A method for rendering display analysis of public art models

Abstract

This application relates to a rendering and display analysis method for public art models, belonging to the field of image analysis and processing technology. By acquiring the 3D model of the public art and its display environment parameters, a comprehensive rendering quality score is calculated using a pre-set evaluation model. The score is based on the model's polygon count, texture loading time, and the ratio of real-time rendering frame rate to a threshold, combined with weighted coefficients. Based on the score results, the system maps the model to predefined levels such as high quality, balanced, or performance, and automatically executes corresponding preset optimization actions. The evaluation model can dynamically adjust the weighted coefficients according to environmental parameters such as terminal GPU performance, memory, and network bandwidth. Through quantitative evaluation, dynamic level mapping, and adaptive optimization strategies, it achieves an intelligent balance between rendering quality and display smoothness of the public art model in diverse terminal and network environments, aiming to ensure a stable and high-quality user experience on the online display platform.

Description

Technical Field

[0001] This application relates to the field of image analysis and processing technology, specifically to a rendering and display analysis method for public art models. Background Technology

[0002] In the field of digital public art display, especially web-based online 3D model rendering platforms, the core objective is to achieve high-quality, smooth, and stable visualization of art models across diverse user terminal devices and network environments. Traditional display solutions typically rely on fixed rendering pipelines or simple strategies based on limited hardware detection, making it difficult to dynamically balance visual quality and real-time performance in complex real-world application scenarios.

[0003] Traditional methods typically employ preset, static rendering quality levels, or rely solely on limited hardware identifiers (such as GPU model) for coarse grading. This approach fails to adequately consider the combined impact of multiple dynamic factors, including the real-time performance of the terminal device (such as GPU instantaneous load and available memory), network bandwidth fluctuations, and the complexity of the model itself (geometry and texture). The root cause lies in the lack of a real-time, multi-parameter fusion evaluation model capable of quantifying the "overall rendering quality under the current environment." Therefore, when device performance is insufficient or network conditions are poor, forcibly loading high-precision resources can easily lead to severe stuttering, latency, or even crashes; conversely, on high-performance devices, it may fail to fully utilize hardware capabilities to provide optimal visual fidelity, resulting in resource waste and a degraded user experience.

[0004] Furthermore, when the rendering scene changes due to user interaction (such as rapid rotation and scaling) or background task interference, a fixed strategy cannot respond to real-time fluctuations in performance metrics (such as sudden drops in frame rate or memory overflow). In addition, traditional methods typically employ a uniform optimization strategy for the entire model, failing to identify and respond to the differences in user attention to specific areas of the model (such as the core details of an artwork or inscriptions). Summary of the Invention

[0005] This application provides a rendering and display analysis method for public art models. By establishing quantitative quality assessment, dynamic level mapping, and adaptive optimization strategies, it achieves an intelligent balance between rendering quality and display smoothness of public art models in diverse terminal and network environments.

[0006] To achieve the above objectives, this application provides the following technical solution: This application provides a rendering and display analysis method for public art models, including the following steps: S1. Obtain the 3D model data of the public art to be displayed and its associated display environment parameters; S2. Based on the geometric complexity of the public art 3D model data, the texture data volume, and the display environment parameters, calculate the comprehensive rendering quality score Q through a preset rendering quality evaluation model. The calculation formula for the comprehensive rendering quality score Q is as follows: where P2 is the actual number of polygons of the model, P1 is the preset maximum number of polygon threshold acceptable in this display environment, T2 is the actual loading time of the model texture data, T1 is the preset target loading time threshold, F2 is the real-time rendering frame rate, F1 is the preset minimum smooth frame rate threshold, and α, β, γ are weight coefficients, and ; S3. According to the value of the comprehensive rendering quality score Q, map it to multiple predefined rendering quality levels. The multiple rendering quality levels at least include a high-quality level, a balanced level, and a performance level; S4. According to the rendering quality level at which the public art 3D model data is currently located, execute a preset display optimization action set corresponding to this level. The actions in the display optimization action set at least include dynamically adjusting one or more of the model geometric details, texture resolution, lighting calculation complexity, and shadow rendering.

[0007] Further, in step S2, the display environment parameters include the GPU performance level G3 of the terminal device, the available memory M3, and the network bandwidth B3; the preset rendering quality evaluation model dynamically adjusts the values of the weight coefficients α, β, γ according to the GPU performance level G3, the available memory M3, and the network bandwidth B3 of the terminal device.

[0008] Further, the dynamic adjustment rules of the weight coefficients α, β, γ are as follows: when G3 is lower than the first performance threshold G4, increase the value of γ; when M3 is lower than the first memory threshold M4, increase the value of β; when B3 is lower than the first bandwidth threshold B4, increase the value of β and decrease the value of α.

[0009] Further, in step S3, the multiple rendering quality levels are divided by two scoring thresholds Q3 and Q4, where Q3 > Q4: when Q ≥ Q3, it is determined as the high-quality level; when Q4 ≤ Q < Q3, it is determined as the balanced level; when Q < Q4, it is determined as performance level.

[0010] Further, in step S4, the preset display optimization action set executed for different rendering quality levels includes: If it is at the high-quality level, adopt the first optimization strategy. The first optimization strategy includes: loading the highest level of model detail (LOD0), using lossless or low compression rate textures, enabling real-time ray tracing or high-precision global illumination calculation, and rendering full-resolution shadows; If the model is in the balanced setting, the second optimization strategy is adopted. The second optimization strategy includes: dynamically switching the level of detail (LOD) of the model according to the view distance, using lossy compressed textures but maintaining high resolution of core visual feature areas, enabling baked lightmaps and simplified real-time lighting models, and rendering simplified shadows or screen space ambient occlusion (SSAO). If the system is in performance mode, a third optimization strategy is employed, which includes: forcing the use of the lowest level of detail (LSD). The model uses high-compression textures and may reduce color depth, disables complex lighting calculations and uses only basic ambient light and diffuse, and disables real-time shadow rendering.

[0011] Furthermore, in the second optimization strategy, the model's level of detail is dynamically switched based on the viewpoint distance. Specifically, this involves establishing a mapping function between the distance d between the viewpoint and the center of the model's bounding box and the LOD level to be loaded. Where f(d) is a piecewise function, and loading occurs when d is less than the first distance threshold d3. When d is in the range [d3, d4], LOD1 is loaded; when d is greater than the second distance threshold d4, LOD2 or a lower level is loaded.

[0012] Furthermore, after performing the display optimization action in step S4, the method also includes: real-time monitoring of performance indicator changes during the rendering process, wherein the performance indicators include at least the real-time frame rate F2 and the video memory usage rate V5; If it is detected that F2 is continuously lower than the preset alarm frame rate threshold F5 for a first time period Δt3, or V5 is continuously higher than the preset alarm memory threshold V6 for a second time period Δt4, then a rendering strategy downgrade instruction is triggered. The rendering strategy downgrade instruction is used to adjust the current rendering quality level to a lower level and re-execute step S4.

[0013] Furthermore, it also includes: during the display process, collecting user interaction behavior data, including the dwell time in specific feature areas of the model, the frequency of zoom operations, and the viewpoint switching path; based on the user interaction behavior data, predicting user interest areas through a machine learning model, and prioritizing the allocation of computing resources to model components within the predicted user interest areas in subsequent rendering to maintain or improve their rendering quality.

[0014] Furthermore, in step S1, obtaining the 3D model data of the public art to be displayed also includes a step of preprocessing the original model data: The model is meshed to generate multiple Level of Detail (LOD) versions, and the texture image is hierarchically compressed to generate multiple compression ratio versions. These preprocessing steps are performed offline before the model is loaded.

[0015] Furthermore, the method is applied to an online public art display platform on the Web. In step S2, when calculating the comprehensive rendering quality score Q, the real-time rendering frame rate F2 is obtained through WebGL or WebGPU API, and the actual loading time T2 of the model texture data is obtained through browser performance API monitoring.

[0016] This application provides a rendering and display analysis method for public art models. By acquiring 3D model data of the public art and its display environment parameters, and based on the model's geometric complexity, texture data volume, and environmental parameters, a pre-defined rendering quality assessment model is used to calculate a comprehensive rendering quality score. The score calculation comprehensively considers the ratio of the model's actual polygon count to a threshold, the actual texture loading time to a threshold, and the real-time rendering frame rate to a threshold, and introduces weighting coefficients for weighting. Based on the score value, it is mapped to a predefined rendering quality level, such as high quality, balanced quality, or performance quality. Subsequently, the system automatically executes a corresponding set of pre-defined display optimization actions based on the model's current quality level. This set includes dynamic adjustments to elements such as model geometric details, texture resolution, lighting computational complexity, and shadow rendering. The rendering quality assessment model can dynamically adjust the weighting coefficients in the scoring formula based on environmental parameters such as the terminal device's GPU performance, available memory, and network bandwidth. For example, it increases the weighting related to frame rate when GPU performance is low, and increases the weighting related to texture loading when memory or bandwidth is insufficient.

[0017] The rendering tiers are defined by pre-set scoring thresholds, with differentiated optimization strategies for each tier. The high-quality tier prioritizes the highest quality rendering, the balanced tier strikes a balance between visual quality and performance and supports dynamic switching of model detail levels based on viewpoint distance, while the performance tier prioritizes smoothness. After optimization actions are executed, the solution monitors performance metrics such as rendering frame rate and VRAM usage in real time. If these metrics remain abnormal, a rendering strategy downgrade is automatically triggered to maintain stability. Furthermore, the solution collects user interaction data, uses machine learning to predict user interest regions, and prioritizes allocating computing resources to these regions in subsequent rendering. Before model loading, the solution also includes offline preprocessing steps, such as generating multi-detail model versions and hierarchically compressed textures, providing a foundation for dynamic optimization. Attached Figure Description

[0018] Figure 1 A flowchart illustrating a rendering and display analysis method for public art models provided in this application. Detailed Implementation

[0019] The present application will now be described in further detail with reference to embodiments. It is to be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the scope of the application.

[0020] The technical solutions in the embodiments of this application will be clearly and completely described below. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0021] Traditional inpatient monitoring systems typically rely on monitoring a single physiological parameter or alarm mechanisms based on fixed thresholds. However, patients' physiological states exhibit high individual variability and dynamic changes, making it difficult for fixed threshold standards to adapt to the actual conditions of different patients at different stages of disease, easily leading to frequent false alarms or serious missed alarms. Furthermore, monitoring a single parameter often fails to comprehensively reflect the complex pathophysiological changes in patients, making it difficult to accurately identify early abnormal states and thus delaying optimal intervention. Therefore, how to integrate multiple physiological signals and combine them with intelligent models to achieve accurate, dynamic identification and graded response to abnormal states in hospitalized patients has become a pressing technical problem in the field of medical monitoring.

[0022] Based on the above issues, please refer to Figure 1 As shown in the figure, a rendering and display analysis method for public art models is provided in an embodiment of this application. The method includes the following steps: This application provides a rendering and display analysis method for public art models to address the technical challenge of adaptively balancing rendering quality and performance when displaying complex 3D public art models online under diverse terminal devices and network environments. This method establishes a quantitative rendering quality evaluation model and dynamically adjusts the rendering strategy based on the evaluation results, thereby maximizing the visual display effect while ensuring basic smoothness.

[0023] The method first executes step S1 to obtain the 3D model data of the public art to be displayed and its associated display environment parameters. Obtaining the 3D model data also includes a preprocessing step on the original model data, which is completed offline before the model is loaded. Specifically, the preprocessing includes mesh simplification of the model to generate multiple levels of detail (LOD) versions, and hierarchical compression of texture images to generate versions with various compression ratios.

[0024] Public art 3D model data typically contains high-precision mesh geometry information and multiple texture maps; directly loading these may place an excessive burden on the terminal device. The associated display environment parameters define the specific hardware and network conditions under which rendering occurs, and are key constraints for evaluating rendering feasibility.

[0025] The purpose of this preprocessing step is to prepare model data of different levels of detail and texture data of different qualities in advance, laying the foundation for dynamically selecting appropriate resources according to the running environment. This avoids time-consuming simplification and compression calculations during real-time rendering, thereby improving response speed and reducing real-time computing overhead.

[0026] This step is completed offline before model loading, aiming to build optional rendering resources adaptable to different performance levels. The model is meshed to generate multiple versions with varying levels of detail, resolving the issue of wasted rendering resources on long-distance or low-performance devices with a single high-poly model. Texture images are subjected to graded compression to generate multiple compression ratio versions, addressing the problems of slow loading and high VRAM consumption of large textures. By generating a multi-version resource library of models and textures through preprocessing, subsequent dynamic optimization strategies have a switchable material foundation, enabling rapid adaptation to different rendering quality requirements by calling different versions of resources without modifying the original assets.

[0027] Step S2 calculates the overall rendering quality score Q based on the geometric complexity, texture data volume, and display environment parameters of the acquired public art 3D model data using a pre-set rendering quality evaluation model. The formula for calculating the overall rendering quality score Q is as follows: .

[0028] In this formula, P2 represents the actual number of polygons in the model, and P1 represents the preset maximum acceptable polygon count threshold for this display environment. This ratio is used to evaluate geometric load. T2 represents the actual loading time of the model texture data, and T1 represents the preset target loading time threshold. This ratio is used to evaluate texture loading efficiency. F2 represents the real-time rendering frame rate, and F1 represents the preset minimum smooth frame rate threshold. This ratio is used to evaluate real-time rendering smoothness.

[0029] α, β, and γ are weighting coefficients, and The Q-value is used to adjust the relative importance of geometric quality, loading speed, and real-time performance in the overall score. This calculation formula normalizes multiple heterogeneous indicators by comparing the current state with an ideal threshold. The output Q-value directly reflects the degree to which the current rendering configuration meets the target performance. A higher Q-value indicates a more balanced rendering load across quality, loading speed, and performance, and more capacity to handle higher-quality rendering effects.

[0030] This calculation formula normalizes and weights the three dimensions of geometric load, loading efficiency, and rendering smoothness, thereby achieving a comprehensive quantitative evaluation of the load level and output quality of the current rendering task and providing accurate numerical basis for subsequent strategy selection.

[0031] The demonstration environment parameters specifically include the terminal device's GPU performance level G3, available memory M3, and network bandwidth B3. The pre-built rendering quality evaluation model dynamically adjusts the values ​​of weighting coefficients α, β, and γ based on these parameters. The specific rules for the dynamic adjustment of weighting coefficients α, β, and γ are as follows: when the GPU performance level G3 is lower than the first performance threshold G4, the value of γ is increased; when the available memory M3 is lower than the first memory threshold M4, the value of β is increased; when the network bandwidth B3 is lower than the first bandwidth threshold B4, the value of β is increased and the value of α is decreased. This dynamic adjustment rule aims to enable the calculation of the overall rendering quality score Q to adapt to the differences in terminal hardware performance and network conditions.

[0032] When GPU performance is weak, the weight γ of the real-time frame rate F2 is increased to make the score more reflective of smoothness. When memory or bandwidth is tight, the weight β of texture loading time T2 is increased to make the score more sensitive to resource loading efficiency, and the weight α of geometric load is appropriately reduced when bandwidth is insufficient. This guides the system to prioritize runnability in resource-constrained environments and ensures that the evaluation results fit the constraints of the actual operating environment.

[0033] The dynamic adjustment mechanism solves the problem that fixed weights cannot adapt to changing environments, enabling the evaluation model to intelligently adjust its evaluation focus based on device limitations. Specifically, when G3 is lower than the first performance threshold G1, the value of γ is increased. This is because GPU performance is a key factor determining the real-time frame rate F2. When performance is insufficient, ensuring the frame rate (i.e., the F2 / F1 term in the formula) is more urgent than pursuing high model detail (P1 / P2 term) or fast loading (T1 / T2 term). Therefore, increasing the weight of γ makes the frame rate performance have a greater impact on the final Q value, and the system will be more inclined to adopt a strategy that guarantees the frame rate.

[0034] When M3 is lower than the first memory threshold M1, the value of β is increased. This is because when available memory is scarce, loading large textures can easily lead to memory overflow. Therefore, more attention needs to be paid to texture loading time T2. By increasing the weight of β, loading efficiency has a higher weight in the score, guiding the system to prioritize texture versions with high compression ratios and low memory usage.

[0035] When B3 is lower than the first bandwidth threshold B1, the value of β is increased and the value of α is decreased. In scenarios with insufficient network bandwidth, texture data download becomes the main bottleneck, significantly increasing loading time T2, while the transmission of high-poly data (affecting P2) is also affected. By increasing the weight of β, loading performance is emphasized; at the same time, the weight of α is reduced to temporarily weaken the demand for geometric details, thereby guiding the system to prioritize loadability and basic smoothness when the network is limited, rather than pursuing the highest geometric accuracy.

[0036] To specifically illustrate the calculation process of the comprehensive rendering quality score Q, an implementation example is provided. Suppose in a certain display environment, the preset thresholds are as follows: P1 (the maximum acceptable number of polygons) is 2 million, T1 (the target loading time) is 3 seconds, and F1 (the minimum smooth frame rate) is 30 FPS. The actual parameters of the currently obtained model are: P2 (the actual number of polygons) is 1.5 million, T2 (the actual loading time) is 2.5 seconds, and F2 (the real-time frame rate) is 45 FPS. According to the current device environment parameters (for example, the GPU performance is medium G3 = G2, the memory is sufficient M3 > M1, and the bandwidth is good B3 > B1), the weight coefficients dynamically set by the rendering quality evaluation model are: α = 0.4, β = 0.3, γ = 0.3.

[0037] Substitute the above parameters into the calculation formula , and the calculation process is as follows: First, calculate each ratio, . Then perform weighted summation: . The calculated comprehensive rendering quality score Q is 1.343. This result is greater than 1, indicating that the current state exceeds the preset minimum requirements in all three dimensions, and the real-time frame rate performance is particularly outstanding. The system has the potential to adopt higher-quality rendering settings.

[0038] Mapping the value of the comprehensive rendering quality score Q to a predefined multiple rendering quality levels is the decision conversion link of this method. This step aims to convert the continuous score value into discrete and executable policy instructions to simplify the control logic. The multiple rendering quality levels at least include a high-quality level, a balanced level, and a performance level, corresponding to three core policy orientations of prioritizing quality, balancing quality and performance, and prioritizing performance, respectively.

[0039] It is divided by two score thresholds Q_high and Q_low (where Q_high > Q_low): when Q ≥ Q_high, it is determined to be the high-quality level; when Q_low ≤ Q < Q_high, it is determined to be the balanced level; when Q < Q_low, it is determined to be the performance level. The settings of the thresholds Q_high and Q_low are based on a large amount of test data and empirical values. For example, Q_high can be set to 1.2 and Q_low can be set to 0.9. This mapping mechanism simplifies the complex multi-dimensional evaluation results into clear level signals, providing clear trigger conditions for subsequent execution of specific and differentiated optimization actions, making the system behavior deterministic and predictable.

[0040] Step S3 maps the calculated overall rendering quality score Q to multiple predefined rendering quality levels, including at least high quality, balanced quality, and performance quality. These quality levels are divided using two scoring thresholds, Q3 and Q4, where Q3 is greater than Q4. Specifically, the mapping rule is as follows: when Q is greater than or equal to Q3, it is classified as high quality; when Q is greater than or equal to Q4 and less than Q3, it is classified as balanced quality; and when Q is less than Q4, it is classified as performance quality.

[0041] The purpose of this gear mapping step is to convert continuous quantitative scores into discrete quality levels with clear operational orientation, thereby providing a clear decision boundary for the subsequent execution of specific optimization action sets and establishing a bridge from evaluation to execution.

[0042] Step S4 executes a set of preset display optimization actions corresponding to the current rendering quality level of the public art 3D model data. This set of actions includes dynamically adjusting at least one or more of the following: model geometric details, texture resolution, lighting computation complexity, and shadow rendering. For high quality, the first optimization strategy is adopted, which includes loading the model's highest level of detail (LOD0), using lossless or low-compression textures, enabling real-time ray tracing or high-precision global illumination calculations, and rendering full-resolution shadows. This strategy aims to maximize the visual fidelity of the model and the realism of the rendering effect when hardware resources are sufficient.

[0043] This strategy is suitable for environments with ample hardware performance and good network conditions. Its core goal is to present art models with the highest visual fidelity while ensuring smooth interaction. The first optimization strategy includes: loading the model's highest level of detail (usually marked as LOD0), using lossless or low-compression textures, enabling real-time ray tracing or high-precision global illumination calculations, and rendering full-resolution shadows. Loading LOD0 models ensures that viewers can observe the most complete geometric details from any distance. Using high-quality textures avoids color distortion and detail blurring caused by compression. Enabling advanced lighting and shadow calculations maximizes the reproduction of the light and shadow interactions of materials, enhancing the model's three-dimensionality and realism. This series of actions works together to achieve the ultimate visual quality output within the limits of performance.

[0044] For the balanced setting, a second optimization strategy is adopted, which includes dynamically switching the level of detail of the model according to the view distance, using lossy compressed textures but maintaining high resolution in the core visual feature areas, enabling baked lightmaps and simplifying the real-time lighting model, and rendering simplified shadows or screen-space ambient occlusion.

[0045] Specifically, the model's level of detail (LOD) is dynamically switched based on the viewpoint distance. This involves establishing a mapping function L=f(d) between the distance *d* between the viewpoint and the center of the model's bounding box and the appropriate LOD level (LOD). Here, *f(d)* is a piecewise function; for example, LOD0 is loaded when *d* is less than a first distance threshold *d3*, LOD1 is loaded when *d* is between *d3* and *d4*, and LOD2 or a lower level is loaded when *d* is greater than a second distance threshold *d4*. This second optimization strategy aims to balance visual quality and performance consumption by effectively reducing rendering overhead within a range imperceptible to the user through distance-dependent LOD management, targeted texture compression, and simplified lighting models.

[0046] This strategy is applicable to most mainstream devices and aims to achieve the best balance between visual quality and interactive smoothness, which is the core application scenario of the method. The second optimization strategy includes: dynamically switching the level of detail (LOD) of the model based on the viewpoint distance, using lossy compressed textures while maintaining high resolution in core visual feature areas, enabling baked lightmaps and simplifying real-time lighting models, and rendering simplified shadows or screen-space ambient occlusion. Dynamically switching LOD based on the viewpoint distance is the key to this strategy, which is specifically achieved by establishing a mapping function L = f(d) between the distance d between the viewpoint and the center of the model bounding box and the LOD level to be loaded, where f(d) is a piecewise function.

[0047] For example, when the distance d is less than the near distance threshold (e.g., 5 meters) Load LOD0, when d is at When the distance d is within a certain range (e.g., 5 meters to 20 meters), LOD1 is loaded; when d is greater than the distance threshold d_far (e.g., 20 meters), LOD2 or a lower level is loaded. This mechanism significantly reduces the vertex processing burden on the GPU by automatically switching to a version with fewer faces for more distant parts of the model, while the loss of visual detail due to distance is not perceptible. Selective high-resolution textures are used, maintaining high resolution in the core visual feature areas of the model (e.g., the face of a sculpture, inscriptions), while using high-compression-rate textures in non-critical areas (e.g., the back of the base, internal structures), thus controlling texture memory usage while preserving key visual information for viewing.

[0048] By using baked lightmaps to replace some real-time lighting calculations, the lighting effects of static environments are pre-calculated and stored in the textures, and then sampled directly during rendering, significantly reducing real-time computational overhead. Shadows are simplified or SSAO technology is used, preserving the basic lighting atmosphere while avoiding the costly calculation of full-scene shadow mapping. These actions work together to systematically reduce the computational and storage pressure at each stage of the rendering pipeline without excessively sacrificing subjective visual quality.

[0049] For performance-critical settings, a third optimization strategy is employed. This strategy includes forcing the use of the lowest level of detail (LODn) model, using high-compression textures and potentially reducing color depth, disabling complex lighting calculations and using only basic ambient light and diffuse, and disabling real-time shadow rendering. The purpose of this strategy is to prioritize the basic smooth operation of the rendering process when resources are severely limited, sacrificing rendering quality to maintain a stable frame rate.

[0050] This strategy is suitable for low-performance terminals, old devices, or extremely poor network conditions. Its primary goal is to ensure basic interactivity and model recognizability. The third optimization strategy includes: forcing the use of the Lowest Level of Detail (LODn) model, using high-compression textures and potentially reducing color depth (e.g., from RGBA8888 to RGB565), disabling complex lighting calculations and using only basic ambient light and diffuse, and disabling real-time shadow rendering. Forcing the use of the LODn model minimizes the amount of geometric data.

[0051] Employing textures with high compression rates and even reduced color depth significantly reduces texture data size and memory usage, accelerating loading speed. Disabling all unnecessary lighting and shadow calculations simplifies shading calculations to a minimum. While these measures significantly reduce visual richness, they ensure that models can be loaded quickly and displayed at an interactive frame rate, avoiding stuttering, long wait times, or crashes, and guaranteeing the basic usability of the display functionality.

[0052] After performing the display optimization action in step S4, the process also includes real-time monitoring of performance metrics changes during the rendering process. These performance metrics include at least the real-time frame rate F2 and the video memory usage V5. If the real-time frame rate F2 is continuously lower than the preset alarm frame rate threshold F5 for a first time period Δt3, or the video memory usage V5 is continuously higher than the preset alarm video memory threshold V6 for a second time period Δt4, a rendering strategy downgrade instruction is triggered.

[0053] The rendering strategy degradation instruction is used to adjust the current rendering quality level to a lower level and re-execute step S4. This real-time monitoring and degradation mechanism is designed to cope with potential performance fluctuations or unexpected load increases during runtime. Through a performance feedback loop, it enables dynamic adaptive adjustment of the rendering strategy, preventing application stuttering or crashes due to continuous performance degradation, thus enhancing the system's robustness.

[0054] Performance metrics include at least real-time frame rate With video memory usage This monitoring mechanism establishes a closed-loop feedback loop, enabling the system to make dynamic corrections based on runtime conditions, rather than making one-off decisions. If a [missing information] is detected... The frame rate remains consistently below the preset alarm frame rate threshold. Reaching the first time interval Δt1, or The alarm memory threshold remains consistently higher than the preset value. Reaching the second time period If this happens, a rendering strategy downgrade instruction will be triggered.

[0055] The rendering strategy is triggered only after the value remains below or above the threshold for a sustained period, avoiding frequent policy fluctuations caused by momentary volatility. The rendering strategy downgrade instruction is used to adjust the current rendering quality level to a lower level and re-execute the corresponding optimization actions.

[0056] For example, it can downgrade from high quality to balanced, or from balanced to performance. Through this degradation mechanism, the system can proactively adopt more conservative rendering settings when it detects a trend of performance degradation, thereby stabilizing the frame rate or video memory usage within an acceptable range.

[0057] It also includes a step of collecting user interaction behavior data during the demonstration and using this data to predict regions of interest to optimize resource allocation. This step addresses the problem that traditional "one-size-fits-all" strategies in rendering optimization may ignore users' subjective focus, aiming to prioritize the allocation of limited computing resources to areas that users are truly interested in, thus achieving subjective perception optimization of resources. The collected user interaction behavior data includes the duration of dwell time in specific feature areas of the model, the frequency of zoom operations, and the path of viewpoint switching. This behavioral data indirectly reflects the degree of user attention to different model components.

[0058] Based on this data, a lightweight machine learning model (such as logistic regression or a shallow neural network) can predict the user's regions of interest. In subsequent rendering, computational resources are prioritized for model components within the predicted regions of user interest to maintain or improve their rendering quality. For example, even if the overall rendering is at a balanced level, higher-precision LODs or textures can still be loaded for predicted high-interest regions; conversely, lower-quality rendering settings can be used for completely uninterested regions. This mechanism transforms resource allocation from equalization to differentiation, improving the user's perceived rendering quality within the same overall performance budget.

[0059] This method can be applied to online public art display platforms on the web. In step S2, when calculating the overall rendering quality score Q, the real-time rendering frame rate F2 is obtained through the WebGL or WebGPU API, and the actual loading time T2 of the model texture data is obtained through browser performance API monitoring. The purpose of this specific implementation is to clarify the technical implementation path of the method in the web environment, utilize the standard graphics interface and performance monitoring interface provided by modern browsers to obtain key performance data, and ensure the feasibility and cross-platform consistency of the method.

[0060] The system first acquires preprocessed 3D model data of public art with multiple levels of detail and compression levels of texture, and collects environmental parameters such as GPU performance, memory, and network bandwidth of the terminal device. Subsequently, the rendering quality evaluation model dynamically weights and calculates a comprehensive rendering quality score Q by combining the model's geometric and texture data volume, real-time performance indicators, and environmental parameters.

[0061] The score Q is mapped to a predefined level, such as high quality, balanced, or performance. Based on the level, the system automatically calls the corresponding set of optimization strategies to adjust the model's geometric detail, texture quality, lighting model complexity, and shadow rendering methods in a coordinated manner.

[0062] During rendering, the system continuously monitors frame rate and video memory usage. If performance falls below a safe threshold, it automatically triggers a downgrade, switching to a more conservative rendering strategy. Simultaneously, the system analyzes user interaction behavior, predicts regions of interest, and prioritizes rendering resources for those regions.

[0063] Through the iterative execution and linkage of the above steps, this method enables public art 3D models to adaptively adjust the rendering pipeline in online environments such as the Web, based on real-time hardware resources, network conditions, and user concerns, while ensuring basic smooth operation and dynamically optimizing visual display quality.

[0064] This method begins with the preprocessing of the original model data and the acquisition of runtime environment parameters. Through a dynamically weighted comprehensive quality assessment model, it quantifies the state of three dimensions—geometric complexity, loading efficiency, and real-time performance—into a single score. The core of this scoring mechanism lies in the fact that its weighting coefficients can be dynamically adjusted according to environmental parameters such as GPU, memory, and network, making the evaluation criteria no longer static but closely related to device capabilities.

[0065] Subsequently, the scores are mapped to three distinct strategy tiers—high quality, balanced, and performance—through threshold comparisons. Each tier corresponds to a predefined set of collaborative optimization actions covering multiple rendering stages such as model LOD, textures, lighting, and shadows. This ensures the systematic and consistent execution of the strategies. Real-time performance monitoring creates a closed-loop feedback loop, automatically triggering strategy degradation when performance exceeds limits, thus guaranteeing the stability of the basic user experience.

Claims

1. A rendering, display, and analysis method for public art models, characterized in that, It includes the following steps: S1. Obtain the 3D model data of public art to be displayed and its associated display environment parameters; S2. Based on the geometric complexity of the 3D model data of public art, the amount of texture data, and the display environment parameters, calculate the comprehensive rendering quality score Q through a preset rendering quality evaluation model. The calculation formula of the comprehensive rendering quality score Q is: Where P2 is the actual number of polygons in the model, P1 is the preset maximum acceptable polygon count threshold for this display environment, T2 is the actual loading time of the model texture data, T1 is the preset target loading time threshold, F2 is the real-time rendering frame rate, F1 is the preset minimum smooth frame rate threshold, and α, β, and γ are weighting coefficients. ; S3. According to the value of the comprehensive rendering quality score Q, map it to multiple predefined rendering quality levels. The multiple rendering quality levels at least include a high-quality level, a balanced level, and a performance level; S4. According to the rendering quality level at which the 3D model data of public art currently is, execute a preset set of display optimization actions corresponding to this level. The actions in the set of display optimization actions at least include dynamically adjusting one or more of the model geometric details, texture resolution, lighting calculation complexity, and shadow rendering.

2. The method for rendering display analysis of public art models of claim 1, wherein, In step S2, the display environment parameters include the GPU performance level G3 of the terminal device, the available memory M3, and the network bandwidth B3; the preset rendering quality evaluation model dynamically adjusts the values of the weight coefficients α, β, and γ according to the GPU performance level G3, the available memory M3, and the network bandwidth B3 of the terminal device.

3. The method for rendering display analysis of public art models of claim 2, wherein, The dynamic adjustment rules of the weight coefficients α, β, and γ are: when G3 is lower than the first performance threshold G4, increase the value of γ; when M3 is lower than the first memory threshold M4, increase the value of β; when B3 is lower than the first bandwidth threshold B4, increase the value of β and decrease the value of α.

4. The method for rendering display analysis of public art models of claim 1, wherein, In step S3, the multiple rendering quality levels are divided by two scoring thresholds Q3 and Q4, where Q3 > Q4: when Q ≥ Q3, it is determined as the high-quality level; when Q4 ≤ Q < Q3, it is determined as the balanced level; when Q < Q4, it is determined as the performance level.

5. The method for rendering display analysis of public art models of claim 4, wherein, In step S4, the preset set of display optimization actions executed for different rendering quality levels includes: If in high quality, a first optimization strategy is employed, which includes loading the highest level of detail of the model (LOD) using lossless or low compression rate textures, enabling real-time ray tracing or high precision global illumination calculations, rendering full resolution shadows; If it is at the balanced level, adopt the second optimization strategy. The second optimization strategy includes: dynamically switching the level of detail (LOD) of the model according to the viewing distance, using lossy compressed textures but maintaining the high resolution of the core visual feature area, enabling baked light maps and simplifying the real-time lighting model, and rendering simplified shadows or screen space ambient occlusion (SSAO); If the system is in performance mode, a third optimization strategy is employed, which includes: forcing the use of the lowest level of detail (LSD). The model uses high-compression textures and may reduce color depth, disables complex lighting calculations and uses only basic ambient light and diffuse, and disables real-time shadow rendering.

6. The method for rendering display analysis of public art models of claim 5, wherein, In the second optimization strategy, the model's level of detail is dynamically switched based on the viewpoint distance. Specifically, this involves establishing a mapping function between the distance d between the viewpoint and the center of the model's bounding box and the appropriate LOD level L. Where f(d) is a piecewise function, and loading occurs when d is less than the first distance threshold d3. Loading when d is in the interval [d3, d4] When d is greater than the second distance threshold d4, load LOD2 or a lower level.

7. The rendering, display, and analysis method for public art models according to claim 1, characterized in that, After executing the display optimization actions in step S4, it further includes: real-time monitoring of the changes in performance metrics during the rendering process. The performance metrics at least include the real-time frame rate F2 and the video memory occupancy rate V5; If F2 is continuously lower than the preset alarm frame rate threshold and F5 reaches the first time period Or, if V5 continuously exceeds the preset alarm memory threshold, or if V6 reaches the second time period... If this occurs, a rendering strategy downgrade instruction is triggered. This instruction is used to adjust the current rendering quality level to a lower level and re-execute step S4.

8. The method for rendering display analysis of public art models of claim 7, wherein, It also includes: During the display process, collect user interaction behavior data. The user interaction behavior data includes the residence time in specific feature areas of the model, the scaling operation frequency, and the viewing angle switching path; based on the user interaction behavior data, predict the user's interested areas through a machine learning model, and preferentially allocate computing resources to the model components in the predicted user's interested areas in subsequent rendering to maintain or improve their rendering quality.

9. The method for rendering display analysis of public art models of claim 1, wherein, In step S1, the obtaining of the 3D model data of public art to be displayed further includes the step of preprocessing the original model data: The model is meshed to generate multiple Level of Detail (LOD) versions, and the texture image is hierarchically compressed to generate multiple compression ratio versions. These preprocessing steps are performed offline before the model is loaded.

10. The method for rendering display analysis of public art models of claim 1, wherein, The method is applied to an online public art display platform on the Web. In step S2, when calculating the comprehensive rendering quality score Q, the real-time rendering frame rate F2 is obtained through WebGL or WebGPU API, and the actual loading time T2 of the model texture data is obtained through browser performance API monitoring.

Images