermore, there is no specific mechanism for ensuring the authenticity of core cultural features such as patterns, inscriptions, and damage to artifacts. The simplification process can easily result in the loss of cultural information, which does not meet the professional requirements for displaying historical scenes.

[0005] The adaptive scheduling lacks sufficient dimensions and energy efficiency awareness: Current adaptive LOD only considers view distance and frame rate feedback, without incorporating the remaining battery power of the terminal device and the real-time power consumption of the CPU / GPU into the scheduling criteria. VR and AR devices generally have limited battery capacity, and the battery life under high-fidelity rendering is usually less than 2 hours, which cannot meet the needs of long-term guided tours. Moreover, the heat generated by high power consumption will also trigger the device to reduce its frequency, further degrading the experience.

[0006] Pre-generated models are wasteful of resources and do not meet the requirements of low-carbon and environmental protection: Traditional LOD mostly adopts offline pre-generation of full-level models, lacks progressive dynamic generation capabilities, and has a low reuse rate of intermediate-level models. This not only occupies a lot of storage and computing resources, but also wastes computing resources and power energy. At the same time, full cloud rendering solutions will exacerbate data center energy consumption, which is not in line with the development direction of green computing under the dual-carbon goal.

[0007] Poor cross-device compatibility: The computing power of different VR and AR terminals varies greatly. From high-end PCVR to lightweight AR glasses, the same LOD strategy cannot be used universally. It is necessary to repeatedly manufacture for different devices, which increases the production cost and carbon emissions of digital assets.

[0008] Therefore, developing an adaptive LOD generation system that can ensure the accuracy of historical scene semantics and cultural features, significantly reduce terminal power consumption, reduce computing resource waste, and adapt to various VR and AR devices has important technical value and social significance. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an adaptive LOD generation system and method for historical scenes that is compatible with VR and AR. This invention solves the technical problems of traditional LOD technology, such as lack of cultural feature fidelity, insufficient energy efficiency optimization, serious waste of pre-generated resources, and poor cross-device compatibility. While ensuring the professionalism of historical and cultural display, it achieves the technical effects of terminal energy saving and consumption reduction and low-carbon environmental protection throughout the entire process.

[0010] Technical solution

[0011] To achieve the above objectives, this invention provides an adaptive LOD generation system for historical scenes adapted to VR and AR, including a historical scene semantic asset library module, a multi-dimensional LOD hierarchical generation module, an adaptive scheduling engine module, a power consumption-aware rendering optimization module, and an edge-device collaborative low-carbon rendering module. The core invention of this system is LOD generation constrained by cultural feature fidelity and adaptive LOD scheduling driven by energy consumption budgets, where LOD0 is the highest precision level and LOD4 is the lowest precision level. The specific composition of each module is as follows:

[0012] The Historical Scene Semantic Asset Library module is used to perform semantic annotation, hierarchical classification, and cultural feature weight mapping on 3D assets in historical scenes, establishing a reusable and lightweight asset library. Assets are divided into four semantic levels based on historical value and visual importance: core cultural relics, key buildings, scene components, and environmental decorations. For core cultural relics, pixel-level feature masks for patterns, inscriptions, and damaged areas are generated through archaeological data annotation combined with texture semantic segmentation. These 2D masks are then mapped to 3D mesh vertices to obtain the cultural feature weights of each vertex, providing a basis for subsequent constraint simplification. Simultaneously, recurring architectural components are instantiated and encapsulated as assets, supporting cross-scene reuse.

[0013] A multi-dimensional LOD (Level of Detail) grading generation module, connected to the historical scene semantic asset library module, is used to generate grading LOD models based on semantic level and cultural feature weights. Different geometric facet reduction ratios and texture resolution attenuation coefficients are set for different semantic levels. A quadratic error measure (QEM) edge folding algorithm incorporating cultural feature weights is used for mesh simplification, forcibly preserving the cultural feature accuracy of core cultural relics. The module includes geometric simplification units, texture compression units, and semantic shape-preserving units. It pre-generates three anchor-level models (LOD0, LOD2, and LOD4) and complete edge folding record sequences. Based on Progressive Mesh (PM) technology, it supports dynamic generation of transitional levels such as LOD1 and LOD3 at runtime through vertex splitting operations.

[0014] The adaptive scheduling engine module connects to the multi-dimensional LOD level generation module and the power consumption-aware rendering optimization module, respectively, to dynamically determine the LOD level of each scene region based on real-time multi-dimensional input. Input dimensions include the user's focal area, viewing distance, device computing power level, remaining battery power, and scene semantic weight. The scheduling strategy employs a two-order mechanism of precision scoring-level mapping, combining semantic gain, viewing distance attenuation, and power consumption constraints to calculate continuous precision scores, which are then output as discrete LOD levels after boundary clamping. Focal areas and assets with high semantic weights are prioritized for high-level precision allocation. When battery power falls below a preset threshold, the LOD level of non-focal areas and low semantic weight areas is progressively reduced. The module incorporates three energy efficiency strategies: performance mode, balanced mode, and energy-saving mode, which automatically switch based on remaining battery power.

[0015] The power consumption-aware rendering optimization module connects to the adaptive scheduling engine module to dynamically adjust rendering parameters in conjunction with LOD scheduling. It monitors CPU and GPU utilization, power consumption, frame rate, and device temperature in real time, employing a closed-loop feedback mechanism. Data is collected every 100ms. When the actual frame rate is lower than 90% of the target frame rate, LOD downgrading is triggered. When the frame rate is higher than 120% of the target frame rate and there is sufficient power consumption, the LOD level of the secondary focus area is appropriately increased. It also features a temperature protection mechanism that proactively reduces the load when the device temperature exceeds a threshold, lowering overall device power consumption and extending battery life without the user noticing.

[0016] The edge-to-edge collaborative low-carbon rendering module connects to the multi-dimensional LOD hierarchical generation module and the adaptive scheduling engine module to achieve hierarchical collaborative rendering between the terminal and edge nodes. A hierarchical offloading strategy is adopted: high-precision models and complex lighting calculations at LOD0 and LOD1 levels are offloaded to edge nodes for rendering as video streams, while low-to-medium precision models at LOD2 to LOD4 levels are rendered locally on the terminal. The system dynamically adjusts the offloading ratio based on network bandwidth. The module features pose synchronization, depth occlusion detection, and asynchronous reprojection correction mechanisms to achieve seamless compositing of edge-rendered and locally rendered images. Simultaneously, on-demand transmission and local caching mechanisms reduce network bandwidth consumption and overall computing power consumption.

[0017] Furthermore, the historical scene semantic asset library module has a built-in cultural relic value labeling system, supports the import of archaeological reports and cultural relic grading data, and automatically matches semantic levels; the asset library adopts an instantiation and texture sharing mechanism, allowing similar building components to reuse the same set of LOD assets, avoiding repeated modeling and repeated LOD generation.

[0018] Furthermore, the three-level energy efficiency strategy of the adaptive scheduling engine module is as follows:

[0019] When the battery level is greater than 60%, the default performance mode is used, prioritizing visual accuracy at the LOD level across all scenarios.

[0020] When the battery level is between 30% and 60%, the balance mode will be automatically switched, reducing the LOD level of non-sight focus areas by 1 level and the environmental decoration areas by 2 levels.

[0021] When the battery level is less than 30%, the system will force the system to enter energy-saving mode. Only the core cultural relic's focal area will maintain its original level, while the rest of the area will be downgraded to the maximum extent. At the same time, secondary lighting and shadow effects will be turned off.

[0022] Another aspect of the present invention provides an adaptive LOD generation method for historical scenes that is compatible with VR and AR, comprising the following steps:

[0023] S1: Semantic classification and cultural feature annotation of historical scene assets: The historical scene 3D assets are divided into four semantic levels according to historical value: core cultural relics level, key building level, scene component level and environmental decoration level; for core cultural relics level assets, the pattern, inscription and damage feature areas are identified and pixel-level feature masks are generated and mapped to the cultural feature weights of the 3D mesh vertices.

[0024] S2: Generation of multi-level LOD models constrained by cultural features: Based on the semantic level setting of differentiated simplification strategies, the QEM edge folding algorithm with cultural feature weights is used to simplify the mesh, and three anchor point level models of LOD0, LOD2 and LOD4 are pre-generated, while the edge folding record sequence is generated at the same time.

[0025] S3: Energy Budget-Driven Adaptive LOD Scheduling: Collects terminal device battery power, real-time power consumption, and computing power level, combines user line of sight focus, line of sight distance, and asset semantic weight, calculates LOD recommendation level through accuracy scoring formula, and outputs final LOD scheduling instruction after boundary clamping;

[0026] S4: Power-Aware Closed-Loop Rendering Optimization: Real-time monitoring of device frame rate, power consumption and temperature data to form a closed-loop feedback, dynamically adjust LOD level and rendering parameters, and maintain a dynamic balance between performance and power consumption.

[0027] S5: Edge-to-device collaborative hierarchical rendering and image compositing: A hierarchical offloading strategy is adopted. High-precision LOD level and complex lighting calculations are rendered as video streams at edge nodes and pushed to the terminal, while medium and low-precision LOD levels are rendered locally on the terminal. The two parts of the image are composited and output through pose synchronization, depth occlusion judgment and reprojection correction.

[0028] Beneficial effects

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] Precise grading of cultural features ensures the professionalism of historical scene display: By incorporating the historical value of cultural relics into the LOD grading criteria and establishing a cultural feature fidelity link of "feature recognition - vertex mapping - simplification constraints", the high-precision presentation of core cultural relics and reasonable simplification of secondary areas are achieved, avoiding the damage to cultural information caused by the traditional "one-size-fits-all" simplification of LOD. It is especially suitable for VR and AR guided tours of professional historical scenes such as museums and archaeological sites.

[0031] Multi-dimensional energy efficiency scheduling significantly improves terminal battery life: By incorporating battery power, real-time power consumption, and device temperature into the LOD scheduling dimension and coordinating with rendering parameter optimization, the battery life of VR and AR devices for historical scene rendering is significantly extended without being noticeable to the user. This effectively solves the battery life problem of mobile VR and AR devices, while reducing device heat generation and hardware wear.

[0032] The end-to-end low-carbon design aligns with the concept of green environmental protection: by reusing asset instantiation, the computational power consumption of repeated modeling is reduced; by generating runtime transition levels based on progressive meshes, the waste of offline pre-generation resources is reduced; and edge-to-edge collaborative on-demand rendering replaces full cloud rendering. The entire process from asset production and offline processing to runtime rendering reduces the energy consumption of computing resources and significantly reduces carbon emissions throughout the entire lifecycle of digital culture projects.

[0033] Cross-device adaptive adaptation reduces development costs: The system can automatically identify the terminal's computing power level and match the optimal LOD strategy and basic accuracy benchmark. The same set of scene assets can be adapted to multiple devices such as PCVR, all-in-one VR, and AR glasses, avoiding repeated production and further saving R&D resources and costs.

[0034] Obviously, based on the above description of the present invention, and according to common technical knowledge and conventional methods in the field, various other modifications, substitutions or alterations can be made without departing from the basic technical concept of the present invention.

[0035] The following detailed embodiments further illustrate the above-described content of the present invention. However, this should not be construed as limiting the scope of the present invention to the following examples. All technologies implemented based on the above-described content of the present invention fall within the scope of the present invention. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the overall system architecture of the present invention;

[0037] Figure 2 This is a schematic diagram of the offline preprocessing structure of the present invention;

[0038] Figure 3 This is a schematic diagram illustrating the prior art and the inventive step of the present invention. Detailed Implementation

[0039] Example 1

[0040] like Figures 1-3 As shown, this embodiment is a VR restoration scene of an ancient Song Dynasty building complex, which runs on a mainstream consumer-grade VR all-in-one terminal. The system includes a historical scene semantic asset library module, a multi-dimensional LOD hierarchical generation module, an adaptive scheduling engine module, a power consumption-aware rendering optimization module, and an edge-device collaborative low-carbon rendering module.

[0041] Construction of Historical Scene Semantic Asset Library Module

[0042] The historical scene semantic asset library module serves as the system's data foundation, responsible for the semantic annotation, cultural feature weight mapping, and classification reuse of 3D assets in historical scenes. In this embodiment, the scene includes various assets such as the main hall, stone carvings, wooden decorations, vegetation, and terrain. The module first imports the cultural relic classification data from the archaeological report, classifying Song Dynasty stone carvings and dougong (bracket sets) paintings within the scene as core cultural relics, the main hall's wooden structure and platform as key buildings, ordinary doors and windows, floor tiles, and railings as scene components, and vegetation, distant mountains, and ground paving as environmental decorations.

[0043] For core cultural relics-level assets, cultural characteristic labeling and weight mapping are performed, with the following specific steps:

[0044] The first step is feature mask generation: import high-definition texture maps of cultural relics and archaeological feature annotation data, and use the U-Net semantic segmentation network to perform pixel-level recognition of three types of features: patterns, inscriptions, and damage. Output a binary feature mask map, where a mask value of 1 corresponds to a cultural feature area and 0 corresponds to a non-feature area. Finally, it is verified and corrected by cultural heritage professionals.

[0045] The second step is vertex weight mapping: traverse all vertices of the grid, query the feature mask map based on the vertex UV coordinates, set the cultural feature weight of vertices that fall completely within the feature region to 3.0, set the weight of vertices within a 1cm buffer range of the feature region edge to 1.5, and set the weight of vertices in non-feature regions to 1.0; finally, the simplified vertex weight is the product of the cultural feature weight and the semantic level weight, and the semantic level weight of the core cultural relic is 1.0.

[0046] For frequently recurring components in ancient architecture, such as roof tiles, rafters, and column bases, the module instantiates and encapsulates these assets. Only one LOD model is generated for each type of component, and these models are then batch-retrieved through instantiated rendering within the scene. Compared to traditional component-by-component modeling, instantiation and reuse can reduce the amount of model data by more than 70%, while significantly reducing the computational load of LOD pre-generation. The asset library supports cross-scene calls, allowing similar Song Dynasty architectural components to be reused in multiple historical scene projects, achieving low-carbon and environmentally friendly practices from the asset production stage.

[0047] Execution flow of multi-dimensional LOD hierarchical generation module

[0048] The multi-dimensional LOD hierarchical generation module is connected to the historical scene semantic asset library module to perform differentiated LOD generation. The process is divided into four steps: semantic weight allocation, geometric simplification, texture compression, and quality verification. The module includes geometric simplification units, texture compression units, and semantic shape-preserving units, and uses the QEM edge folding algorithm with cultural feature weights to achieve feature fidelity simplification.

[0049] Geometric simplification units are configured with differentiated facet reduction ratios for different semantic levels:

[0050] Core cultural relic level: The face shrinkage ratio from LOD0 to LOD4 is 100%, 70%, 50%, 30%, and 15% respectively. The vertex weight of the cultural feature region is locked by semantic shape preservation unit, and the vertex density of the feature region is forcibly preserved during simplification.

[0051] Key building category: Shrinkage ratios of 100%, 60%, 35%, 20%, and 10%, preserving the overall building outline and key structures;

[0052] Scene component level: shrinkage ratios of 100%, 50%, 25%, 10%, and 5%;

[0053] Environmentally friendly decorative grade: shrinkage ratios of 100%, 40%, 15%, 5%, and 2%, allowing for significant simplification to save computing power.

[0054] The weight constraint calculation method for QEM edge folding is as follows: for the edge to be folded First, calculate the original QEM folding error. ,in , Let be the quadratic error matrix of the vertex; then, introduce vertex simplification weights to correct the folding cost: The simplification process performs edge folding in ascending order of folding cost. Feature regions, due to their higher weight and greater folding cost, are ranked lower and are not easily simplified, thus prioritizing the preservation of the geometric accuracy of cultural features in low-face-count models.

[0055] The texture compression unit adopts a differentiated compression strategy: the resolution of core cultural relic-level textures is gradually reduced from 4K to 512px, using a high-fidelity compression format; the resolution of environmental decoration-level textures is gradually reduced from 1K to 64px, using a high compression ratio format.

[0056] This module uses progressive mesh technology to dynamically generate transition levels, as detailed below:

[0057] Offline generation stage: pre-generate complete models at three anchor levels: LOD0, LOD2, and LOD4, and simultaneously save the complete edge folding record sequence from LOD0 to LOD4. Each record contains the vertex index of the folded edge, the coordinates of the new vertex, the texture coordinates, the normal vector, and the information of adjacent faces. Compared with offline generation of all 5 levels, it can reduce offline computation and storage usage by 40%.

[0058] Runtime generation phase: Based on the scheduling instructions and the corresponding anchor point model, the vertex splitting operation is performed in reverse order from the edge folding sequence to quickly generate transition levels such as LOD1 and LOD3; the transition level texture is generated by bilinear interpolation of adjacent anchor point textures, and the generation time of a single asset transition is <2ms with no visual stuttering.

[0059] Adaptive scheduling engine module scheduling logic

[0060] The adaptive scheduling engine module is the core of the system's decision-making, receiving five types of input data in real time: the coordinates of the user's gaze focus obtained from eye tracking, the viewing distance calculated from scene depth information, the computing power level of the terminal device, the remaining battery power, and the semantic weight of the assets. The scheduling adopts a two-order mechanism of "continuous accuracy scoring - discrete level mapping".

[0061]

[0062] in This is the baseline accuracy value corresponding to the computing power of the equipment. , , These are the normalized values ​​for semantic, line-of-sight, and power consumption constraints, respectively. , , These are the corresponding weighting coefficients;

[0063] The first step is to calculate the continuous accuracy score. (The smaller the score, the higher the accuracy and the smaller the LOD level value), the calculation formula is:

[0064]

[0065] Parameter definitions and values:

[0066] The baseline accuracy value is determined by the computing power level of the device; in this embodiment, the VR all-in-one machine takes the value of 2.

[0067] Semantic gain coefficient, fixed value 2.0;

[0068] Semantic level normalization value (0~1): core cultural relic level = 1, key building level = 0.7, scene component level = 0.4, environmental decoration level = 0.2;

[0069] Line-of-sight attenuation weight, fixed value 2.5;

[0070] Normalized sight distance value (0~1), 0 when sight distance ≤ 1m, 1 when sight distance ≥ 20m, and linear interpolation in between;

[0071] Power consumption constraint penalty coefficient, fixed value 1.5;

[0072] : Power consumption constraint normalized value (0~1), 0 when the power consumption is >60% and there is a power surplus, 1 when the power consumption is <20% and there is a high load, and linear interpolation in between.

[0073] The second step, discrete rating mapping and boundary handling: After rounding down the continuous ratings, a clamping function is used to restrict them to the interval of 0-4, resulting in the final LOD rating. Here, clamp(x,min,max) is a boundary clamping function to prevent outliers with negative or excessive levels from occurring.

[0074] In this embodiment, the engine has a built-in three-level energy efficiency strategy that automatically switches according to the battery level:

[0075] Performance Mode: Enabled when battery level > 60%, target frame rate 90Hz, prioritizes visual quality across all scenes, and only significantly downgrades the visual quality of distant environmental decorations;

[0076] Balanced mode: Automatically triggered when battery level is 30%-60%, target frame rate is maintained at 90Hz, LOD in non-focus areas is generally reduced by 1 level, soft shadows at long distances are turned off, and terminal power consumption is reduced by about 20%.

[0077] Energy Saving Mode: Forced to be enabled when battery level is below 30%, the target frame rate is reduced to 72Hz, only the core artifacts in the field of view are maintained at LOD1 level, and the rest of the area is reduced to LOD3-LOD4 at most. At the same time, anti-aliasing and ambient light occlusion are turned off, and the overall power consumption of the terminal is significantly reduced to meet the user's needs to complete the remaining tour.

[0078] When a user's gaze is focused on a core artifact, the artifact is automatically assigned the highest LOD level, while the level of surrounding architectural components within a certain range is appropriately increased. When the gaze moves quickly, the system predicts the movement trajectory and loads the LOD data for the next view in advance to avoid lag when switching between views.

[0079] Power consumption-aware rendering optimization module closed-loop control

[0080] The power consumption-aware rendering optimization module works in conjunction with the adaptive scheduling engine module to form a closed-loop control. The module collects CPU and GPU utilization, power consumption, actual frame rate, and device temperature every 100ms through the system's underlying interface.

[0081] When a frame rate of 3 consecutive frames is detected to be lower than 90% of the target value, a downgrade request is sent to the scheduling engine to prioritize reducing the LOD level of low semantic regions.

[0082] When the frame rate is detected to be consistently higher than the target value of 120% and the GPU utilization is lower than 50%, the LOD level of the secondary focus area is appropriately increased to make full use of the surplus computing power to improve image quality.

[0083] When the device temperature exceeds the preset threshold, it actively triggers LOD downgrading and rendering parameter reduction to avoid sudden stuttering caused by device overheating and frequency reduction, thus improving the stability of the experience.

[0084] Edge-to-edge collaborative low-carbon rendering module operation mechanism

[0085] The edge-to-edge collaborative low-carbon rendering module is deployed on the edge computing nodes of the cultural and museum park. It adopts a hierarchical offloading architecture and has a complete screen synchronization and compositing mechanism.

[0086] Tiered offloading strategy: High-precision model rendering, global illumination, real-time lighting and other complex calculations for LOD0 and LOD1 levels are completed at edge nodes, and the rendering results are pushed to the terminal as a low-latency video stream; medium and low-precision models from LOD2 to LOD4 are rendered directly on the terminal. The system dynamically adjusts the offloading ratio according to real-time network bandwidth: 70% of high-precision areas are rendered at the edge in 5G environments, significantly reducing terminal GPU power consumption; the offloading ratio is adjusted to 40% in WiFi environments; and all rendering is switched to local rendering in weak network environments to ensure continuous experience.

[0087] Image synchronization and compositing process:

[0088] Pose synchronization: Edge nodes and the terminal maintain a unified millisecond-level timestamp, and the terminal sends head pose data every 10ms; the edge nodes render the picture based on the pose and embed the corresponding timestamp and view matrix in the video stream; after receiving the data, the terminal performs motion compensation based on the pose difference to offset the transmission delay.

[0089] Depth occlusion detection: When rendering high-precision images at edge nodes, a depth map is output simultaneously, along with color... Figure 1 The same encoding is used for transmission; a local depth cache is generated synchronously when the terminal renders the local model; during the compositing stage, the depth values ​​are compared pixel by pixel, and objects with smaller depth values ​​are displayed first to achieve correct occlusion relationships and avoid clipping.

[0090] Asynchronous reprojection correction: To address the viewing angle deviation caused by transmission delay, the terminal uses asynchronous time warp (ATW) technology to perform geometric transformation correction on the edge image; the edge areas not covered by the transformed edge image are filled with local low-precision images to ensure the integrity of the field of view.

[0091] Layered compositing output: The final image is composed of three layers: the bottom layer is a low-precision environment scene rendered by the terminal, the middle layer is a high-precision scene with corrected edges, and the top layer is the terminal UI and interactive elements. After depth testing and Alpha mixing, it is output to the display device.

[0092] In addition, the module adopts an on-demand loading and caching mechanism, transmitting only the LOD data within the user's current field of view. The LOD resources of the areas that have been visited are cached locally on the terminal, so there is no need to retransmit and calculate them when visiting the same area again, which further saves bandwidth and computing resources.

[0093] Performance testing verification

[0094] This embodiment is based on a mainstream consumer-grade VR all-in-one device (Snapdragon XR2 chip, 8GB RAM, 5500mAh battery). The test scene has a total of 12 million triangles and 8GB of textures. The control group is a traditional general LOD scheduling scheme based solely on view distance. The test results are as follows:

[0095] Battery life performance: In balanced mode, the continuous rendering battery life of this system terminal is improved by about 41% compared with the control group, from 1.8 hours to 2.54 hours; in power saving mode, the battery life is improved by about 47%, reaching 2.65 hours.

[0096] Power consumption performance: Under the same visual accuracy, the average power consumption of the terminal is reduced by about 38% compared with the control group; in the edge-device collaborative mode, the total computing power energy consumption of the entire link is reduced by about 42% compared with the pure cloud rendering solution.

[0097] Image quality: The geometric simplification error in the core cultural relic area is reduced by about 65% compared with the traditional undifferentiated LOD scheme, and there is no obvious simplification distortion in the pattern and inscription areas;

[0098] Smoothness: The average frame rate remains stable at 90Hz across all scenes, with frame rate fluctuations of less than 5%, and there are no obvious visual jumps when switching LODs.

[0099] Example 2

[0100] This embodiment describes an augmented reality guided tour application scenario for historical districts on AR glasses. Compared to Embodiment 1, the adaptive scheduling engine module in this embodiment additionally incorporates two input dimensions: spatial positioning accuracy and ambient light intensity.

[0101] Since AR glasses have significantly lower computing power and battery capacity than VR all-in-one devices, the system defaults to starting in balanced mode, with semantic weights further tilted towards core cultural relics. When a user points their camera at a historical building plaque, only the building facade is upgraded to a high Level of Detail (LOD), while the surrounding streetscape remains at a low level. When the device detects sufficient ambient light and the AR display brightness can be reduced, it automatically lowers the LOD level of non-core areas, further saving energy in conjunction with screen brightness adjustment.

[0102] In this embodiment, the edge-to-edge collaborative low-carbon rendering module undertakes the main computing tasks, with 80% of the high-precision model rendering completed at the edge nodes. The AR glasses are only responsible for image overlay and posture tracking, significantly improving device battery life from the original 1.5 hours to meet all-day navigation needs. Furthermore, by offloading a large amount of computing to shared edge nodes, the redundant computing power consumption of independent high-precision rendering on each terminal is avoided, resulting in more significant energy-saving and emission-reduction benefits under large-scale application.

[0103] Existing technologies do not simultaneously possess both the core technologies of "LOD generation with cultural feature fidelity constraints" and "adaptive scheduling driven by end-to-end energy consumption budget," nor are they specifically designed for the cultural attributes of historical scenes. This invention solves the technical problem of balancing "cultural fidelity" and "energy efficiency" in VR / AR rendering of historical scenes by combining the two, possessing outstanding substantive features and significant progress.

[0104] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0105] The above description is merely a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural transformations made using the contents of the present invention's specification and drawings under the inventive concept of the present invention, or direct / indirect applications in other related technical fields, are included within the patent protection scope of the present invention.

[0106] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for adaptive LOD generation of historical scenes compatible with VR and AR, characterized in that, Includes the following steps: S1: Semantic classification and cultural feature annotation of historical scene assets: The historical scene 3D assets are divided into four semantic levels according to historical value: core cultural relics level, key building level, scene component level and environmental decoration level; for core cultural relics level assets, the pattern, inscription and damage feature areas are identified and pixel-level feature masks are generated and mapped to the cultural feature weights of the 3D mesh vertices. S2: Generation of multi-level LOD models constrained by cultural features: Based on a simplification strategy with differentiated semantic level settings, the QEM edge folding algorithm with cultural feature weights is used to simplify the mesh, and three anchor level models of LOD0, LOD2 and LOD4 are pre-generated, while edge folding record sequences are generated at the same time; among them, LOD0 is the highest precision and LOD4 is the lowest precision. S3: Energy Budget-Driven Adaptive LOD Scheduling: Collects terminal device battery power, real-time power consumption, and computing power level, combines user line of sight focus, line of sight distance, and asset semantic weight, calculates LOD recommendation level through accuracy scoring formula, and outputs final LOD scheduling instruction after boundary clamping; S4: Power-Aware Closed-Loop Rendering Optimization: Real-time monitoring of device frame rate, power consumption and temperature data to form a closed-loop feedback, dynamically adjust LOD level and rendering parameters, and maintain a dynamic balance between performance and power consumption. S5: Edge-to-device collaborative hierarchical rendering and image compositing: A hierarchical offloading strategy is adopted. High-precision LOD level and complex lighting calculations are rendered as video streams at edge nodes and pushed to the terminal, while medium and low-precision LOD levels are rendered locally on the terminal. The two parts of the image are composited and output through pose synchronization, depth occlusion judgment and reprojection correction.

2. The adaptive LOD generation method for historical scenes adapted to VR and AR according to claim 1, characterized in that, In step S1, the mapping method for cultural feature weights is as follows: For each grid vertex, the feature mask map is queried based on its texture coordinates; the weight of vertices that fall completely within the feature region is set to 3.0, the weight of vertices in the feature edge buffer region is set to 1.5, and the weight of vertices in the non-feature region is set to 1.0; the final simplified vertex weight is the product of the cultural feature weight and the semantic level weight.

3. The adaptive LOD generation method for historical scenes adapted to VR and AR according to claim 1, characterized in that, In step S2, the QEM edge folding algorithm incorporating cultural feature weights is specifically as follows: For the edge to be folded, after calculating the original QEM folding error, multiply it by the average simplified weight of the two vertices at the two ends of the edge to obtain the corrected folding cost; By folding edges in ascending order of folding cost, the edges of the feature regions become more difficult to simplify, thus forcibly preserving the accuracy of cultural features.

4. The adaptive LOD generation method for historical scenes adapted to VR and AR according to claim 1, characterized in that, Step S2 also includes runtime transitional LOD generation: Three anchor point levels, LOD0, LOD2, and LOD4, are pre-generated. The transitional levels, LOD1 and LOD3, are generated at runtime through vertex splitting operations. Based on the adjacent low-precision anchor point models, the vertex splitting is performed in reverse order from the edge folding record sequence to restore the target precision level.

5. The adaptive LOD generation method for historical scenes adapted to VR and AR according to claim 1, characterized in that, In step S3, the accuracy scoring formula is: in This is the baseline accuracy value corresponding to the computing power of the equipment. , , These are the normalized values ​​for semantic, line-of-sight, and power consumption constraints, respectively. , , These are the corresponding weighting coefficients; The final LOD level is obtained by rounding down and boundary clamping, and the value range is limited to 0 to 4.

6. The adaptive LOD generation method for historical scenes adapted to VR and AR according to claim 1, characterized in that, Step S3 incorporates three energy efficiency strategies that automatically switch based on battery level: When the battery level is above 60%, it enters performance mode, prioritizing visual accuracy across all scenarios. When the battery level is between 30% and 60%, it enters the balanced mode, which reduces the LOD level of non-sight focus areas by 1 level and the environmental decoration areas by 2 levels. When the battery level is less than 30%, it enters energy-saving mode, maintaining the original level only in the core cultural relic focus area, while the other areas are downgraded to the maximum extent, and secondary lighting effects are turned off.

7. The adaptive LOD generation method for historical scenes adapted to VR and AR according to claim 1, characterized in that, In step S5, the image compositing method for edge-level rendering is as follows: When rendering high-precision images at edge nodes, a depth map is output simultaneously, and the terminal generates a local depth cache. The depth values ​​are compared pixel by pixel to determine occlusion relationships. At the same time, the edge images are asynchronously reprojected and corrected based on the latest pose of the terminal. Finally, a complete image is output after depth testing and layered synthesis.

8. A historical scene adaptive LOD generation system adapted to VR and AR, characterized in that, include: The historical scene semantic asset library module is used to semantically classify 3D assets of historical scenes according to their historical value, complete cultural feature annotation and vertex weight mapping, and establish a reusable lightweight asset library. The multi-dimensional LOD hierarchical generation module is connected to the historical scene semantic asset library module. It is used to perform QEM mesh simplification with cultural feature weight constraints, generate multi-level anchor point LOD models and edge folding record sequences, and support runtime vertex splitting to generate transition levels. The adaptive scheduling engine module is connected to the multi-dimensional LOD level generation module and the power consumption-aware rendering optimization module, respectively. It is used to combine semantic weight, view distance, line focus, device computing power and battery power to dynamically determine the LOD level of each region based on the accuracy scoring formula. The power consumption-aware rendering optimization module is connected to the adaptive scheduling engine module to monitor device power consumption, frame rate and temperature in real time, forming a closed-loop feedback and adjusting the LOD level and rendering parameters accordingly. The edge-to-edge collaborative low-carbon rendering module is connected to the multi-dimensional LOD hierarchical generation module and the adaptive scheduling engine module, respectively. It is used to realize hierarchical offloading of high-precision content edge rendering and low-precision content local rendering, and to complete image synchronization, depth occlusion correction and compositing output.

Images