LOD data scheduling method based on interoperation

Abstract

The invention discloses an interoperation-based LOD data scheduling method, which comprises the following steps of: inputting an operation scene; extracting key content of the scene; constructing an operation scene factor; establishing a scheduling scene; setting a scheduling level; according to the method, the operation intention of a user is matched in real time through three modes of dynamic scene factor construction: dotted diffusion, viewpoint radiation, linear translation, strip-shaped extension in the movement direction, planar collapse and collapse from the boundary to the center, intelligent scheduling level linkage is adopted, six levels of bandwidth strategies including 0-5 levels are set, the fifth level is 100% bandwidth, the sixth level is 100% bandwidth, and the fifth level is 100% bandwidth. 0 level: 0% bandwidth is dynamically adjusted based on a real-time frame rate.

Description

Technical Field

[0001] This invention relates to the field of real-scene 3D technology, and more specifically to an interoperable LOD data scheduling method. Background Technology

[0002] Level of Detail (LOD) in Realistic 3D modeling is a technique that adjusts the rendering detail of objects based on their distance from the observer or camera, aiming to improve rendering efficiency and performance. In Realistic 3D modeling, it generates multiple versions of the model with varying levels of detail and selects the appropriate level of detail for rendering based on the observer's distance, thereby reducing computational resource consumption without sacrificing visual quality. This technique is widely used in urban planning, Geographic Information Systems (GIS), and virtual reality, effectively handling large-scale complex scenes and improving model loading speed and interactivity.

[0003] Traditional LOD data scheduling relies on computer vision, prioritizing the scheduling of real-world 3D tile data within the viewport. This approach suffers from several drawbacks. For example, scheduling all data at once using multi-threading can lead to high instantaneous network concurrency, resulting in incomplete model loading when network bandwidth is insufficient. Furthermore, the lack of focus in model scheduling results in a large amount of data without significant meaning being scheduled, crowding out channels for important data. Therefore, there is an urgent need to design an interoperable LOD data scheduling method to address these issues. Summary of the Invention

[0004] The purpose of this invention is to provide an interoperable LOD data scheduling method to address the aforementioned shortcomings in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: The interoperability-based LOD data scheduling method includes the following steps: S1 Input Operation Scenario: Acquire interactive scene data containing 3D models, viewpoint positions, and motion vectors; S2 extracts key scene content: Based on viewpoint position and motion vectors, it determines high-interest areas in the scene through entropy calculation, using the formula: E=−∑ n i=1n P i log2P i； Where E is the view entropy, Pi represents the pixel proportion of the i-th region in the field of view, and regions with entropy values ​​higher than the threshold θ are marked as key content; S3 constructs operational scene factors: Based on motion vectors, dynamic scheduling modes are generated, including point-like diffusion (radiating diffusion centered on the viewpoint), linear translation (extending in a strip along the direction of motion), and surface collapse (collapse from the edge of the scene to the center). S4 sets up a scheduling scenario: combining operation scenario factors and key content, divide the data into high, medium and low priority data blocks, and assign an initial LOD level; S5 sets scheduling levels: Defines a total of 6 bandwidth allocation strategies from level 0 to level 5: Level 5 (100% bandwidth), Level 4 (80%), Level 3 (60%), Level 2 (40%), Level 1 (20%), and Level 0 (0%). The level is dynamically adjusted according to the real-time frame rate F. S6 scene scheduling method: Based on scheduling level and operation scene factor, dynamically load or unload LOD model, with priority order as follows: point diffusion type > line translation type > surface collapse type.

[0006] Preferably, the operation scenario includes a virtual reality environment, a geographic information system, or an industrial simulation platform, and the scenario data is stored through distributed nodes to support concurrent access by multiple users; The scenario type must be compatible with heterogeneous platforms (such as VR headsets and GIS map engines). Distributed nodes use spatial block indexes (such as R* trees) to ensure that data blocks are quickly located according to viewpoint position; when multiple users access the site, priority sorting is synchronized through timestamps to avoid resource conflicts.

[0007] Preferably, the extraction of key scene content includes: High texture complexity regions are identified by convolutional neural networks and weighted by viewpoint entropy. The weight formula is W=αE+βT, where T is the texture complexity, and α and β are dynamic coefficients. When W>γ, it is marked as a key region and assigned the minimum LOD simplification rate.

[0008] Preferably, the point-like diffusion method specifically refers to: expanding in layers according to the spherical radius with the viewpoint as the center, allocating the highest scheduling level to models within the radius r, and attenuating the scheduling level according to distance outside r: Level outer =5×e −kr k is the attenuation coefficient, and the sphere center is updated to the latest viewpoint coordinates every frame.

[0009] Preferably, the linear translation method specifically involves: constructing a strip-shaped buffer zone along the direction of motion, where the buffer zone width d is positively correlated with the motion velocity v: d = μv + d0, where μ is a scaling factor; and reducing the LOD of the model outside the buffer zone according to the patch density, with a density threshold ρ = ρ max ×Level / 5.

[0010] Preferably, the planar collapse method specifically involves: collapsing layer by layer from the scene boundary toward the viewpoint, with the number of collapse layers N being negatively correlated with the frame delay Δt: N = ⌈λ / Δt⌉, where λ is the collapse rate, and during each layer of collapse, the LOD level of non-critical areas is forcibly reduced to 0.

[0011] Preferably, the bandwidth allocation at the scheduling level is achieved through dynamic weights: weights wp, wl, and ws are assigned to point, line, and area factors, satisfying wp + wl + ws = 1. The total bandwidth is split according to weights: Bp = B × wp, where B is the total bandwidth at the current level, and the weights are adjusted with the motion vector direction angle Φ: wp = 1 / (1 + e^(-1 / 2)). −Φ ).

[0012] Preferably, the LOD model loading includes transition region culling: when switching between adjacent LOD levels, a geometric deformation transition region is inserted, with the deformation δ=∥Vhigh−Vlow∥×sin(ωt), where V is the vertex coordinate and ω is the angular frequency. When δ<ϵ, the low-level model is removed to avoid visual jumps.

[0013] Preferably, the method further includes edge computing node collaboration: scheduling instructions are distributed to edge nodes, the nodes cache the LOD model library locally, and when the bandwidth level is ≤2, the local simplification engine is enabled to generate low-level models in real time, with a simplification rate η=1−Level×0.2.

[0014] Preferably, the method is executed via a GPU parallel pipeline: the scheduled task is split into compute shader thread groups, each group processing one data block, with the number of threads M = ⌈Block. size / 32⌉, and synchronize the priority queue via an atomic counter.

[0015] In the above technical solution, the interoperability-based LOD data scheduling method provided by the present invention has the following beneficial effects: (1) The present invention constructs dynamic scene factors: through three modes, namely point diffusion (viewpoint radiation), linear translation (movement direction band extension), and surface collapse (boundary collapse towards the center), it matches the user's operation intention in real time.

[0016] (2) The present invention features intelligent scheduling level linkage: setting a total of 6 bandwidth strategies from level 0 to level 5 (level 5: 100% bandwidth → level 0: 0% bandwidth), which are dynamically adjusted based on real-time frame rate.

[0017] (3) The present invention uses an edge node collaboration mechanism: scheduling instructions are distributed to edge nodes and LOD model libraries are cached locally; when the bandwidth is ≤ Level 2, the real-time simplification engine is enabled and the simplification rate η = 1 − Level × 0.2. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0019] Figure 1 This is a schematic diagram illustrating the steps of an embodiment of the LOD data scheduling method based on interoperability of the present invention. Detailed Implementation

[0020] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0021] like Figure 1 As shown, the LOD data scheduling method based on interoperability provided in this embodiment of the invention includes the following steps: S1 Input Operation Scenario: Acquire interactive scene data containing 3D models, viewpoint positions, and motion vectors; S2 extracts key scene content: Based on viewpoint position and motion vectors, it determines high-interest areas in the scene through entropy calculation, using the formula: E=-∑ n i＝1n P i log2P i Where E is the view entropy, Pi represents the pixel proportion of the i-th region in the field of view, and regions with entropy values ​​higher than the threshold θ are marked as key content; S3 constructs operational scene factors: Based on motion vectors, dynamic scheduling modes are generated, including point-like diffusion (radiating diffusion centered on the viewpoint), linear translation (extending in a strip along the direction of motion), and surface collapse (collapse from the edge of the scene to the center). S4 sets up a scheduling scenario: combining operation scenario factors and key content, divide the data into high, medium and low priority data blocks, and assign an initial LOD level; S5 sets scheduling levels: Defines a total of 6 bandwidth allocation strategies from level 0 to level 5: Level 5 (100% bandwidth), Level 4 (80%), Level 3 (60%), Level 2 (40%), Level 1 (20%), and Level 0 (0%). The level is dynamically adjusted according to the real-time frame rate F. S6 scene scheduling method: Based on scheduling level and operation scene factor, dynamically load or unload LOD model, with priority order as follows: point diffusion type > line translation type > surface collapse type.

[0022] Preferably, the operation scenario includes a virtual reality environment, a geographic information system, or an industrial simulation platform, and the scenario data is stored through distributed nodes to support concurrent access by multiple users; The scenario type must be compatible with heterogeneous platforms (such as VR headsets and GIS map engines). Distributed nodes use spatial block indexes (such as R* trees) to ensure that data blocks are quickly located according to viewpoint position; when multiple users access the site, priority sorting is synchronized through timestamps to avoid resource conflicts.

[0023] Preferably, the extraction of key scene content includes: High texture complexity regions are identified by convolutional neural networks and weighted by viewpoint entropy. The weight formula is W=αE+βT, where T is the texture complexity, and α and β are dynamic coefficients. When W>γ, it is marked as a key region and assigned the minimum LOD simplification rate.

[0024] Preferably, the point-like diffusion method specifically refers to: expanding in layers according to the spherical radius with the viewpoint as the center, allocating the highest scheduling level to models within the radius r, and attenuating the scheduling level according to distance outside r: Level outer =5×e −kr k is the attenuation coefficient, and the sphere center is updated to the latest viewpoint coordinates every frame.

[0025] Preferably, the linear translation method specifically involves: constructing a strip-shaped buffer zone along the direction of motion, where the buffer zone width d is positively correlated with the motion velocity v: d = μv + d0, where μ is a scaling factor; and reducing the LOD of the model outside the buffer zone according to the patch density, with a density threshold ρ = ρ max ×Level / 5.

[0026] Preferably, the planar collapse method specifically involves: collapsing layer by layer from the scene boundary toward the viewpoint, with the number of collapse layers N being negatively correlated with the frame delay Δt: N = ⌈λ / Δt⌉, where λ is the collapse rate, and during each layer of collapse, the LOD level of non-critical areas is forcibly reduced to 0.

[0027] Preferably, the bandwidth allocation at the scheduling level is achieved through dynamic weights: weights wp, wl, and ws are assigned to point, line, and area factors, satisfying wp + wl + ws = 1. The total bandwidth is split according to weights: Bp = B × wp, where B is the total bandwidth at the current level, and the weights are adjusted with the motion vector direction angle Φ: wp = 1 / (1 + e^(-1 / 2)). −Φ ).

[0028] Preferably, the LOD model loading includes transition region culling: when switching between adjacent LOD levels, a geometric deformation transition region is inserted, with the deformation δ=∥Vhigh−Vlow∥×sin(ωt), where V is the vertex coordinate and ω is the angular frequency. When δ<ϵ, the low-level model is removed to avoid visual jumps.

[0029] Preferably, the method further includes edge computing node collaboration: scheduling instructions are distributed to edge nodes, the nodes locally cache the LOD model library, and when the bandwidth level is ≤2, a local simplification engine is enabled to generate low-level models in real time, with a simplification rate η=1−Level×0.2. Preferably, the method is executed via a GPU parallel pipeline: the scheduled task is split into compute shader thread groups, each group processing one data block, with the number of threads M = ⌈Block. size / 32⌉, and synchronize the priority queue via an atomic counter.

[0030] Example 1 See Figure 1 The process involves inputting the operation scenario, extracting key content from the scenario, constructing operation scenario factors, building a scheduling scenario, setting the scheduling level, and generating a scenario scheduling method.

[0031] Input the operation scenario, display a 3D real-world view of a square and its surroundings; extract key content of the scenario, the square; construct operation scenario factors, point-to-point diffusion scheduling factors; build scheduling scenarios, single-point point-to-point diffusion scenarios; set scheduling levels; generate scenario scheduling methods.

[0032] Example 2 See Figure 1 The process involves inputting the operation scenario, extracting key content from the scenario, constructing operation scenario factors, building a scheduling scenario, setting the scheduling level, and generating a scenario scheduling method.

[0033] The input operation scenario is to display the real-world 3D scenery along both banks of the Haihe River in Tianjin; the key content of the scene is the Haihe River in Tianjin; the operation scenario factor is a linear translational scheduling factor; the scheduling scenario is constructed as a linear translational scheduling along the riverbank of the Haihe River, with the translation line following the center point diffusion scheduling factor, and the scheduling scenario is constructed in combination; the scheduling level is set; and the scene scheduling method is generated. The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. An interoperability-based LOD data scheduling method, comprising the following steps: S1 Input Operation Scenario: Acquire interactive scene data containing 3D models, viewpoint positions, and motion vectors; S2 extracts key scene content: Based on viewpoint position and motion vectors, it determines high-interest areas in the scene through entropy calculation, using the formula: E=−∑ n i=1n P i log2P i； Where E is the view entropy, Pi represents the pixel proportion of the i-th region in the field of view, and regions with entropy values ​​higher than the threshold θ are marked as key content; S3 constructs operational scene factors: Based on motion vectors, dynamic scheduling modes are generated, including point-like diffusion (radiating diffusion centered on the viewpoint), linear translation (extending in a strip along the direction of motion), and surface collapse (collapse from the edge of the scene to the center). S4 sets up a scheduling scenario: combining operation scenario factors and key content, divide the data into high, medium and low priority data blocks, and assign an initial LOD level; S5 sets scheduling levels: Defines a total of 6 bandwidth allocation strategies from level 0 to level 5: Level 5 (100% bandwidth), Level 4 (80%), Level 3 (60%), Level 2 (40%), Level 1 (20%), and Level 0 (0%). The level is dynamically adjusted according to the real-time frame rate F. S6 scene scheduling method: Based on scheduling level and operation scene factor, dynamically load or unload LOD model, with priority order as follows: point diffusion type > line translation type > surface collapse type.

2. The LOD data scheduling method based on interoperability according to claim 1, characterized in that, The operation scenarios include virtual reality environments, geographic information systems, or industrial simulation platforms, and the scenario data is stored through distributed nodes, supporting concurrent access by multiple users. The scenario type must be compatible with heterogeneous platforms. Distributed nodes use spatial block indexing to ensure that data blocks are quickly located according to viewpoint position; when multiple users access the site, priority sorting is synchronized through timestamps to avoid resource conflicts.

3. The LOD data scheduling method based on interoperability according to claim 2, characterized in that, The key content extracted from the scenario includes: High texture complexity regions are identified by convolutional neural networks and weighted by viewpoint entropy. The weight formula is W=αE+βT, where T is the texture complexity, and α and β are dynamic coefficients. When W>γ, it is marked as a key region and assigned the minimum LOD simplification rate.

4. The LOD data scheduling method based on interoperability according to claim 1, characterized in that, The point-like diffusion method specifically refers to: expanding in layers according to the spherical radius with the viewpoint as the center, assigning the highest scheduling level to models within the radius r, and attenuating the scheduling level according to distance outside r: Level outer =5×e −kr k is the attenuation coefficient, and the sphere center is updated to the latest viewpoint coordinates every frame.

5. The LOD data scheduling method based on interoperability according to claim 1, characterized in that, The linear translation method specifically involves: constructing a strip-shaped buffer zone along the direction of motion, where the buffer width d is positively correlated with the motion velocity v: d = μv + d0, where μ is a scaling factor; and reducing the LOD of the model outside the buffer zone based on patch density, with a density threshold ρ = ρ max ×Level / 5.

6. The LOD data scheduling method based on interoperability according to claim 5, characterized in that, The planar collapse method specifically refers to the collapse layer by layer from the scene boundary towards the viewpoint. The number of collapse layers N is negatively correlated with the frame delay Δt: N=⌈λ / Δt⌉, where λ is the collapse rate. During each layer collapse, the LOD level of non-critical areas is forcibly reduced to 0.

7. The interoperability-based LOD data scheduling method according to claim 2, characterized in that, The bandwidth allocation at the scheduling level is achieved through dynamic weights: weights wp, wl, and ws are assigned to point, line, and area factors, satisfying wp + wl + ws = 1. The total bandwidth is split according to the weights: Bp = B × wp, where B is the total bandwidth of the current level, and the weights are adjusted with the motion vector direction angle Φ: wp = 1 / (1 + e^(wp)). −Φ ).

8. The LOD data scheduling method based on interoperability according to claim 1, characterized in that, The LOD model loading includes transition region culling: when switching between adjacent LOD levels, a geometric deformation transition region is inserted, with the deformation δ=∥Vhigh-Vlow∥×sin(ωt), where V is the vertex coordinate and ω is the angular frequency. When δ<∈, the low-level model is removed to avoid visual jumps.

9. The LOD data scheduling method based on interoperability according to claim 1, characterized in that, The method also includes edge computing node collaboration: scheduling instructions are distributed to edge nodes, the nodes cache the LOD model library locally, and when the bandwidth level is ≤2, the local simplification engine is enabled to generate low-level models in real time, with a simplification rate η=1−Level×0.

2.

10. The LOD data scheduling method based on interoperability according to claim 1, characterized in that, The method is executed via the GPU parallel pipeline: the scheduled task is split into compute shader thread groups, each group processes one data block, and the number of threads M = ⌈Block. size / 32⌉, and synchronize the priority queue via an atomic counter.

Images