Edge AI computing cluster based on micro-container cloud

Abstract

The invention discloses an edge AI computing cluster based on edge micro-container cloud. The cluster is applied to the technical field of edge computing, artificial intelligence, cloud computing and5G communication, and aims at solving the problem that in the prior art, the real-time comprehensive analysis requirements of city management for multi-mode, full-time-period, full-target, full-background, full-tracking, full-cognition and whole-domain (whole city) data of the data are difficult to meet. According to the method, by constructing the edge micro-container cluster and the edge AI computing cluster, the computing tasks are automatically split and combined, so that computing resources of the edge cluster are utilized to the maximum extent. Meanwhile, the task execution efficiency isgreatly improved in a parallel computing mode, and centralized training and distributed deployment of the model can be realized at the total cloud through a multi-level cloud architecture.

Description

technical field [0001] The invention belongs to the technical fields of edge computing, artificial intelligence, cloud computing and 5G communication, and in particular relates to a construction technology of an edge AI computing cluster based on a micro-container cloud. Background technique [0002] As safe cities, smart transportation, and city brains have higher and higher requirements for urban intelligent management, surveillance cameras and other sensing devices are deployed more and more densely at the edge of cities. Therefore, how to monitor these edge data in real time Computing and intelligent analysis, especially the real-time detection, identification, tracking and recognition of moving objects appearing in surveillance video, has become an urgent need to realize the refined, intelligent, precise and agile management of cities. [0003] Currently, computing and processing of edge data mainly includes the following methods: [0004] One is to use edge acquisitio...

AI Technical Summary

Problems solved by technology

For example, using induction coils to take pictures of motor vehicles running red lights, using radar to detect urban pipelines, etc. However, this method usually has the following problems: First, the hardware configuration and computing power of proprietary devices are usually fixed and limited. When the computing power provided by the hardware When the capacity cannot meet the needs of edge computing, new equipment must be replaced, resulting in waste of resources and increased costs; secondly, this kind of proprietary equipment usually presets the algorithm model in advance, and for the algorithm model that cannot be preset in advance, It will be difficult to achieve fast and real-time update of the algorithm model; the third is that such proprietary equipment is usually closed and not connected to the Internet, so it will be difficult to achieve real-time sharing, fusion and comparative analysis of data for cross-collection and cross-device data analysis

[0005] The second is to use edge-specific computing equipment. Although the edge adopts multi-core ARM processors, multi-core GPUs, multi-core DSPs, etc., and has certain computing capabilities, but because it is still a single device, when the edge handles complex tasks or the calculation scale expands, This kind of computing power that only relies on a single device will also face the situation of insufficient computing power

[0006] The third is to use microcomputer equipment (such as Raspberry Pi) as the edge computing device, place it in the computer room, and one acquisition device corresponds to one edge computing device. This method usually increases the data transmission delay, and when the edge needs to When processing complex computing tasks in real time, a single edge computing device will not be able to provide sufficient computing resources and computing power to meet real-time processing requirements

[0008] Based on the above current methods of edge computing, limited by the processing capability of a single edge device, edge devices either use proprietary devices to process specific data, or use proprietary computing devices, or use micro-devices / terminals, which have insufficient computing power It is difficult to flexibly expand, the acquisition function and type of equipment are single, iterative update of artificial intelligence models is difficult or costly, multi-modal data lacks fusion analysis, and the lack of unified management and control of multi-level cloud computing will make it difficult to meet urban management. Real-time comprehensive analysis requirements for multi-modal, full-time, full-target, full-background, full-tracking, full-cognition and global (whole city) data

Images