An adaptive transmission scheduling method for AR applications

Abstract

This invention discloses an adaptive transmission scheduling method for AR applications, relating to the fields of network communication and AR technology. Addressing issues such as transmission and rendering latency, main thread blocking, and poor environment adaptability in AR applications, this invention establishes a client-server communication architecture with dual communication links. The client detects network bandwidth and rendering rate in real time and reports this information to the server via WebSocket. The server abstracts each LOD level of the 3D model into independent jobs, defining two sequential processes: transmission and rendering. With the goal of minimizing total completion time, the Johnson scheduling algorithm is used to generate and distribute the optimal job scheduling sequence. The client utilizes a WebWorker pool to asynchronously load and process model data using multiple threads, while the main rendering thread continuously renders, updating model accuracy using an intelligent sharpness management strategy. A closed-loop adaptive mechanism is constructed to dynamically reschedule when network, device computing power, or scene changes. This invention achieves globally optimal scheduling for transmission and rendering, shortens loading time, avoids main thread blocking, adapts to heterogeneous dynamic environments, and effectively improves the smoothness of AR interaction and resource utilization efficiency.

Description

Technical Field

[0001] This invention relates to the field of network communication technology, and more specifically to an adaptive transmission scheduling method for AR applications. Background Technology

[0002] In current AR applications, especially on mobile devices, direct transmission is typically used to present high-precision 3D models to users. After the client initiates a request, the server sends the complete, high-precision model file, such as in GLB or glTF format, to the client all at once. Upon receiving the complete file, the client decodes, parses, loads textures, and performs final rendering in its main rendering thread. To optimize loading speed, some solutions employ Level-of-Depth (LOD) technology, prioritizing the transmission and display of lower-precision models. However, this is usually based on a preset, fixed loading order and fails to dynamically adapt to environmental changes.

[0003] For the transmission of immersive media content, the industry has extensively researched chunking, multi-resolution coding, and adaptive streaming technologies such as MPEG-DASH. International standards organizations such as 3GPP have also developed device media capability specifications for AR / VR services (e.g., 3GPP TS 26.119) and VR streaming profiles (e.g., 3GPP TS 26.118), aiming to promote industry standardization and interoperability, ensuring efficient transmission of XR content across devices with different capabilities. In the field of Web XR development, industry practices also emphasize the importance of optimizing performance through asynchronous or multi-threaded processing.

[0004] The existing technical solutions mainly have the following technical problems: High latency and poor experience: High-precision model files are huge in size, and direct transmission results in excessively long download times. Users have to endure a long "white screen" wait before the scene is fully loaded, which seriously affects the immersive experience.

[0005] Lack of environmental adaptability: Most existing solutions employ fixed transmission strategies, which cannot adapt to the dynamically changing wireless network environment in which user terminals operate, such as differences in bandwidth, latency, jitter, and device performance. When network conditions are poor or device computing power is limited, problems such as loading timeouts and screen stuttering are easily caused.

[0006] Main thread blocking causes interactive stuttering: Placing computationally intensive tasks such as model decoding and parsing in the main rendering thread will severely block the thread, causing the UI to become unresponsive and the rendering frame rate to drop sharply, thus disrupting the smoothness of real-time interaction in AR applications.

[0007] The existing technologies suffer from the aforementioned problems because they treat transmission and rendering as two isolated processes, lacking a global, collaborative scheduling mechanism aimed at minimizing total latency. The primary technical problem addressed by this invention is: how to intelligently determine an optimal data transmission and rendering order based on the user's dynamically changing network and device performance, in order to minimize total processing time and ensure smooth rendering. Summary of the Invention

[0008] To achieve the above objectives, the present invention provides the following technical solution: An adaptive transmission scheduling method for AR applications, the method comprising the following steps: Step 1: Complete the initialization of the client-server communication architecture, deploy each functional module, and establish dual communication links; Step 2, Performance Detection and Reporting: The client detects and predicts bandwidth B and rendering rate in real time. And report to the server via WebSocket; Step 3, Job Abstraction and Scheduling: The server abstracts each LOD level of the 3D model into an independent job, defines two sequential processes: transmission and rendering, calculates the processing time of each job, and executes Johnson scheduling to obtain the optimal job scheduling order π with the goal of minimizing the total completion time from the client to the server, and then sends it to the client. Step 4, Multi-threaded asynchronous execution: The client executes model data processing in the optimal job scheduling order π, loads and processes model data asynchronously through the WebWorker pool, the main rendering thread continues rendering, and the intelligent rendering management module updates the model accuracy; Step 5, Closed-loop adaptive scheduling: The client continuously monitors the environmental status. If the network, device computing power, or scenario changes, steps 2-4 are re-executed to dynamically update the optimal job scheduling order π.

[0009] Specifically, in step 1, the client deploys a performance monitoring module, a multi-threaded processing module, an intelligent rendering management module, and a 3D rendering engine to respectively realize performance index detection, asynchronous processing of model data, intelligent management of rendering accuracy, and AR scene rendering; the server deploys a scheduling algorithm module and an AR 3D model storage module to respectively realize job abstraction, scheduling algorithm solving, and multi-LOD level AR 3D model storage. The performance monitoring module performs network testing, evaluates and obtains the predicted bandwidth B, and estimates the rendering rate based on device hardware information and historical frame rate data. ; In step 3, after receiving the performance metrics, the server performs job abstraction, which involves abstracting each LOD level of each 3D model into an independent job. The operation Carrying the size of the 3D model data corresponding to the LOD level Number of dough sheets The network transmission process of the 3D model is defined as transmission step M1, and the terminal parsing and rendering process of the 3D model is defined as rendering step M2. M1 and M2 are two sequential steps that all jobs must execute in turn, based on the predicted bandwidth B and rendering rate. Calculate each job During the transmission time of transmission process M1 and rendering time in rendering process M2 With all assignments The optimization objective is to minimize the total client-to-server completion time from the execution of job M1 to the completion of job M2. The Johnson scheduling algorithm is then applied to all jobs. The tasks are grouped and then sorted by total completion time from smallest to largest. The optimal job scheduling order π is obtained by solving the problem and encapsulating the optimal job scheduling order π into a scheduling instruction, which is then sent to the client via a WebSocket link. In step 4, the multi-threaded asynchronous execution specifically involves the multi-threaded processing module calling the Web Worker background thread pool to perform asynchronous loading and processing of 3D model data, the 3D rendering engine continuously executing AR scene rendering on the main rendering thread, and the intelligent rendering management module updating the model accuracy of the AR scene based on the intelligent sharpness management strategy.

[0010] Specifically, the data size in step 2 The time required for transmission process M1 is: , Number of pieces The time required for rendering process M2 is: in, For homework During the processing time of transmission process M1; For homework During the processing time of rendering process M2; For homework The size of the 3D model data corresponding to the LOD level. For homework The number of facets in a 3D model corresponding to the LOD level.

[0011] Specifically, the optimization objective of the Johnson scheduling algorithm in step 2 is to find the optimal job scheduling order π, so that all jobs... Total client-to-server completion time Minimum, the total completion time Let π be the completion time of the last job in the optimal job scheduling order on rendering process M2. The objective function for optimization is: Where N is the job to be loaded. The total number of jobs, where πN is the last job in the optimal job scheduling order π. Let πN be the completion time of job πN in rendering process M2.

[0012] Specifically, the optimal job scheduling order in step 2 is as follows: Where N is the number of 3D mesh models, and Li is the LOD level of the i-th 3D model. The completion time of task πk in transport process M1 is calculated using the following formula: , The calculation formula is as follows: πk is the completion time of task πk in rendering process M2. , The start time of job πk in rendering process M2 is calculated using the following formula: Specifically, the steps for executing the Johnson scheduling algorithm in step 2 are as follows: S5.1 will handle all tasks The jobs are divided into two groups, G1 and G2. The first group, G1, consists of jobs whose transmission time is less than or equal to the rendering time, i.e., Ti≤Ri. The second group, G2, consists of jobs whose transmission time is greater than the rendering time, i.e., Ti>Ri. S5.2 sorts the two groups of jobs separately: G1 is sorted in ascending order by transmission time Ti, and G2 is sorted in descending order by rendering time Ri. S5.3 Arrange the sorted G1 and G2 in order to obtain the final optimal job scheduling order π.

[0013] Specifically, the asynchronous loading and processing method of the 3D model data in step 3 is as follows: the multi-threaded processing module selects an idle Web Worker from its maintained Web Worker pool, and delivers the parsing and decompression tasks of the model data to the idle Web Worker for background execution. The client's main rendering thread does not participate in the model data processing and continues to render the current AR scene at a high frame rate.

[0014] Specifically, the rendering process described in step 3, which combines the intelligent rendering management module, involves the main rendering thread updating the AR scene through the intelligent clarity management strategy of the intelligent rendering management module after receiving the model data processed by the Web Worker. This automatically upgrades or replaces the model quality or skips low-quality rendering, avoiding redundant calculations.

[0015] According to claim 1, an adaptive transmission scheduling method for AR applications is characterized in that, in the closed-loop adaptive scheduling mechanism, the change in environment includes one or more of the following: the client detects network bandwidth fluctuations, changes in device computing load, and AR scene updates. After the client detects the environmental change, the performance monitoring module re-detects real-time performance indicators and reports them to the server, triggering the server to re-execute the Johnson scheduling algorithm. Compared with the prior art, the beneficial effects of the present invention are as follows: 1. Achieve globally optimal scheduling for transmission and rendering, significantly reducing total time: Innovatively abstract each LOD level of the AR model into a pipeline operation, and define transmission and rendering as two sequential processes. The optimal job scheduling order with the minimum total completion time is obtained by solving the Johnson scheduling algorithm, which theoretically guarantees the highest efficiency of end-to-end transmission and rendering, minimizes the idle waiting time of the rendering device, and significantly shortens the total time from when the user initiates a request to when they see the complete AR scene; 2. Decouple data processing and rendering threads to ensure smooth interaction: Implement multi-threaded asynchronous loading and processing of model data through the Web Worker pool, place computationally intensive tasks to be executed in background sub-threads, and ensure that the main rendering thread is completely unblocked, so as to ensure smooth UI response and stable rendering frame rate of AR application during data loading, and meet the core requirements of real-time interaction of mobile AR application. 3. Construct a closed-loop adaptive mechanism to adapt to heterogeneous environments: Through a closed-loop process of real-time client performance detection - dynamic server scheduling - client execution - rescheduling due to environmental changes, the transmission scheduling strategy can be dynamically generated according to each user's real-time network conditions and device performance. In different wireless network environments such as 5G and Wi-Fi, as well as on mobile terminals with different configurations, it can automatically adjust to the optimal working mode to achieve a stable and high-quality experience in heterogeneous environments. 4. Intelligent management of rendering resources to improve resource utilization efficiency: Intelligent resolution management strategy enables intelligent decision-making for model rendering, automatically upgrades model quality or skips low-quality rendering, avoids ineffective rendering calculations and resource waste, and efficiently utilizes the limited computing power and storage resources of mobile terminals. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall process of the present invention. Detailed Implementation

[0017] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0018] like Figure 1 As shown, Figure 1 As shown, the present invention provides an adaptive transmission scheduling method for AR applications, applicable to real-time loading and interactive scenarios of high-precision 3D AR models on mobile devices using a client-server (C / S) architecture wireless network. The following detailed embodiments, in conjunction with the accompanying drawings and actual application scenarios, elaborate on the technical solution of the present invention, clarifying the execution logic of each step, parameter calculation, and the actual application process of the core algorithm. Specific examples are also given for wireless network adaptation and the calculation of the optimal job scheduling order π. This embodiment is only used to explain the present invention and is not intended to limit the scope of protection of the present invention.

[0019] Step 1: Complete the initialization of the client-server communication architecture, deploy each functional module, and establish dual communication links. Wireless Network Environment: This implementation uses a 5G cellular network / Wi-Fi 6 low-latency wireless communication network as the communication carrier between the client and the server. The end-to-end transmission latency of this wireless network is ≤20ms, supports the transmission of large file 3D model data via HTTP / 2 protocol, and long-connection low-latency signaling interaction via Web Socket protocol, meeting the requirements of AR applications for real-time and stable data transmission. Client Configuration: The client is a mobile device with AR rendering capabilities, such as a smartphone or AR glasses, with built-in performance monitoring module, multi-threaded processing module, intelligent rendering management module, and 3D rendering engine; the multi-threaded processing module maintains a WebWorker background thread pool containing 4 independent threads, which is dedicated to performing computationally intensive model data processing tasks, and the main rendering thread is completely decoupled from the background threads; the client hardware supports AR spatial positioning and high-definition image rendering, and can collect local data such as device hardware computing power and historical rendering frame rate; Server configuration: The server is a cloud server that stores multiple LOD versions of various AR 3D models (low-precision LOD versions have fewer faces and smaller data volume, while high-precision LOD versions have richer details and larger data volume and number of faces). It is equipped with a scheduling algorithm module that supports parsing performance indicators reported by clients, job abstraction, and solving the Johnson scheduling algorithm. The server also supports Web Socket long-connection signaling interaction and HTTP protocol model data push, and can send model data to clients according to the optimal job scheduling order. Data interaction rules: The client and server communicate with each other via WebSocket for low-latency signaling of performance metrics and scheduling instructions, and transmit 3D model data in an orderly manner via HTTP. All operations are performed on the aforementioned 5G / Wi-Fi 6 wireless network.

[0020] The method of this invention is executed in the following order: client performance detection and reporting - server job abstraction and Johnson scheduling - client multi-threaded asynchronous execution - closed-loop adaptive scheduling. The specific execution logic, parameter calculation and operation details of each step are as follows: Step 2: Client performance detection and reporting This step is executed by the client's performance monitoring module. The trigger condition is the launch of the AR application / the client entering a new AR interactive scene. The core is to obtain two performance metrics representing network transmission capabilities and terminal rendering capabilities, and report them to the server via a WebSocket link. Specific operations are as follows: Obtaining the predicted bandwidth B: Under the current 5G / Wi-Fi 6 wireless network environment, the performance monitoring module performs 5 rounds of continuous network speed tests, and takes the average download speed of each round of speed test to obtain the predicted bandwidth B (unit: MB / s), which represents the network transmission capacity. This value is the core basis for the server to calculate the transmission time of the model. Rendering speed Acquisition: The performance monitoring module reads the client's hardware computing power information (CPU / GPU model, RAM), combines it with historical rendering frame rate data and historical model rendering time records of the AR application, and estimates the rendering rate, which characterizes the terminal's rendering capability, through data fitting. (Unit: 10,000 polygons / second), this value is the core basis for the server to calculate the model rendering time; Performance metrics reporting: The performance monitoring module will predict bandwidth B and rendering rate. The performance reporting signaling, encapsulated in JSON format, is pushed to the server in real time via a WebSocket long connection. After receiving the signaling, the server parses it to provide basic parameters for subsequent Johnson scheduling.

[0021] In this embodiment, the client detects a predicted bandwidth B = 20MB / s and a rendering rate of... =100,000 pieces / second.

[0022] Step 3: Server Job Abstraction and Scheduling This step is executed by the server's scheduling algorithm module. The trigger condition is that the server receives a performance reporting signal from the client. Its core function is to complete job abstraction, process time calculation, and Johnson's scheduling algorithm solution, ultimately obtaining the optimal job scheduling order π and sending it to the client. This is the core step of this invention. This step will illustrate the entire process of job abstraction, parameter calculation, and π solution with specific examples.

[0023] Sub-step 2.1: Job Abstraction The scheduling algorithm module abstracts each LOD level of each 3D model to be loaded on the server into an independent job. For each job Assign a unique identifier and associate it with two core attributes: model data size. (Model file size corresponding to LOD level, unit: MB), number of model faces (Number of model faces corresponding to the LOD level, unit: 10,000 faces); Simultaneously, the process of transmitting the 3D model via wireless network over an HTTP link is divided into transmission process M1, and the process of parsing and rendering the 3D model on the client terminal is divided into rendering process M2. M1 and M2 are clearly defined as the processes for all operations. The two sequential steps must be executed in turn: all jobs must first complete the M1 transfer and then execute the M2 rendering.

[0024] Specific example: Assume there are two AR 3D models (Model A and Model B) to be loaded on the server. Each model contains two Levels of Detail (LOD1, low precision; LOD2, high precision), abstracted into four independent jobs J1, J2, J3, and J4. Each job... , The attributes are as follows: Sub-step 2.2: Calculate the task Transmission time Rendering time The scheduling algorithm module is based on the predicted bandwidth B and rendering rate reported by the client. Calculate each task according to the following formula. During the transmission time of transmission process M1 During the rendering time of rendering process M2 : , in, For homework Processing time in M1 (in seconds). For homework Processing time in M2 (in seconds).

[0025] Specific example calculation: Combining the exemplary data from step 1, B = 20MB / s, =100,000 faces / second, and the job attributes in sub-step 2.1, calculate the four jobs. , The results are as follows: Sub-step 2.3: Execute the Johnson scheduling algorithm to solve for the optimal job scheduling order π. The scheduling algorithm module uses all jobs Total end-to-end completion time from executing M1 to completing M2 With the goal of minimizing, the Johnson scheduling algorithm is executed for all jobs. The algorithm performs grouping, sorting, and synthesis to obtain the optimal job scheduling order π. The specific operations executed by the algorithm strictly follow the specifications in the claims and consist of three steps: Job grouping: Divide all jobs Ji into two groups, G1 and G2. Group G1 consists of jobs whose transfer time is less than or equal to the rendering time (Ti≤Ri), and group G2 consists of jobs whose transfer time is greater than the rendering time (Ti>Ri). Grouping and sorting: Group G1 is sorted in ascending order by transmission time Ti, and Group G2 is sorted in descending order by rendering time Ri; Sequential composition: The sorted G1 group and G2 group are concatenated in sequence to obtain the final optimal job scheduling order π.

[0026] Specific example for calculating π: Combining the examples above... , Calculate the results and execute the Johnson scheduling algorithm: Grouping: Since Ti ≤ Ri for all 4 jobs, all jobs are assigned to group G1, and there is no group G2; Group G1: J1, J2, J3, J4; Sort: Group G1 is sorted in ascending order of Ti. The sorted result is: J1 (T=0.1) → J3 (T=0.2) → J2 (T=0.4) → J4 (T=0.5). Synthesis: Since there is no G2 group, the optimal job scheduling order π is the sorting result of G1 group, that is: π=(π1,π2,π3,π4)=(J1,J3,J2,J4) If a new task J5 (S5=10MB, F5=50,000 faces) is added, the calculated time is T5=0.5s, R5=0.5s (Ti=Ri, assigned to G1); task J6 (S6=8MB, F6=40,000 faces) is calculated, the calculated time is T6=0.4s, R6=0.4s (Ti=Ri, assigned to G1); task J7 (S7=5MB, F7=20,000 faces) is calculated, the calculated time is T7=0.25s, R7=0.2s (Ti>Ri, assigned to G2); group G2 is sorted in descending order of Ri (only J7), and the final result is π=(J1,J3,J2,J4,J5,J6,J7).

[0027] Sub-step 2.4: Issue the optimal job scheduling order π The scheduling algorithm module encapsulates the optimal job scheduling order π obtained from the solution into a scheduling instruction, which is then sent to the client via a WebSocket long connection. After receiving the scheduling instruction, the client completes the parsing and provides the execution order basis for the subsequent loading and processing of model data.

[0028] Step 4: Asynchronous execution using multiple threads This step is executed collaboratively by the client's multi-threaded processing module, 3D rendering engine, and intelligent rendering management module. The trigger condition is that the client receives a scheduling instruction from the server. The core is to achieve asynchronous loading and processing of model data and smooth rendering of the AR scene according to the optimal job scheduling order π, strictly adhering to the multi-threaded execution logic in the claims. Specific operations are as follows: Web Worker background thread pool call: The multi-threaded processing module selects an idle Web Worker thread from its maintained Web Worker background thread pool, and assigns computationally intensive tasks such as downloading, decompressing, and parsing the 3D model data corresponding to the job to the idle Web Worker thread for background execution according to the optimal job scheduling order π; the client's main rendering thread does not participate in any model data processing operations, fundamentally avoiding main thread blocking; Main rendering thread continuously renders: In the main rendering thread, the 3D rendering engine continuously renders the AR real-world scene at a high frame rate (≥60fps) based on the real-time transmission status of the current 5G / Wi-Fi 6 wireless network, ensuring that the UI is unresponsive and the screen is smooth during user interaction. Intelligent Sharpness Management Strategy Execution: The intelligent rendering management module receives model data processed by the Web Worker background thread in real time, and executes intelligent sharpness management strategies based on the user's perspective and display area in the current AR scene. If the area corresponding to the model data is the display area of ​​the current AR scene, the accuracy of the AR 3D model in that area will be immediately upgraded to the LOD level corresponding to the model data, thus achieving a smooth improvement in image quality. If the area corresponding to the model data is a non-display area of ​​the current AR scene, the model accuracy update will not be performed for the time being. The model data will be cached locally, and the model loading and accuracy upgrade will be completed when the user's viewpoint switches to that area, thus avoiding redundant calculations and saving computing power on the client terminal.

[0029] Step 5: Closed-loop adaptive scheduling This step is initiated by the client's performance monitoring module and executed in conjunction with the server's scheduling algorithm module. Its core function is to achieve real-time monitoring of the wireless network environment and client device status. When the environment changes, the scheduling process is re-triggered to ensure that the AR application is always in the optimal transmission and rendering state. Specific operations are as follows: Real-time environment monitoring: The client's performance monitoring module continuously monitors the environmental status during the operation of the AR application. The monitoring content includes: bandwidth fluctuations of the 5G / Wi-Fi6 wireless network (predicted changes in bandwidth B), changes in the computing load of the client device (CPU / GPU utilization, memory usage), and scene updates of the AR application (entering a new AR interactive scene). Re-triggering of the scheduling process: If the performance monitoring module detects a change in any of the above environmental states, it will immediately re-execute the client performance detection and reporting operation in step 1 to obtain the new predicted bandwidth B′ and rendering rate. And report it to the server; Regenerating the optimal scheduling order π: After receiving the new performance metrics, the server re-executes the job abstraction and Johnson scheduling operations from step 2, based on the new B′, Recalculate each job ′、 The algorithm calculates the optimal job scheduling order π′ and then sends it to the client. Client updates execution strategy: After receiving the new optimal job scheduling order π′, the client immediately updates the model data loading, processing and rendering strategy, and re-executes the multi-threaded asynchronous execution operation in step 3 according to the order of π′, so as to realize the dynamic adaptive adjustment of the transmission scheduling strategy.

[0030] If the client switches from a 5G network to a regular Wi-Fi network during the AR application's operation, and the performance monitoring module detects a drop in predicted bandwidth from B=20MB / s to B′=5MB / s, it immediately re-reports the performance metrics; the server then recalculates the performance metrics for each job based on B′=5MB / s. (e.g., T1′=2 / 5=0.4s for J1), and re-execute Johnson scheduling to obtain π′; the client prioritizes loading low-precision LOD level jobs with small data volume and few faces according to π′ to ensure the smooth rendering of AR scenes in low-bandwidth wireless network environments.

[0031] Glossary AR (Augmented Reality): A technology that cleverly integrates virtual information with the real world. It uses computers to generate virtual information such as text, images, and 3D models, and applies them to the real world to "enhance" reality.

[0032] LOD (Level of Detail): Multiple versions of the same 3D model with different levels of complexity. Lower LOD versions have fewer faces and less data, resulting in faster rendering; higher LOD versions offer richer details but consume more resources.

[0033] Johnson's rule: A greedy algorithm for finding the optimal solution to the two-machine flow scheduling problem.

Claims

1. An adaptive transmission scheduling method for AR application, characterized in that, The method includes the following steps: Step 1: Complete the initialization of the client-server communication architecture, deploy each functional module, and establish dual communication links; Step 2, performance detection and reporting: the client real-time detects the predicted bandwidth B, rendering rate and reports the server through WebSocket; Step 3, Job Abstraction and Scheduling: The server abstracts each LOD level of the 3D model into an independent job, defines two sequential processes: transmission and rendering, calculates the processing time of each job, and executes Johnson scheduling to obtain the optimal job scheduling order π with the goal of minimizing the total completion time from the client to the server, and then sends it to the client. Step 4, Multi-threaded asynchronous execution: The client executes model data processing in the optimal job scheduling order π, loads and processes model data asynchronously through the WebWorker pool, the main rendering thread continues rendering, and the intelligent rendering management module updates the model accuracy; Step 5, Closed-loop adaptive scheduling: The client continuously monitors the environmental status. If the network, device computing power, or scenario changes, steps 2-4 are re-executed to dynamically update the optimal job scheduling order π.

2. The adaptive transmission scheduling method for AR application according to claim 1, wherein, In step 1, the client deploys a performance monitoring module, a multi-threaded processing module, an intelligent rendering management module, and a 3D rendering engine to respectively realize performance index detection, asynchronous processing of model data, intelligent management of rendering accuracy, and AR scene rendering; the server deploys a scheduling algorithm module and an AR 3D model storage module to respectively realize job abstraction, scheduling algorithm solving, and multi-LOD level AR 3D model storage. The performance monitoring module performs network test, evaluates to obtain predicted bandwidth B, and the performance monitoring module estimates rendering rate according to device hardware information and historical frame rate data ; In step 3, after receiving the performance metrics, the server performs job abstraction, which involves abstracting each LOD level of each 3D model into an independent job. The operation Carrying the size of the 3D model data corresponding to the LOD level Number of dough sheets The network transmission process of the 3D model is defined as transmission process M1, and the terminal parsing and rendering process of the 3D model is defined as rendering process M2. M1 and M2 are the processes for all operations. The two sequential steps need to be executed in turn, based on the predicted bandwidth B and the rendering rate. Calculate each job During the transmission time of transmission process M1 and rendering time in rendering process M2 With all assignments The optimization objective is to minimize the total client-to-server completion time from the execution of job M1 to the completion of job M2. The Johnson scheduling algorithm is then applied to all jobs. The tasks are grouped and then sorted by total completion time from smallest to largest. The optimal job scheduling order π is obtained by solving the problem and encapsulating the optimal job scheduling order π into a scheduling instruction, which is then sent to the client via a WebSocket link. In step 4, the multi-threaded asynchronous execution specifically involves the multi-threaded processing module calling the WebWorker background thread pool to perform asynchronous loading and processing of 3D model data, the 3D rendering engine continuously executing AR scene rendering on the main rendering thread, and the intelligent rendering management module updating the model accuracy of the AR scene based on the intelligent sharpness management strategy.

3. The adaptive transmission scheduling method for AR applications according to claim 2, characterized in that, Data size in step 2 The time required for transmission process M1 is: ， Number of pieces The time required for rendering process M2 is: in, For homework During the processing time of transmission process M1; For homework During the processing time of rendering process M2; For homework The size of the 3D model data corresponding to the LOD level. For homework The number of facets in a 3D model corresponding to the LOD level.

4. The adaptive transmission scheduling method for AR applications according to claim 3, characterized in that, The optimization objective of the Johnson scheduling algorithm in step 2 is to find the optimal job scheduling order π, so that all jobs... Total client-to-server completion time Minimum, the total completion time Let π be the completion time of the last job in the optimal job scheduling order on rendering process M2. The objective function for optimization is: Where N is the job to be loaded. The total number of jobs, where πN is the last job in the optimal job scheduling order π. Let πN be the completion time of job πN in rendering process M2.

5. The adaptive transmission scheduling method for AR applications according to claim 4, characterized in that, The optimal job scheduling order in step 2 is: Where N is the number of 3D mesh models, and Li is the LOD level of the i-th 3D model. The completion time of task πk in transport process M1 is calculated using the following formula: ， The calculation formula is as follows: πk is the completion time of task πk in rendering process M2. ， The start time of job πk in rendering process M2 is calculated using the following formula: 。 6. The adaptive transmission scheduling method for AR applications according to claim 5, characterized in that, The specific steps for executing the Johnson scheduling algorithm in step 2 are as follows: S5.1 will handle all tasks The jobs are divided into two groups, G1 and G2. The first group, G1, consists of jobs whose transmission time is less than or equal to the rendering time, i.e., Ti≤Ri. The second group, G2, consists of jobs whose transmission time is greater than the rendering time, i.e., Ti>Ri. S5.2 sorts the two groups of jobs separately: G1 is sorted in ascending order by transmission time Ti, and G2 is sorted in descending order by rendering time Ri. S5.3 Arrange the sorted G1 and G2 in order to obtain the final optimal job scheduling order π.

7. The adaptive transmission scheduling method for AR applications according to claim 1, characterized in that, The asynchronous loading and processing method of the 3D model data in step 3 is as follows: the multi-threaded processing module selects an idle WebWorker from its maintained WebWorker pool, and delivers the parsing and decompression tasks of the model data to the idle WebWorker for background execution. The client's main rendering thread does not participate in the model data processing and continues to render the current AR scene at a high frame rate.

8. The adaptive transmission scheduling method for AR applications according to claim 7, characterized in that, The rendering process described in step 3, which combines the intelligent rendering management module, specifically involves the main rendering thread updating the AR scene through the intelligent clarity management strategy of the intelligent rendering management module after receiving the model data processed by the WebWorker. This automatically upgrades or replaces the model quality or skips low-quality rendering, avoiding redundant calculations.

9. The adaptive transmission scheduling method for AR applications according to claim 1, characterized in that, In the closed-loop adaptive scheduling mechanism, the change in environment includes one or more of the following: the client detects network bandwidth fluctuations, changes in device computing load, and AR scene updates. After the client detects the change in environment, the performance monitoring module re-detects the real-time performance indicators and reports them to the server, triggering the server to re-execute the Johnson scheduling algorithm.

Images