Calculation network scheduling method and device for road detection task

Abstract

The invention relates to the technical field of expressway detection, and provides a calculation network scheduling method and device for a road detection task, and the method comprises the steps: obtaining a road detection task issued by an upper application; sensing the current computing power resource state of each computing power node in the computing network; extracting a resource demand of the road detection task, determining at least one computing power node of which the computing power resource state meets the resource demand as a target computing power node based on the task type of the road detection task, and scheduling the road detection task to the target computing power node for execution; wherein each computing power node in the computing network is deployed with a large model and an AI video detection model for executing different detection tasks, and the target computing power node executes the road detection task based on the large model and the AI video detection model and returns an execution result to the upper-layer application. The road detection task processing efficiency is improved, the processing pressure on high-speed detection related data is relieved, and the effect of bearing the whole flow of high-speed monitoring management is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of highway detection, and particularly to an arithmetic network scheduling method and device for highway detection tasks. Background Art

[0002] At present, highway (pavement) detection is mainly divided into two categories: manual detection and automated detection. Among them, manual detection has disadvantages such as heavy workload, low efficiency, poor safety, and large result differences, and cannot meet the requirements of fast and accurate detection. Existing automated detection equipment is expensive, and manual assistance is required to complete the entire detection work, so there are still certain limitations.

[0003] On the other hand, in the context of intelligent highways, there are a large number of data computing requirements, such as the need to analyze and process a large amount of video and picture data, and to monitor the process in real time, etc., which bring great pressure to data analysis, processing and other tasks, resulting in low processing efficiency and low data processing accuracy (such as pavement fault detection based on the captured video); at the same time, as more and more data processing, computing and other tasks sink to the edge side, due to the limited computing power resources of the edge-side computing power nodes, they cannot bear the entire process of high-speed monitoring and management. Summary of the Invention

[0004] The present invention provides an arithmetic network scheduling method and device for highway detection tasks to solve the problems of large data processing pressure, low efficiency, low accuracy in the processing of highway detection tasks in the prior art, and inability to bear the entire process of high-speed monitoring and management.

[0005] The present invention provides an arithmetic network scheduling method for highway detection tasks, including the following steps: Obtain the highway detection task issued by the upper-layer application; Perceive the current computing power resource status of each computing power node in the arithmetic network; Extract the resource requirements of the highway detection task, and based on the task type of the highway detection task, determine at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node, and schedule the highway detection task to the target computing power node for execution; Wherein, a large model and an AI video detection model for executing different detection tasks are deployed on each computing power node in the arithmetic network, and the target computing power node executes the highway detection task based on the large model and the AI video detection model, and returns the execution result to the upper-layer application.

[0006] According to the arithmetic network scheduling method for highway detection tasks provided by the present invention, determining at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node based on the task type of the highway detection task includes: When the task type of the highway detection task is delay-sensitive or interactive, at least one computing power node located at the edge or end side of the computing network and whose computing power resource status meets the resource requirements is determined as the target computing power node; When the task type of the highway detection task is non-delay-sensitive and non-interactive, at least one computing power node located on the cloud side of the computing network and whose computing power resource status meets the resource requirements is determined as the target computing power node.

[0007] According to a computing network scheduling method for highway detection tasks provided by the present invention, when the task type of the highway detection task is delay-sensitive or interactive and there is no computing power node on the edge or end side whose computing power resource status meets the resource requirements, all computing power nodes on the edge or end side are traversed, and a computing power node that is executing a non-delay-sensitive or non-interactive task is selected as the target computing power node, and the non-delay-sensitive or non-interactive task being executed is forwarded to a computing power node on the cloud side or another computing power node with a computing power performance lower than that of the target computing power node.

[0008] According to a computing network scheduling method for highway detection tasks provided by the present invention, determining at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node based on the task type of the highway detection task includes: Dividing the highway detection task into multiple subtasks; Based on the task type of the subtask, determining at least one computing power node whose computing power resource status meets the resource requirements of the subtask as the target computing power node.

[0009] According to a computing network scheduling method for highway detection tasks provided by the present invention, it further includes: during the low-demand period of the highway detection task volume, releasing the highway detection tasks executed on some computing power nodes on the edge or end side, and the released computing power nodes are used to execute non-delay-sensitive or non-interactive tasks.

[0010] According to a computing network scheduling method for highway detection tasks provided by the present invention, the target computing power node executes the highway detection task based on a large model and an AI video detection model, and returns the execution result to the upper-layer application, including: Extracting the user prompt words in the highway detection task, inputting the user prompt words into the large model, so that the large model automatically identifies and diverts the monitoring video stream according to the user prompt words, and intercepts according to the user prompt words to obtain a video segment that conforms to the scene corresponding to the user prompt words; Sending the video segment to the AI video detection model corresponding to the scene to obtain a detection result; Input the detection result and the user prompt into the large model, so that the large model generates a task report according to the detection result and the user prompt, and returns the task report as the execution result to the upper-layer application.

[0011] The present invention also provides a computing network scheduling device for highway detection tasks, including the following modules: A task acquisition module, configured to acquire a highway detection task issued by an upper-layer application; A computing network perception module, configured to perceive the current computing power resource status of each computing power node in the computing network; A computing network scheduling module, configured to extract the resource requirements of the highway detection task, and based on the task type of the highway detection task, determine at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node, and schedule the highway detection task to the target computing power node for execution; Wherein, a large model and an AI video detection model for executing different detection tasks are deployed on each computing power node in the computing network, and the target computing power node executes the highway detection task based on the large model and the AI video detection model, and returns the execution result to the upper-layer application.

[0012] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and running on the processor. When the processor executes the program, it implements the computing network scheduling method for highway detection tasks as described in any one of the above.

[0013] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the computing network scheduling method for highway detection tasks as described in any one of the above.

[0014] The present invention also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the computing network scheduling method for highway detection tasks as described in any one of the above.

[0015] The computing network scheduling method and device for highway detection tasks provided by the present invention determine at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node according to the task type of the highway detection task, and schedule the highway detection task to the target computing power node for execution. Moreover, a large model and an AI video detection model for executing different detection tasks are deployed on each computing power node, and the target computing power node executes the highway detection task based on the large model and the AI video detection model, which improves the processing efficiency of highway detection tasks, reduces the processing pressure on high-speed detection-related data, and realizes the effect of carrying the entire process of high-speed monitoring and management. Description of the Drawings

[0016] To more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 It is one of the schematic flowcharts of the computing and network scheduling method provided by the present invention for highway detection tasks.

[0018] Figure 2 It is the schematic architecture diagram for implementing the computing and network scheduling method provided by the present invention for highway detection tasks.

[0019] Figure 3 It is the second schematic flowchart of the computing and network scheduling method provided by the present invention for highway detection tasks.

[0020] Figure 4 It is the schematic structural diagram of the computing and network scheduling device provided by the present invention for highway detection tasks.

[0021] Figure 5 It is the schematic structural diagram of the electronic device provided by the present invention. Detailed implementation manners

[0022] To make the objectives, technical solutions, and advantages of the present invention clearer, the following will clearly and completely describe the technical solutions in the present invention in conjunction with the drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. Based on the embodiments in the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the scope of protection of the present invention.

[0023] The computing and network scheduling method for highway detection tasks in the embodiments of the present invention, as Figure 1 and 2 shown, includes steps S110 to S130.

[0024] Step S110: Obtain the highway detection task issued by the upper-layer application. Specifically, the upper-layer application refers to the business application at the user layer. For example, for the detection of expressways, the upper-layer application is the intelligent expressway application component, which realizes functions such as AI video detection, intelligent question answering, intelligent query, intelligent prediction, intelligent pre-plan, and information release for high-speed safety hazards and abnormal events, and provides safe, intelligent, and efficient services for high-speed supervisors, traveling public, etc. For example, for the intelligent query component, if it is necessary to query which sections of the road surface have cracks, a corresponding crack detection task will be generated, and the execution of this task is to target cracks and perform target detection on the cracks in the video or image of each section of the road surface. Another example: for the intelligent pre-plan component, if it is necessary to conduct video surveillance on the expressway to handle traffic accidents or congestion in a timely manner, a video surveillance task will be generated to monitor whether traffic accidents or congestion occur in the surveillance video.

[0025] Step S120: Sense the current computing power resource status of each computing power node in the computing network. Among them, the computing network sensing component of the computing network can be called to sense the candidate computing power nodes that meet the resource requirements in the computing network, and uniformly access, analyze, and share the status of computing network resources, services, applications, etc., specifically including: computing network index sensing engine, resource unified management engine, and access task management engine.

[0026] The computing network index sensing engine, on the one hand, provides unified monitoring of resources such as computing power, network, and storage, accesses information such as available computing power resources and overall network load, and monitors the resource occupancy rate at the cluster level, user level, and task level; on the other hand, by providing interfaces to business systems, etc., it collects user-related business requirement data, as well as business service quality data such as service success rate and user SLA (Service Level Agreement) achievement level, and maps the user's business requirements (i.e., the resource requirements of the highway detection task in this embodiment) to standardized requirements for underlying computing power, network, and service capabilities.

[0027] The resource unified management engine provides unified management of computing power, network, and storage resources, including resource activation, resource sharing, and resource classification, etc., and constructs a unified standard model according to different computing power characteristics to form standard constraints for the same type of resources, including standard constraints on data and interfaces, etc. Through access, management, and opening, it realizes unified management of computing power nodes, various networks, and platform services.

[0028] The access task management engine provides unified access to various data sources, including data source management, adapter management, and task management, etc., dynamically manages, real-time monitors, and executes tracking of various access tasks, and retries or repeats the execution of abnormal and error tasks according to rules.

[0029] In this step, by invoking the computing network perception component, the current computing power resource status of each computing power node in the computing network can be obtained. For example: the remaining amounts of resources such as the current CPU (number of cores), GPU, and storage space of a certain computing power node.

[0030] Step S130: Extract the resource requirements of the road detection task. Based on the task type of the road detection task, determine at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node, and schedule the road detection task to the target computing power node for execution. Among them, the resource requirement is how many resources such as CPU (number of cores), GPU, and storage space need to be occupied when the road detection task is executed. This resource requirement can be sent to the computing network as an attribute item of the road detection task.

[0031] In this embodiment, the computing network scheduling component of the computing network can be invoked to specifically implement task scheduling. The computing network scheduling component splits and generates tasks such as functions or meta-computations according to different types of tasks, such as service type, content type, computing type, latency-sensitive type, and interactive type, as well as preset rules and strategies, etc., and forms an evaluation of resources such as computing power, network, and storage, and the perception results based on the periodic and real-time perception of indicators such as computing network resources, computing network services, and task SLA, etc., as the index input for optimized scheduling, to achieve resource and task scheduling in multi-user and multi-task scenarios. Specifically, it includes a computing network fusion orchestration engine, a scheduling strategy optimization engine, and a computing network scheduling execution engine.

[0032] The computing network fusion orchestration engine, based on container technology, extends container orchestration capabilities to the cloud, edge, end, as well as intelligent computing and fusion full-stack orchestration and management capabilities, realizes the integrated orchestration ability of tasks on cross-domain computing power resources, and achieves the goal of globally distributing application tasks.

[0033] The scheduling strategy optimization engine can construct an analysis model for application task deployment characteristics based on machine learning algorithms and manual processing, etc. For the analysis results and combined with user requirements and resource performance, it disassembles, deconstructs, and recombines each part module of the application task, and provides optimization suggestions, including: migration, scaling, and online / offline, etc. Specifically, the matching of computing power nodes during scheduling includes exact matching and fuzzy matching. The exact matching accurately matches resources in a prescriptive or rule-based manner according to the resource requirements of the road detection task to obtain corresponding supported resources. The fuzzy matching, when the resource requirements of the road detection task have no optimal or complete resource support, uses strategies such as lossy accuracy and lossy time to schedule relevant resources.

[0034] The computing network scheduling execution engine distributes each task of the application to the corresponding computing power nodes through the computing power network, and starts the construction and deployment of the application based on the fully containerized environment, realizing the automatic deployment of application modules and the automatic opening of network policies. Specifically, it includes task priority scheduling and task tidal scheduling, etc.

[0035] For the task priority scheduling, weights are first assigned according to the importance of tasks and sorted by priority. Combining task characteristics and priority analysis, for example, different task types have different task priorities. Based on the task priorities, resources (computing, network) are pre-allocated, and then business requests are forwarded according to the task priorities. Overall management of cloud, edge, and terminal computing power offloading and network bandwidth allocation tasks, and dynamic adaptive transmission and processing strategies are adopted at different time periods and different nodes respectively to improve the task service quality when the number of users is too high in a resource-constrained environment. Through the integrated scheduling of resources and applications, the resource preemption relationships in aspects such as computing, network, IO, and storage between non-delay-sensitive tasks and delay-sensitive tasks such as high-speed traffic event detection are reduced; and for complex applications containing multiple subtasks, the best forwarding paths of each service and the computing power nodes for offloading are determined.

[0036] For the task tidal scheduling, business migration is carried out according to the local demand hotspots in time and space. By sinking computing power resources, key business demand scenarios are preferentially guaranteed. And when local resources are preempted, low-latency requirement tasks are forwarded to nodes at a longer distance or with lower performance through task forwarding, etc. To maximize the user service quality under the premise of ensuring data transmission and security management requirements during busy hours, and effectively improve the overall service quality; for idle hours, to ensure the service quality of all services and improve the resource utilization rate on the supply side (i.e., the edge and terminal side). For example, in the high-speed video surveillance scenario, there is a large amount of static data in the night video, and the resource demand for the terminal side is significantly lower than that in the day. Therefore, some video surveillance tasks can be shut down at night, and the computing power nodes used for video surveillance can be reallocated to deploy non-real-time batch processing business applications such as OCR and text processing, effectively improving the computing power resource utilization rate.

[0037] In this step, it is only necessary to judge whether the current computing power resource status of each computing power node meets the resource requirements of the road detection task. For example, whether the remaining resources such as CPU, GPU, and storage space of a certain computing power node meet the requirements of the road detection task for CPU, GPU, and storage space, and based on the task type of the road detection task, call the computing network scheduling component to schedule the road detection task to the computing power node on the cloud side, edge side, or terminal side that is suitable for the task type.

[0038] Moreover, a large model and an AI video detection model for performing different detection tasks are deployed on each computing power node in the computing network. The target computing power node executes the highway detection task based on the large model and the AI video detection model, and returns the execution result to the upper-layer application. By fusing the large model and the AI video detection model to process the highway detection task, even in the case of limited computing power resources on the edge side, especially the end side, it is possible to improve the task processing efficiency, reduce the processing pressure on high-speed detection-related data, and achieve the effect of carrying out the entire process of high-speed monitoring and management.

[0039] It should be noted that in this embodiment, the intelligent highway application component captures pictures from the real-time monitoring video stream, provides training and inference of video data based on AI algorithms such as object detection and combined with the computing network scheduling ability, accurately identifies highway pavement and shallow diseases such as cracks and potholes, as well as abnormal events such as highway congestion and abnormal parking, and gives early warnings in a timely manner, providing services for highway supervisors, traveling public, etc. The intelligent highway application component specifically includes an AI vision detection engine (realizing intelligent query and intelligent prediction), a vision question-answering engine, an intelligent pre-plan engine, and an information release engine.

[0040] The AI vision detection engine captures pictures from the real-time monitoring video stream, provides training and inference of video data based on AI algorithms such as object detection and combined with the computing network scheduling ability, accurately identifies highway pavement and shallow diseases such as cracks and potholes, as well as abnormal events such as highway congestion and abnormal parking, and gives early warnings in a timely manner. Among them, highway pavement and shallow diseases such as cracks and potholes, as well as abnormal events such as highway congestion and abnormal parking are all the execution results of the corresponding highway detection tasks.

[0041] The vision question-answering engine is used for visual feature extraction and parsing user questions, providing multi-round intelligent question-answering capabilities. The answers to the questions are the execution results of the corresponding highway detection tasks, and are applied to scenarios such as real-time monitoring and alarm location.

[0042] The intelligent pre-plan engine is used to automatically generate intuitive and effective emergency disposal pre-plans according to the video detection information and alarm situations of highway road diseases, traffic events, etc. (the execution results of the corresponding highway detection tasks), and combined with historical case libraries, knowledge bases, etc., effectively improving the event disposal efficiency and intelligent level.

[0043] The information release engine is mainly used to intelligently release the pre-plan (the execution result of the corresponding highway detection task) in multi-modal ways such as pictures, texts, and sounds, through channels such as highway information boards, text messages, 5G messages, and radio stations, including detours, slowdowns, etc., providing travel information services for the public. It can be understood that the method and corresponding architecture of this embodiment are not only applicable to highways, but also applicable to the scheduling of highway detection tasks such as national roads, rural roads, and urban roads.

[0044] In the computing network scheduling method for highway detection tasks in this embodiment, at least one computing power node whose computing power resource status meets the resource requirements is determined as the target computing power node according to the task type of the highway detection task, and the highway detection task is scheduled to the target computing power node for execution. Moreover, a large model and AI video detection models for executing different detection tasks are deployed on each computing power node. The target computing power node executes the highway detection task based on the large model and the AI video detection models, improving the processing efficiency of the highway detection task, reducing the processing pressure on the data related to highway detection, and achieving the effect of carrying out the entire process of highway monitoring and management.

[0045] In some embodiments, step S130 specifically includes: in the case where the task type of the highway detection task is delay-sensitive (requiring short task processing and task result transmission times) or interactive, determining the candidate computing power nodes located at the edge or end side of the computing network from the candidate computing power node set as the target computing power nodes; in the case where the task type of the highway detection task is non-delay-sensitive and non-interactive, determining the candidate computing power nodes located at the cloud side of the computing network from the candidate computing power node set as the target computing power nodes.

[0046] Delay-sensitive highway detection tasks such as high-speed accident monitoring, and interactive highway detection tasks such as high-speed situation Q&A. For these two types of highway detection tasks, the results need to be fed back to the user in a timely manner. Since the edge and end sides in the computing network, especially the end side, are closer to the user and have low transmission delays, and the computing power nodes on the edge and end sides are both deployed with a large model and an AI video detection model fusion architecture, while improving the task processing efficiency, the transmission delay is low, greatly meeting the user's task processing requirements.

[0047] In some embodiments, when the task type of the highway detection task is delay-sensitive or interactive, and there are no candidate computing power nodes on the edge or end side in the candidate computing power node set, all the computing power nodes on the edge or end side are traversed, and a computing power node that is executing a non-delay-sensitive or non-interactive task is selected as the target computing power node, and the non-delay-sensitive or non-interactive task being executed is forwarded to a computing power node on the cloud side or another computing power node with a computing power performance lower than that of the target computing power node. In this embodiment, the task migration is executed by the computing network scheduling component, and the computing power resources sink to the edge and end sides, giving priority to processing delay-sensitive or interactive highway detection tasks, giving priority to ensuring the key business demand scenarios of users, achieving the maximization of user service quality under the premise of ensuring data transmission and security management requirements during busy hours, and effectively improving the overall service quality.

[0048] Furthermore, to avoid frequent task migrations, the computing and network scheduling component can reduce resource contention relationships in aspects such as computing, network, IO, and storage between non-latency-sensitive tasks and latency-sensitive tasks such as high-speed traffic event detection through the integrated scheduling of computing power resources and application tasks. It can be understood that when scheduling highway detection tasks, multiple highway detection tasks are scheduled to a single computing power node when the resource status of the computing power node meets the resource requirements of the tasks, so as to improve the resource utilization rate of a single computing power node as much as possible and leave more edge-side computing power nodes to support latency-sensitive tasks.

[0049] In some embodiments, determining at least one target computing power node from the set of candidate computing power nodes based on the task type of the highway detection task includes: dividing the highway detection task into multiple subtasks; and determining the target computing power node corresponding to each subtask from the set of candidate computing power nodes based on the task type of the subtask.

[0050] Specifically, a relatively large highway detection task usually includes multiple subtasks. In this embodiment, the computing and network scheduling component can determine the best forwarding path for each service and different target computing power nodes, and schedule different subtasks to different target computing power nodes for execution. In particular, each subtask is scheduled to target computing power nodes at different levels, achieving efficient coordination of multi-level computing power resources and further improving the execution efficiency of highway detection tasks.

[0051] In some embodiments, the computing and network scheduling method for highway detection tasks further includes: during the low-demand period of highway detection tasks, releasing some of the highway detection tasks executed on edge or terminal computing power nodes, and the released computing power nodes are used to execute non-latency-sensitive or non-interactive tasks, achieving the improvement of the resource utilization rate on the supply side (i.e., the edge side) under the condition of ensuring the quality of all services during idle time. For example, in the high-speed video surveillance scenario, there is a large amount of static data in the night video, and the resource demand for the terminal side is significantly lower than that during the day. Therefore, some video surveillance tasks can be shut down at night, and the computing power nodes used for video surveillance can be reallocated to deploy non-real-time batch processing service applications such as OCR and text processing, effectively improving the utilization rate of edge-side computing power resources.

[0052] In some embodiments, as Figure 2 and 3 shown, the target computing power node executes the highway detection task based on a large model and an AI video detection model, and returns the execution result to the upper-layer application, including the following steps S310 to S330.

[0053] Step S310: Extract the user prompt words in the highway detection task, and input the user prompt words into the large model, so that the large model automatically identifies and diverts the monitoring video stream according to the user prompt words, and intercepts according to the user prompt words to obtain video segments that conform to the scenarios corresponding to the user prompt words. Among them, the scenarios include: scenarios of tasks such as congestion, reverse driving, and road surface cracks.

[0054] Specifically, the highway detection task contains user prompt words for task description. For example: Check which section of the highway is congested. The large model determines that the highway detection task is a congestion detection scenario according to the prompt words, diverts the current monitoring video of the upper-layer application, and intercepts the video segments or images where congestion occurs.

[0055] Step S320: Send the video segment to the AI video detection model corresponding to the scenario to obtain the detection result. For example: For the congestion detection scenario, allocate the video segments or images where congestion occurs to the AI video detection model for detecting congestion, and obtain detection results such as the degree of congestion and the reasons for congestion.

[0056] Step S330: Input the detection result and the user prompt words into the large model, so that the large model generates a task report according to the detection result and the user prompt words, and returns the task report as the execution result to the upper-layer application. In this step, the large model's video analysis ability is used to generate a comprehensive report, which effectively improves the automation level of highway detection task execution while ensuring the processing accuracy.

[0057] In this embodiment, by combining the multi-scenario generalization ability of the vision large model and the scene-based high-precision characteristics of the traditional AI video detection model, the efficient execution of highway detection tasks is achieved, and high-precision execution results are obtained.

[0058] In this embodiment, the large model mainly adopts technologies such as P-Tuning and LoRA, and uses a distributed training method for model fine-tuning. Since the number of fine-tuning scenario samples is relatively small and the model size is large, the model parallel method is mainly used for training to realize the construction of a large model for high-speed scenarios. Deploy the fine-tuned professional large model to the computing and networking nodes of the computing and networking, dock with the multi-modal model, and open the API interface to the outside world to dock with the high-speed business system in the computing and networking nodes to realize the application of the large model.

[0059] Furthermore, by guiding lightweight small models to imitate the behavior of more powerful large models, the knowledge of large models is transferred to small models, enabling small models to learn the soft classification information of large models, thereby approaching the performance of large models, effectively reducing the resource requirements of the models, expanding the model deployment scope, and accelerating the model inference speed to achieve a rapid response to high-speed services. Deploying the lightweight small models on the edge side, through the collaborative optimization of "small models on the edge side + large models on the cloud", cross-cloud / edge and heterogeneous computing power deployment of large and small models is achieved. Multiple-step inference of small models is carried out on the edge and end sides, and the inference results are sent to the large models on the cloud side for optimization, so as to improve the inference efficiency and quality of the models, thereby optimizing the efficiency and response speed of video analysis, quickly responding to traffic events, and improving the timeliness of traffic management.

[0060] For the AI video detection model, through the "detection - tracking" combined framework, capabilities such as the detection, tracking of multiple individuals and multiple types of targets, and the recognition and analysis of complex traffic events are realized. Specifically, it includes two major functions: video detection and video tracking.

[0061] For the video detection function, by extracting frames from the video stream and performing operations such as normalization, scaling, rotation, and cropping on the image data to achieve data augmentation and adapt to the input of the recognition model, neural network models such as YOLO can be used to extract the feature information in the image, convert the spatial information of the objects in the image into vector information, and realize the detection and recognition of targets.

[0062] For the video tracking function, by adopting target tracking algorithms, extracting and matching the features of the detected objects, judging whether the objects in different frames of the video are the same target, and further analyzing the movement trajectory and direction of the target, the detection, tracking of multiple individuals and multiple types of targets, and the detection of complex traffic events in high-speed traffic scenarios are realized.

[0063] Next, the network computing scheduling device for highway detection tasks provided by the present invention will be described. The network computing scheduling device for highway detection tasks described below can be mutually referred to corresponding to the network computing scheduling method for highway detection tasks described above.

[0064] The network computing scheduling device for highway detection tasks in the embodiments of the present invention, as Figure 4 shown, includes the following modules 410 to 430.

[0065] The task acquisition module 410 is used to acquire the highway detection tasks issued by the upper-layer application.

[0066] The network computing perception module 420 is used to perceive the current computing power resource status of each computing power node in the network computing.

[0067] The computing and network scheduling module 430 is configured to extract the resource requirements of the highway detection task, determine at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node based on the task type of the highway detection task, and schedule the highway detection task to the target computing power node for execution.

[0068] Among them, each computing power node in the computing and network is deployed with a large model and an AI video detection model for executing different detection tasks. The target computing power node executes the highway detection task based on the large model and the AI video detection model, and returns the execution result to the upper-layer application.

[0069] The computing and network scheduling device for highway detection tasks in this embodiment determines at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node according to the task type of the highway detection task, schedules the highway detection task to the target computing power node for execution, and each computing power node is deployed with a large model and an AI video detection model for executing different detection tasks. The target computing power node executes the highway detection task based on the large model and the AI video detection model, improving the processing efficiency of highway detection tasks, reducing the processing pressure on data related to highway detection, and achieving the effect of carrying out the entire process of highway monitoring and management.

[0070] In some embodiments, the computing and network scheduling module 430 is specifically configured to, when the task type of the highway detection task is time-delay sensitive or interactive, determine at least one computing power node located at the edge or end side of the computing and network and whose computing power resource status meets the resource requirements as the target computing power node; when the task type of the highway detection task is non-time-delay sensitive and non-interactive, determine at least one computing power node located at the cloud side of the computing and network and whose computing power resource status meets the resource requirements as the target computing power node.

[0071] In some embodiments, the computing and network scheduling module 430 is specifically configured to, when the task type of the highway detection task is time-delay sensitive or interactive and there is no computing power node at the edge or end side whose computing power resource status meets the resource requirements, traverse all computing power nodes at the edge or end side, select a computing power node that is executing a non-time-delay sensitive or non-interactive task as the target computing power node, and forward the non-time-delay sensitive or non-interactive task being executed to a computing power node at the cloud side or another computing power node with computing power performance lower than that of the target computing power node.

[0072] In some embodiments, the computing and network scheduling module 430 is specifically configured to divide the highway detection task into multiple subtasks; determine at least one computing power node whose computing power resource status meets the resource requirements of the subtask as the target computing power node based on the task type of the subtask.

[0073] In some embodiments, the computing network scheduling device for highway detection tasks further includes: a task release module, configured to release highway detection tasks executed on some edge-side or end-side computing power nodes during the low-demand period of highway detection tasks, and the released computing power nodes are used to execute non-delay-sensitive or non-interactive tasks.

[0074] In some embodiments, the target computing power node executes the highway detection task based on a large model and an AI video detection model, and returns the execution result to the upper-layer application, including: Extract the user prompt words in the highway detection task, input the user prompt words into the large model, so that the large model automatically identifies and diverts the monitoring video stream according to the user prompt words, and intercepts according to the user prompt words to obtain a video segment that conforms to the scene corresponding to the user prompt words.

[0075] Send the video segment to the AI video detection model corresponding to the scene to obtain a detection result.

[0076] Input the detection result and the user prompt words into the large model, so that the large model generates a task report according to the detection result and the user prompt words, and returns the task report as the execution result to the upper-layer application.

[0077] Figure 5 An example of the physical structure diagram of an electronic device is shown as Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530, and a communication bus 540. Among them, the processor 510, the communication interface 520, and the memory 530 complete mutual communication through the communication bus 540. The processor 510 can call the logical instructions in the memory 530 to execute the computing network scheduling method for highway detection tasks, and this method includes the following steps: Obtain the highway detection task issued by the upper-layer application.

[0078] Perceive the current computing power resource status of each computing power node in the computing network.

[0079] Extract the resource requirements of the highway detection task, and based on the task type of the highway detection task, determine at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node, and schedule the highway detection task to the target computing power node for execution.

[0080] Among them, each computing power node in the computing network is deployed with a large model and an AI video detection model for performing different detection tasks. The target computing power node executes the road detection task based on the large model and the AI video detection model, and returns the execution result to the upper-layer application.

[0081] In addition, when the logical instructions in the above-mentioned memory 530 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0082] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the computing network scheduling method for road detection tasks provided by the above-mentioned various methods. The method includes the following steps: Obtain the road detection task issued by the upper-layer application.

[0083] Perceive the current computing power resource status of each computing power node in the computing network.

[0084] Extract the resource requirements of the road detection task. Based on the task type of the road detection task, determine at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node, and schedule the road detection task to the target computing power node for execution.

[0085] Among them, each computing power node in the computing network is deployed with a large model and an AI video detection model for performing different detection tasks. The target computing power node executes the road detection task based on the large model and the AI video detection model, and returns the execution result to the upper-layer application.

[0086] On yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the computing network scheduling method for road detection tasks provided by the above-mentioned various methods. The method includes the following steps: Obtain the road detection tasks sent by the upper-layer application.

[0087] Perceive the current computing power resource status of each computing power node in the computing network.

[0088] Extract the resource requirements of the road detection task, and based on the task type of the road detection task, determine at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node, and schedule the road detection task to the target computing power node for execution.

[0089] Among them, each computing power node in the computing network is deployed with a large model and an AI video detection model for performing different detection tasks. The target computing power node executes the road detection task based on the large model and the AI video detection model, and returns the execution result to the upper-layer application.

[0090] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0091] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solutions, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0092] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A computing network scheduling method for highway detection tasks, characterized in that, Including: Obtain the road detection task issued by the upper-layer application; Perceive the current computing power resource status of each computing power node in the computing network; Extract the resource requirements of the road detection task, and based on the task type of the road detection task, determine at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node, and schedule the road detection task to the target computing power node for execution; Among them, each computing power node in the computing network is deployed with a large model and an AI video detection model for executing different detection tasks. The target computing power node executes the road detection task based on the large model and the AI video detection model, and returns the execution result to the upper-layer application.

2. The network computing scheduling method for highway detection tasks according to claim 1, wherein Based on the task type of the road detection task, determining at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node includes: In the case where the task type of the road detection task is delay-sensitive or interactive, determine at least one computing power node located at the edge or end of the computing network and whose computing power resource status meets the resource requirements as the target computing power node; In the case where the task type of the road detection task is non-delay-sensitive and non-interactive, determine at least one computing power node located on the cloud side of the computing network and whose computing power resource status meets the resource requirements as the target computing power node.

3. The computing and network scheduling method for highway detection tasks according to claim 2, wherein, In the case where the task type of the road detection task is delay-sensitive or interactive and there are no computing power nodes on the edge or end whose computing power resource status meets the resource requirements, traverse all the computing power nodes on the edge or end, select a computing power node that is executing a non-delay-sensitive or non-interactive task as the target computing power node, and forward the non-delay-sensitive or non-interactive task being executed to a computing power node on the cloud side or another computing power node with a computing power performance lower than that of the target computing power node.

4. The computing and network scheduling method for highway detection tasks according to claim 1, wherein Based on the task type of the road detection task, determining at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node includes: Divide the road detection task into multiple subtasks; Based on the task type of the subtask, determine at least one computing power node whose computing power resource status meets the resource requirements of the subtask as the target computing power node.

5. The computing and network scheduling method for highway detection tasks according to claim 1, wherein It also includes: During the low-demand period of the road detection task volume, release some of the road detection tasks executed on the edge or end computing power nodes, and the released computing power nodes are used to execute non-delay-sensitive or non-interactive tasks.

6. The computing network scheduling method for highway detection tasks according to any one of claims 1 to 5, characterized in that The target computing power node executes the road detection task based on the large model and the AI video detection model, and returns the execution result to the upper-layer application, including: Extract the user prompt words in the road detection task, input the user prompt words into the large model, so that the large model automatically identifies and diverts the monitoring video stream according to the user prompt words, and intercepts according to the user prompt words to obtain video segments that conform to the scenes corresponding to the user prompt words; Send the video segments to the AI video detection model corresponding to the scene to obtain the detection results; Input the detection result and the user prompt into the large model, so that the large model generates a task report based on the detection result and the user prompt, and returns the task report as the execution result to the upper-layer application.

7. A computing network scheduling device for highway detection tasks, characterized in that, It includes: A task acquisition module, configured to acquire a road detection task issued by an upper-layer application; A computing network perception module, configured to perceive the current computing resource status of each computing power node in the computing network; A computing network scheduling module, configured to extract the resource requirements of the road detection task, and based on the task type of the road detection task, determine at least one computing power node whose computing power resource status meets the resource requirements as the target computing power node, and schedule the road detection task to the target computing power node for execution; Wherein, a large model and an AI video detection model for executing different detection tasks are deployed on each computing power node in the computing network, and the target computing power node executes the road detection task based on the large model and the AI video detection model, and returns the execution result to the upper-layer application.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the computing network scheduling method for road detection tasks according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the computing network scheduling method for road detection tasks according to any one of claims 1 to 6.

10. A computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the computing network scheduling method for road detection tasks according to any one of claims 1 to 6.

Images