Large-scale 3D model dynamic rendering system based on multi-thread mechanism
Through multi-threading mechanism and cache sharing optimization, the problem of high rendering latency on mobile devices is solved, efficient 3D model rendering on low-performance devices is achieved, and it adapts to heterogeneity and network conditions, improving rendering efficiency and cache management.
Patent Information
- Application Number
- CN202510731970.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-03
- Publication Date
- 2025-10-17
AI Technical Summary
The heterogeneity of mobile devices leads to high rendering latency for MW3D applications on low-performance devices. Remote rendering calculations cause transmission pressure and browser memory challenges, making it difficult to achieve real-time and efficient 3D model rendering.
A dynamic rendering system based on a multi-threaded mechanism is adopted. Through the collaborative work of the cloud server layer, edge service layer and terminal device layer, the 3D model is subdivided into small blocks, and the visible blocks in the current viewport are selectively loaded and rendered. The multi-threaded communication and cache sharing mechanism are used to achieve rendering calculation offloading and optimization.
Significantly reduce rendering calculation complexity and memory usage, improve rendering efficiency and cache management efficiency, achieve low-latency and efficient 3D model rendering, and adapt to different device performance and network conditions.
Smart Images
Figure CN120807736A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of scene rendering, and particularly relates to a large 3D model dynamic rendering system based on a multi-thread mechanism. BACKGROUND
[0002] With the development of 5G communication technology, mobile augmented reality (AR), virtual reality (VR) and 3D applications have achieved remarkable success on mobile devices such as mobile phones and smart glasses. As a new type of information interaction mode, 3D model interaction can provide more intuitive and efficient information presentation than traditional 2D methods. Mobile Web 3D applications (MW3D) can quickly establish three-dimensional information interaction through various mobile devices at any time and anywhere due to their universality and convenience.
[0003] The core of MW3D service is 3D model / scene rendering, but high resource consumption and unacceptable delay make it difficult to achieve satisfactory real-time experience for large-scale scene MW3D applications. The heterogeneity of mobile devices (computing power, form, network access, etc.) can cause service differences, especially in the initialization rendering calculation delay: high-performance devices (such as new mobile phones) have strong rendering computing power and low delay, but they lack portability and endurance; low-performance devices (such as smart glasses) are the opposite. This forces users to face a dilemma: (1) sacrificing portability to choose high-performance devices; (2) accepting high delay to choose portable devices.
[0004] To make up for the rendering defects of low-performance devices, the rendering calculation of other computing nodes needs to be offloaded. The current mainstream scheme performs rendering through remote computing nodes and returns the results in the form of multimedia streams (such as pictures / videos). However, due to the stereoscopic presentation characteristics, the MW3D rendering data is huge, which brings huge transmission pressure to the remote server. In large-scale MW3D or high-concurrency scenarios, massive downlink rendering data can easily cause congestion, and the browser memory also faces serious challenges. SUMMARY
[0005] In view of the above problems, the application provides a large 3D model dynamic rendering system based on a multi-thread mechanism.
[0006] To solve the above technical problems, the application adopts the following technical solutions: A large 3D model dynamic rendering system based on a multi-thread mechanism includes a cloud server layer, an edge service layer and a terminal device layer, In the cloud server layer, based on the geometric topology of the mesh 3D model and the visual features of the scene, the mesh 3D model is subdivided into smaller model blocks, and a model block sequential loading method is adopted, so that only the model blocks visible in the current viewport are selectively loaded and rendered, and the invisible model blocks are temporarily ignored, while the visible model blocks are preferentially rendered according to their visibility and user viewpoint; When the user initializes the service, the cloud server layer network module provides continuous data service to the edge server in the form of multi-thread multi-modal stream according to the model block number sequence of each 3D model and the pre-defined real-time dynamic scheduling algorithm; In the mobile edge layer, transmission sequence support is established for 3D model blocks, including a cooperative real-time scheduling module, a multi-thread data scheduling module and a cooperative data caching module. The cooperative real-time scheduling module determines the allocation scheme of the rendering calculation task based on the dynamic request of the client and the visual perception characteristics, and transmits the task allocation parameters to the cloud server and the mobile device. On the connection channel between the mobile device layer and the edge service layer, a multi-thread communication link based on the HTTP / 2 protocol is established, and rendering data is continuously transmitted to the mobile device through a multi-thread asynchronous communication mode. The cooperative data caching module utilizes the caching resources of the mobile edge server to establish a real-time dynamic caching sharing mechanism between the mobile device web browser cache and the mobile edge service cache. In the terminal device layer, rendering calculation and rendering data management functions for large-scale 3D models are provided. The terminal device adopts a multi-thread parallel processing mode and performs rendering calculation tasks according to the 3D model block transmission sequence provided by the edge server. After service initialization is completed, the terminal device visualizes the large-scale 3D model based on the visual features of the field of view angle, and establishes a rendering calculation offloading mechanism for the remote server, decomposes the rendering calculation task into multiple sub-tasks, and creates a corresponding number of independent threads in the terminal browser to execute these sub-tasks in parallel.
[0007] In one possible implementation, the subdividing of the mesh 3D model into smaller model blocks includes analyzing the entire 3D model based on image features or texture information, identifying surface or region fragments thereof, converting the model into a graph structure, and performing image region segmentation using a min-cut max-flow algorithm, thereby dividing the large-scale 3D model into multiple 3D model blocks.
[0008] In a possible implementation, the continuous transmission of rendering data to the mobile device through the multi-thread asynchronous communication mode comprises: after the HTTP / 2 connection is established, the mobile terminal creates multiple streams in the connection, each stream represents an independent request / response channel, supports concurrent transmission of multiple requests and responses, the mobile browser determines the 3D model blocks and corresponding resolution sequence parameters according to the field of view range defined by the user, initiates a request / creates a new stream and allocates a unique identifier on the remote server, the request can be divided into one or more frames for transmission according to needs, the frame is the minimum transmission unit of HTTP / 2, and is composed of a frame header and a frame payload, in this process, multiple streams are transmitted through the same connection, and concurrent transmission of requests and responses is realized without waiting for a previous request response to be completed; the client sends an HTTP request frame through the created stream, the HTTP request frame includes encapsulated request method, uniform resource identifier URI, and request header information; the server returns an HTTP response frame including response status code, response header, and response body after processing the request, and divides the generated response into one or more frames and sends the response back to the client according to the corresponding stream identifier.
[0009] In a possible implementation, the terminal device adopts a multi-thread parallel processing mode, which comprises: in the data receiving and caching stage, the browser synchronously enables multiple loading threads to receive and store 3D model block data from the remote server.
[0010] In a possible implementation, the terminal device adopts a multi-thread parallel processing mode, which comprises: in the data merging and rendering stage, the main thread delivers the data processed by the loading thread to the rendering engine for final rendering and display.
[0011] In a possible implementation, the mobile device web browser cache is implemented in the following manner: the field of view FOV information of the user is captured to optimize cache management, the current field of view range of the user is perceived through sensor technology or browser interface detection of the visual behavior of the user, a cache eviction strategy based on priority is adopted, the cache priority of data outside the field of view range is reduced to release cache space for data about to enter the field of view range, for foreground data in the field of view FOV region, a prefetching and depth-first strategy is adopted to optimize cache management, and the prefetching strategy preloads data that is likely to enter the field of view by analyzing user behavior and data reference mode.
[0012] In a possible implementation, a linked list method is adopted to optimize the mobile web browser cache.
[0013] In a possible implementation, the adoption of the linked list method to optimize the mobile web browser cache specifically comprises: Let the current user's view S within the 3D model block set as: Block_FOV∈S = [blocks-m,...,blocks,..., blocks+m], wherein the model block block_i corresponding to the view area is recorded as area_i, and m is a positive integer; The rendering data of the model block in the view S is represented as: render(i, res_i∈S), wherein res_i∈S represents the resolution required by the model block under the current view S (res_i∈S ∈ Res); The rendering data set in the current view is represented as: Ren_FOV=S = [render(s-m, res_(s-m)∈S),..., render(s, res_s∈S),..., render(s+m, res_(s+m)∈S)] When the user's perspective changes, resulting in a displacement of the view center Δs, the corresponding view becomes ΔS; at this time, the rendering resolution of each model block in the view will change, and the new resolution of the i-th model block is recorded as res_i∈ΔS, and the resolution change is represented as: Δres_i∈ΔS = res_i∈ΔS - res_i∈S; The sequential access feature of the linked list structure allows it to quickly locate the head data without global traversal, thereby reducing data search time. According to the visual characteristic analysis, the rendering data render(s, res_s∈S) corresponding to the view center point s has the highest interaction probability, so it is used as the head of the linked list to realize priority access when the 3D model application is initialized. Subsequently, the left region data render(s-1, res_(s-1)∈S) and the right region data render(s+1, res_(s+1)∈S) are linked as successor nodes. By predicting the trend of user perspective changes, the rendering data of high-interaction-probability adjacent model blocks are constructed as linked list nodes. Starting from the center point render(s, res_s∈S), sequential traversal is performed until all model block rendering data in the current view S are stored in the cache linked list. When the mobile browser needs to obtain rendering data from the cache: start searching from the head of the linked list, and preferentially draw the 3D scene of the region corresponding to the view center point; sequentially traverse the linked list, and draw each node data on both sides of the center model block in turn; loop until the rendering and presentation of all 3D model blocks in the current view are completed; When the user's viewpoint changes, the current field of view FOV will change accordingly, and due to the continuity of the user's visual features, the transition of the current field of view will occur, and the cached data will change as follows: when the viewpoint moves to the left: the rendering data render(s-1, res_(s-1) e S-1) of block_{s-1} becomes the center area of the current field of view, the chain table connection relationship is modified, the mobile device sets the forward pointer of render(s-1, res_(s-1) e S-1) to null, so as to become the new head node of the chain table, the head value of render(s-2, res_(s-2) e S-1) is modified to point to the corresponding node of render(s-1, res_(s-1) e S-1), the forward pointer of render(s, res_s e S-1) is set to point to the corresponding node of render(s-2, res_(s-2) e S-1), and the process continues until all display area data in the current field of view S-1 is stored in the chain table, and the same processing procedure is adopted when the viewpoint moves to the right.
[0014] In a possible implementation, the working process of the cooperative data caching module is as follows: the multi-resolution rendering data in the mobile edge server cache is stored in a tree structure, each tree is represented as Tree_i, i e Block, corresponding to the rendering data of a 3D model block, and stores multi-resolution data [render(i, res0), render(i, res1),..., render(i, resm)], the root node of the tree stores the current field of view resolution rendering data render(i, rescurrent) requested by the mobile web browser at present, the child nodes of the root node store the rendering data of the adjacent resolution of the current field of view area: the left child node stores the rendering data corresponding to the resolution rescurrent-1, and the right child node stores the rendering data corresponding to the resolution rescurrent+1, and each node further stores the address of the connected parent node and child node, so as to realize fast traversal of the tree, and when the user's visual angle changes, the traversal pointer of the tree structure is written into the root node.
[0015] The present application has at least the following beneficial effects: The model blocks visible in the current viewport are selectively loaded and rendered, the invisible blocks are temporarily ignored, and the visible blocks are preferentially rendered according to the visibility and the user's viewpoint, and the division based on the visual features and the task priority significantly reduces the calculation complexity and the memory occupation during the rendering operation.
[0016] The cooperative data caching module of the mobile edge layer makes full use of the rich cache resources of the mobile edge server, and establishes a real-time dynamic cache sharing mechanism between the mobile device web browser cache and the mobile edge server cache, so that the cache management efficiency and response speed are significantly improved.
[0017] (3) By introducing a multi-thread concurrent mechanism, rendering calculation tasks can be processed synchronously on the multi-core processor of the terminal device, thereby significantly improving the rendering calculation efficiency. Meanwhile, for the FOV visual characteristics, the system constructs a visual association stack management mechanism in the browser cache. In addition, a sharing mechanism with the edge server cache is established for low-resolution visual areas. By comprehensively using the viewpoint prediction technology, the multi-thread communication mechanism and the user view interaction technology, the system realizes efficient transmission of high-resolution rendering data. BRIEF DESCRIPTION OF DRAWINGS
[0018] Figure 1 A principle block diagram of a large 3D model dynamic rendering system based on a multi-thread mechanism according to an embodiment of the present application; Figure 2 A diagram for verifying a large-scale 3D model selected in a specific application example of the present application. DETAILED DESCRIPTION
[0019] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.
[0020] Referring to Figure 1 , a principle block diagram of a large 3D model dynamic rendering system based on a multi-thread mechanism according to an embodiment of the present application is shown, including a cloud server layer, an edge service layer and a terminal device layer, In the cloud server layer, based on the geometric topology structure of the grid 3D model and the visual characteristics of the scene, the grid 3D model is subdivided into smaller model blocks, a model block sequential loading method is adopted, and only the model blocks visible in the current viewport are selectively loaded and rendered, the invisible model blocks are temporarily ignored, and the visible model blocks are preferentially rendered according to their visibility and user viewpoint; When the user initializes the service, the cloud server layer network module provides continuous data service to the edge server in the form of multi-thread multi-modal flow according to the model block number sequence of each 3D model and the pre-defined real-time dynamic scheduling algorithm; In the mobile edge layer, transmission sequence support is established for the 3D model blocks, including a cooperative real-time scheduling module, a multi-thread data scheduling module and a cooperative data caching module, the cooperative real-time scheduling module determines a distribution scheme of rendering calculation tasks based on dynamic requests of clients and visual perception characteristics, and transmits task distribution parameters to cloud servers and mobile devices; On the connection channel between the mobile device layer and the edge service layer, a multi-thread communication link based on the HTTP / 2 protocol is established, and rendering data is continuously transmitted to the mobile device through a multi-thread asynchronous communication mode; The cooperative data caching module utilizes the caching resources of the mobile edge server to establish a real-time dynamic caching sharing mechanism between the mobile device web browser cache and the mobile edge service cache; In the terminal device layer, rendering calculation and rendering data management functions for large-scale 3D models are provided, the terminal device adopts a multi-thread parallel processing mode, performs rendering calculation tasks according to the 3D model block transmission sequence provided by the edge server; after service initialization is completed, the terminal device performs visual partitioning on the large-scale 3D model based on the visual features of the field of view angle, and accordingly establishes a rendering calculation offloading mechanism facing the remote server, decomposes the rendering calculation task into multiple subtasks, and creates a corresponding number of independent threads in the terminal browser to execute these subtasks in parallel.
[0021] In another embodiment of the application, subdividing the grid 3D model into smaller model blocks comprises: analyzing the entire 3D model based on image features or texture information, identifying surface or regional fragments thereof; converting the model into a graph structure, and performing image region segmentation using a minimum cut maximum flow algorithm, thereby dividing the large-scale 3D model into multiple 3D model blocks.
[0022] In another embodiment of the present application, the continuous transmission of rendering data to the mobile device by the multi-thread asynchronous communication mode comprises: after the HTTP / 2 connection is established, the mobile terminal creates multiple streams in the connection, each stream representing an independent request / response channel, supporting the concurrent transmission of multiple requests and responses, the mobile browser determines the 3D model blocks and their corresponding resolution sequence parameters according to the user-defined field of view range, initiates a request / creates a new stream and allocates a unique identifier on the remote server, the request can be divided into one or more frames for transmission as needed, the frame is the minimum transmission unit of HTTP / 2, which is composed of a frame header and a frame payload, in this process, multiple streams are transmitted through the same connection to realize the concurrent transmission of requests and responses without waiting for the completion of the previous request response; the client sends an HTTP request frame through the created stream, the HTTP request frame includes the information encapsulating the request method, uniform resource identifier URI, and request header; the server returns an HTTP response frame containing the response status code, response header, and response body after processing the request, and the server divides the generated response into one or more frames and sends them back to the client according to the corresponding stream identifier. Based on the above multi-thread communication mode, different 3D model blocks provide rendering data services for mobile devices through continuous multi-threading during the initialization process of large-scale 3D models; when the user's perspective changes, the mobile browser uses the multi-threading mechanism to realize the sharing of rendering data between the mobile edge server caches.
[0023] In another embodiment of the present application, the terminal device adopts a multi-thread parallel processing mode, which comprises: in the data receiving and caching stage, the browser synchronously enables multiple loading threads to receive and store 3D model block data from the remote server. It can also include: in the data merging and rendering stage, the main thread delivers the data processed by the loading thread to the rendering engine for final rendering and display.
[0024] In another embodiment of the present application, the mobile device web browser cache is implemented by the following methods: the cache management is optimized by capturing the field of view FOV information of the user, the user's current field of view range is perceived through sensor technology or browser interface detection of the user's visual behavior, a priority-based cache eviction policy is adopted, the cache priority of data outside the field of view range is reduced to release cache space for data about to enter the field of view range, for the foreground data in the field of view FOV area, a prefetching and depth-first strategy is used for optimized cache management, the prefetching strategy preloads the data that may enter the field of view by analyzing the user behavior and data reference mode.
[0025] In another embodiment of the present application, a linked list method is used for mobile web browser cache optimization, which specifically comprises: Let the current user's view S within the 3D model block set as: Block_FOV∈S = [blocks-m,...,blocks,..., blocks+m], wherein the model block block_i corresponding to the view area is recorded as area_i, and m is a positive integer; The rendering data of the model block in the view S is represented as: render(i, res_i∈S), wherein res_i∈S represents the resolution required by the model block under the current view S (res_i∈S ∈ Res); The rendering data set in the current view is represented as: Ren_FOV=S = [render(s-m, res_(s-m)∈S),..., render(s, res_s∈S),..., render(s+m, res_(s+m)∈S)] When the user's perspective changes, resulting in a displacement of the view center Δs, the corresponding view becomes ΔS; at this time, the rendering resolution of each model block in the view will change, and the new resolution of the i-th model block is recorded as res_i∈ΔS, and the resolution change is represented as: Δres_i∈ΔS = res_i∈ΔS - res_i∈S; The sequential access feature of the linked list structure allows it to quickly locate the head data without global traversal, thereby reducing data search time. According to the visual characteristic analysis, the rendering data render(s, res_s∈S) corresponding to the view center point s has the highest interaction probability, so it is used as the head of the linked list to realize priority access when the 3D model application is initialized. Subsequently, the left region data render(s-1, res_(s-1)∈S) and the right region data render(s+1, res_(s+1)∈S) are linked as successor nodes. By predicting the trend of user perspective changes, the rendering data of high-interaction-probability adjacent model blocks are constructed as linked list nodes, starting from the center point render(s, res_s∈S) and sequentially traversing until all model block rendering data in the current view S are stored in the cache linked list. When the mobile browser needs to obtain rendering data from the cache: start searching from the head of the linked list, and preferentially draw the 3D scene of the region corresponding to the view center point; sequentially traverse the linked list, and draw each node data on both sides of the center model block in turn; loop until the rendering and presentation of all 3D model blocks in the current view are completed; When the user's viewpoint changes, the current field of view FOV will change accordingly. Since the continuity of the user's visual features determines the transition of the current field of view, the cached data will change as follows: When the viewpoint moves to the left: the rendering data render(s-1, res_(s-1)∈S-1) of block_{s-1} becomes the center area of the current field of view. The linked list connection relationship is modified. The mobile device sets the forward pointer of render(s-1, res_(s-1)∈S-1) to null, making it the new head node of the linked list, and modifies the head value of render(s-2, res_(s-2)∈S-1) to point to the node corresponding to render(s-1, res_(s-1)∈S-1). The forward pointer of render(s, res_s∈S-1) is set to point to render(s-2, res_(s-2)∈S-1), and the process continues until all display area data within the current view S-1 are stored in the linked list. The same processing flow is used when the viewpoint moves right.
[0026] In another embodiment of the present invention, the working process of the collaborative data caching module is as follows: the multi-resolution rendering data in the mobile edge server cache is stored in a tree structure, each tree is represented as Tree_i, i∈Block, corresponding to the rendering data of a 3D model block, and storing multi-resolution data [render(i,res0),render(i,res1),...,render(i,resm)], the root node of the tree stores the current field of view resolution rendering data render(i,rescurrent) currently requested by the mobile web browser, and the child nodes of the root node store rendering data of resolutions adjacent to the current field of view area: the left child node stores rendering data corresponding to the resolution rescurrent-1, and the right child node stores rendering data corresponding to the resolution rescurrent+1. Each node also stores the addresses of its connected parent node and child node to achieve fast traversal of the tree. When the user's perspective changes, the traversal pointer of the tree structure is written to the root node.
[0027] To verify the effectiveness of the method of the embodiment of the present invention, a collaborative computing environment consisting of three PCs was established, including a cloud server, an edge server, and a simulated mobile device. The cloud server was equipped with a 2.9 GHz sixteen-core Intel CPU and 64 GB of memory. The edge server was equipped with an x64 CPU, an i7-8750H @ 2.21 GHz processor, 8 GB of memory, and a GeForce GTX 3080 graphics processing unit. The virtual machine used for mobile device simulation was configured with a simulated CPU frequency of 1.5 GHz and 8 GB of memory. The 64-bit Google Chrome (version 79.0.3945.130) was used as the browser to simulate a mobile web browser. The Google Chrome developer tools and Wonder Shaper were used to configure different client computing resources to ensure stable network conditions.
[0028] Use large-scale 3D models for verification, such as Figure 2 As shown, the model detail level is high, the file size is 82.27MB, and it contains 2,963,572 triangles and 1,513,188 vertices. Each web browser needs to display the 3D model image from a specific angle. The viewing angle is set to ±105 degrees during the verification process.
[0029] In poor network conditions (B < 200 Mbps), this clear The method of the embodiment has significantly better performance than the traditional mechanism because the asynchronous communication method based on visual features effectively alleviates the low bandwidth limitation.
[0030] A dynamic linked list data management mechanism based on visual features and a mobile edge server cache sharing solution are used to provide real-time and stable rendering data services for interactive mobile web browsers. Two main dimensions are verified: (1) the smoothness of MW3D interactions supported by local browser cache; and (2) the efficiency of user interaction responses under different service environments. Dynamic linked list management optimized by visual features can improve data retrieval efficiency, achieve higher frame rates and more stable rendering performance, and meet users' demand for smooth interaction in complex 3D scenes.
[0031] The edge cache response efficiency was verified by simulating a 30° viewpoint offset scenario for a mobile web browser. Compared with the remote server solution and the mobile edge server cache sharing solution, the edge cache sharing solution can provide rendering data with lower latency. When the user's viewpoint changes significantly, the traditional method causes a surge in computing pressure due to continuous data requests. The present invention implements this through a tree-like storage structure, which can progressively traverse and provide continuous resolution data, reducing the demand for stepped resolution data.
[0032] It is to be understood that the example embodiments described herein are illustrative rather than limiting. Although one or more embodiments of the application are described in connection with the accompanying drawings, it will be understood that various modifications in form and detail can be made without departing from the spirit and scope of the application as defined by the appended claims.
Claims
1. A large-scale 3D model dynamic rendering system based on a multi-threaded mechanism, characterized in that: Including cloud server layer, edge service layer and terminal device layer, In the cloud server layer, the mesh 3D model is subdivided into smaller blocks based on its geometric topological structure and the visual characteristics of the scene. A sequential block loading method is used to selectively load and render only the model blocks visible in the current viewport. Invisible model blocks are temporarily ignored, while visible model blocks are rendered first based on their visibility and user viewpoint. When the user initializes the service, the cloud server layer network module provides continuous data services to the edge server in the form of multi-threaded multimodal streaming according to the model block number sequence of each 3D model and the predefined real-time dynamic scheduling algorithm; In the mobile edge layer, a transmission sequence is established for 3D model blocks, including a collaborative real-time scheduling module, a multi-threaded data scheduling module, and a collaborative data caching module. The collaborative real-time scheduling module determines the allocation scheme for rendering computing tasks based on client dynamic requests and visual perception characteristics, and transmits the task allocation parameters to the cloud server and mobile devices. A multi-threaded communication link based on the HTTP / 2 protocol is established on the connection channel between the mobile device layer and the edge service layer, and rendering data is continuously transmitted to the mobile device through multi-threaded asynchronous communication. The collaborative data caching module uses the cache resources of the mobile edge server to establish a real-time dynamic cache sharing mechanism between the mobile device web browser cache and the mobile edge service cache; In the terminal device layer, it is used to provide rendering calculation and rendering data management functions for large-scale 3D models. The terminal device adopts a multi-threaded parallel processing mode and performs rendering calculation tasks according to the 3D model block transmission sequence provided by the edge server; after the service initialization is completed, the terminal device visually partitions the large-scale 3D model based on the visual characteristics of the field of view, and establishes a rendering calculation offloading mechanism for the remote server based on this, decomposing the rendering calculation task into multiple subtasks, and creating a corresponding number of independent threads in the terminal browser to execute these subtasks in parallel.
2. The large-scale 3D model dynamic rendering system based on multi-threading mechanism according to claim 1, characterized in that: The subdividing of the mesh 3D model into smaller model blocks includes: analyzing the entire 3D model based on image features or texture information to identify its surface or region fragments; converting the model into a graph structure and using a minimum cut maximum flow algorithm to perform image region segmentation, thereby dividing the large-scale 3D model into multiple 3D model blocks.
3. The large-scale 3D model dynamic rendering system based on multi-threading mechanism as claimed in claim 1, characterized in that: The method of continuously transmitting rendering data to a mobile device through multi-threaded asynchronous communication includes: after an HTTP / 2 connection is established, the mobile terminal creates multiple streams within the connection, each stream represents an independent request / response channel, and supports concurrent transmission of multiple requests and responses. The mobile browser determines the 3D model block and its corresponding resolution sequence parameters based on the user-defined field of view, initiates a request / creates a new stream on the remote server and assigns a unique identifier. The request can be divided into one or more frames for transmission as needed. The frame is the minimum transmission unit of HTTP / 2 and consists of a frame header and a frame payload. During this process, multiple streams are transmitted through the same connection, realizing concurrent transmission of requests and responses without waiting for the previous request response to be completed; the client sends an HTTP request frame through the created stream, and the HTTP request frame includes information about the encapsulated request method, uniform resource identifier URI, and request header; after the server receives the request and processes it, it returns an HTTP response frame containing a response status code, a response header, and a response body. The server divides the generated response into one or more frames and sends them back to the client according to the corresponding stream identifier.
4. The large-scale 3D model dynamic rendering system based on multi-threading mechanism as claimed in claim 1, characterized in that: The terminal device adopts a multi-threaded parallel processing mode including: in the data receiving and caching stage, the browser synchronously enables multiple loading threads to receive and store 3D model block data from a remote server.
5. The large-scale 3D model dynamic rendering system based on multi-threading mechanism as claimed in claim 1, characterized in that: The terminal device adopts a multi-threaded parallel processing mode including: in the data merging and rendering stage, the main thread transmits the data processed by the loading thread to the rendering engine for final rendering and display.
6. The large-scale 3D model dynamic rendering system based on multi-threading mechanism as claimed in claim 1, characterized in that: The mobile device web browser cache is implemented in the following ways: optimizing cache management by capturing the user's field of view (FOV) information, detecting the user's visual behavior through sensor technology or a browser interface to perceive the user's current field of view range, adopting a priority-based cache elimination strategy, reducing the cache priority of data outside the field of view range, and releasing cache space for data that is about to enter the field of view area. For foreground data within the field of view (FOV) area, prefetching and depth-first strategies are used to optimize cache management. The prefetching strategy pre-loads data that may enter the field of view by analyzing user behavior and data reference patterns.
7. The large-scale 3D model dynamic rendering system based on multi-threading mechanism as claimed in claim 6, characterized in that: Use linked list method to optimize cache of mobile web browsers.
8. The large-scale 3D model dynamic rendering system based on multi-threading mechanism as claimed in claim 7, characterized in that: The mobile web browser cache optimization using the linked list method specifically includes: Suppose the set of 3D model blocks in the current user field of view S is: Block_FOV∈S = [blocks-m, ..., blocks,..., blocks+m], where the field of view area corresponding to the model block block_i is denoted as area_i, and m is a positive integer; The rendering data of the model block in the view S is expressed as: render(i, res_i∈S), where res_i∈S represents the resolution required by the model block under the current view S (res_i∈S ∈ Res); The set of rendered data within the current field of view is represented as: Ren_FOV=S = [render(s−m, res_(sm)∈S),..., render(s, res_s∈S), ..., render(s+m, res_(s+m)∈S)] When the user's perspective changes, causing the center of the field of view to shift by Δs, the corresponding field of view changes to ΔS. At this time, the rendering resolution of each model block in the field of view will change. The new resolution of the i-th model block is recorded as res_i∈ΔS, and the resolution change is expressed as: Δres_i∈ΔS = res_i∈ΔS - res_i∈S. The sequential access feature of the linked list structure allows it to quickly locate the head data without global traversal, thereby reducing data search time. According to visual characteristics analysis, the rendering data render(s, res_s∈S) corresponding to the center point s of the field of view has the highest interaction probability. Therefore, when the 3D model application is initialized, it is used as the head of the linked list for priority access. Subsequently, the data of the two side areas: the left area render(s-1, res_(s-1)∈S) and the right area render(s+1, res_(s+1)∈S) are linked as successor nodes. By predicting the user's perspective change trend, the rendering data of the adjacent model blocks with high interaction probability are constructed as linked list nodes. Starting from the center point render(s, res_s∈S), the sequential traversal is carried out until the rendering data of all model blocks in the current field of view S are stored in the cache linked list. When the mobile browser needs to obtain rendering data from the cache: it starts searching from the head of the linked list, giving priority to drawing the 3D scene of the area corresponding to the center point of the viewport; it traverses the linked list in sequence, drawing the data of each node on both sides of the central model block in turn; and executes the loop until all 3D model blocks in the current viewport are rendered. When the user's viewpoint changes, the current field of view FOV will change accordingly. Since the continuity of the user's visual features determines the transition of the current field of view, the cached data will change as follows: When the viewpoint moves to the left: the rendering data render(s-1, res_(s-1)∈S-1) of block_{s-1} becomes the center area of the current field of view. The linked list connection relationship is modified. The mobile device sets the forward pointer of render(s-1, res_(s-1)∈S-1) to null, making it the new head node of the linked list, and modifies the head value of render(s-2, res_(s-2)∈S-1) to point to the node corresponding to render(s-1, res_(s-1)∈S-1). The forward pointer of render(s, res_s∈S-1) is set to point to render(s-2, res_(s-2)∈S-1), and the process continues until all display area data within the current view S-1 are stored in the linked list. The same processing flow is used when the viewpoint moves right.
9. The large-scale 3D model dynamic rendering system based on multi-threading mechanism as claimed in claim 1, characterized in that: The working process of the collaborative data caching module is as follows: the multi-resolution rendering data in the mobile edge server cache is stored in a tree structure, and each tree is represented as Tree_i, i∈Block, corresponding to the rendering data of a 3D model block, storing multi-resolution data [render(i,res0),render(i,res1),...,render(i,resm)], the root node of the tree stores the current field of view resolution rendering data render(i,rescurrent) currently requested by the mobile web browser, and the child nodes of the root node store rendering data of resolutions adjacent to the current field of view area: the left child node stores rendering data corresponding to the resolution rescurrent-1, and the right child node stores rendering data corresponding to the resolution rescurrent+1. Each node also stores the addresses of its connected parent node and child node to achieve fast traversal of the tree. When the user's perspective changes, the traversal pointer of the tree structure is written to the root node.