CN-121984842-A - Intelligent park distributed monitoring system and method based on heterogeneous calculation power flexible scheduling

CN121984842ACN 121984842 ACN121984842 ACN 121984842ACN-121984842-A

Abstract

The invention discloses an intelligent park distributed monitoring system and method based on heterogeneous computing power elastic scheduling, wherein the system is constructed by adopting a distributed architecture with separated central configuration scheduling nodes and heterogeneous computing power nodes, and comprises a central configuration scheduling node, a plurality of heterogeneous computing nodes and one or more AI model online reasoning units, wherein the central configuration scheduling node is communicated with the heterogeneous computing nodes through a network, the heterogeneous computing nodes are communicated with the AI model online reasoning units through a high-performance RPC protocol, the central configuration scheduling node is responsible for unified management task configuration and resource scheduling, the heterogeneous computing nodes are responsible for execution and state maintenance of task processes, and the AI model online reasoning units are responsible for model reasoning calculation, so that three-layer decoupling of computing tasks, resource scheduling and model reasoning is realized. The scheme solves the problems that the hardware configuration of the existing intelligent monitoring system is stiff, the utilization rate of computing power resources is low, and low-cost elastic expansion cannot be realized.

Inventors

Ping Xiangfan
ZHANG ZHONGLIANG
Gao Dihe

Assignees

启明信息技术股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260206

Claims (7)

1. The intelligent park distributed monitoring system based on heterogeneous computing power elastic scheduling is characterized by being constructed by adopting a distributed architecture with separated central configuration scheduling nodes and heterogeneous computing power nodes, and comprising the central configuration scheduling nodes, heterogeneous computing nodes and an AI model online reasoning unit; The central configuration scheduling node is in charge of unified management of task configuration and resource scheduling, the heterogeneous computing node is in charge of execution and state maintenance of task processes, and the AI model online reasoning unit conducts reasoning calculation on the model.
2. The intelligent campus distributed monitoring system based on heterogeneous computing power flexible scheduling according to claim 1, wherein the number of the central configuration scheduling nodes is one, the number of the heterogeneous computing power nodes is a plurality, and the number of the AI model online reasoning units is one or a plurality.
3. The intelligent park distributed monitoring system based on heterogeneous computing power flexible scheduling according to claim 2, wherein the central configuration scheduling node uniformly manages information of all monitoring task configurations, heterogeneous computing node resource states and an AI model online reasoning unit, and dynamically maintains a mapping relation between tasks and resources; If a node is found to be out of connection, the running task is automatically rescheduled to other healthy heterogeneous computing nodes with redundancy performance, so that the automatic fault transfer is realized; And the communication between the heterogeneous computing node and the AI model online reasoning unit adopts a high-performance RPC framework, so that low-delay and high-throughput model reasoning service call is realized.
4. The intelligent park distributed monitoring method based on heterogeneous computing power elastic scheduling is realized based on the intelligent park distributed monitoring system based on heterogeneous computing power elastic scheduling according to any one of claims 1-3, and is characterized by comprising the following steps: the method comprises the steps of S1, registering a heterogeneous computing node to a central configuration scheduling node after the heterogeneous computing node is started, and reporting the computing power attribute of the heterogeneous computing node; step S2, on a central configuration scheduling node, configuring one or more violation items to be monitored for each monitoring camera, and setting a required detection area and a required detection time plan; the central configuration scheduling node dynamically allocates a heterogeneous computing node to each monitoring task as a task executor according to a preset scheduling strategy, and binds an AI model online reasoning unit capable of providing a required algorithm model as a reasoning server; Step S3, the central configuration scheduling node issues a task starting instruction to the heterogeneous computing node selected in the step S2, and after the heterogeneous computing node receives the instruction, an independent monitoring process is started for each camera task; Step S4, the monitoring process pulls the video stream of the appointed camera and extracts the video frame as required, and the video frame needing intelligent analysis is subjected to remote reasoning by calling the AI model online reasoning unit appointed in the step S2 through a high-performance RPC protocol and the identification result is structured; Step S5, the central configuration dispatching node periodically sends a heartbeat request to each node, if no response is made, the node is marked as abnormal, and meanwhile, the monitoring process on the node is transferred; And S6, storing the recognized violation results to a central database, storing the pictures and the video clips to a local machine, and inquiring by a user through a Web front-end interface and simultaneously providing visual display and alarm functions.
5. The intelligent campus distributed monitoring method based on heterogeneous power resilient dispatch of claim 4, wherein the power attributes in step S1 include CPU architecture, memory, and whether to mount an acceleration unit.
6. The intelligent campus distributed monitoring method based on heterogeneous computing power flexible scheduling according to claim 5, wherein the task start instruction content in step S3 includes camera information, a bound AI model service address, a list of violations to be detected, and a time plan.
7. The intelligent campus distributed monitoring method based on heterogeneous power resilient dispatch of claim 4, wherein step S5 further comprises periodically checking the log of AI model services, and if a response delay is found, reducing cameras allocated by the AI model services.

Description

Intelligent park distributed monitoring system and method based on heterogeneous calculation power flexible scheduling Technical Field The invention relates to the technical field of intelligent monitoring and computers of intelligent parks, in particular to a distributed monitoring system and a distributed monitoring method of an intelligent park based on heterogeneous computing power elastic scheduling. Background Traditional garden security monitoring system highly relies on the security officer to observe the control picture for a long time, can only follow up through backtracking video after illegal event takes place and causes the influence. The method has the problems of untimely early warning, negligence, omission and the like caused by fatigue of personnel. With the development of computer vision technology, the real-time intelligent monitoring system based on artificial intelligence is gradually applied, so that automatic identification and instant alarm can be realized, and the security response efficiency is effectively improved. However, existing intelligent monitoring schemes still have significant drawbacks. One type of scheme employs a centralized data processing center for visual recognition analysis. The method needs to remotely transmit all video streams to a central machine room, has extremely high requirement on network bandwidth, is difficult to meet real-time processing delay requirement, needs to be provided with a high-performance GPU, is expensive in construction and operation and maintenance cost, and is not suitable for small and medium-sized parks lacking professional machine room facilities. Another class of schemes attempts to improve the above problem by a distributed architecture, employing edge compute nodes for local analysis, alleviating the central transmission pressure. However, each edge node still needs to be equipped with a graphics card to run an image recognition algorithm, so that the dependency on special graphics hardware cannot be fundamentally eliminated, the hardware configuration is stiff, the node cost is high, the devices with different computing power types cannot be flexibly incorporated for collaborative scheduling, and the application potential of the system in the aspects of low cost and high elastic expansion is limited. Therefore, aiming at the intelligent monitoring requirements of small and medium-sized parks on low cost, easy expansion and real-time reliability, an intelligent monitoring system architecture which can integrate heterogeneous computing resources, support the participation of nodes without display cards and truly realize the cooperation of elastic scheduling and distribution is needed. Disclosure of Invention Aiming at the technical problems, the invention provides an intelligent park distributed monitoring system and method based on heterogeneous computation power flexible scheduling, which are used for solving the problems of stiff hardware configuration, high cost and poor expansibility of the existing intelligent park intelligent monitoring system. The invention is realized by adopting the following technical scheme: The intelligent park distributed monitoring system based on heterogeneous computing power flexible scheduling is constructed by adopting a distributed architecture with separated central configuration scheduling nodes and heterogeneous computing power nodes, and comprises central configuration scheduling nodes, heterogeneous computing nodes and an AI model online reasoning unit; The central configuration scheduling node is in charge of unified management of task configuration and resource scheduling, the heterogeneous computing node is in charge of execution and state maintenance of task processes, and the AI model online reasoning unit conducts reasoning calculation on the model. Specifically, the number of the central configuration scheduling nodes is one, the number of the heterogeneous computing power nodes is a plurality of the heterogeneous computing power nodes, and the number of the AI model online reasoning units is one or a plurality of the AI model online reasoning units. Specifically, the central configuration scheduling node uniformly manages the configuration of all monitoring tasks, the resource state of heterogeneous computing nodes and the information of an AI model online reasoning unit, and dynamically maintains the mapping relation between tasks and resources; If a node is found to be out of connection, the running task is automatically rescheduled to other healthy heterogeneous computing nodes with redundancy performance, so that the automatic fault transfer is realized; And the communication between the heterogeneous computing node and the AI model online reasoning unit adopts a high-performance RPC framework, so that low-delay and high-throughput model reasoning service call is realized. On the other hand, the intelligent park distributed monitoring method based on heterogeneous computation po