| With the development of 5G and the Internet of Things,the amount of data in networks is showing an explosive growth trend,especially the data generated by monitoring devices at the network edge.In the traditional cloud computing model,video data from terminal devices on the edge is transmitted over the network to central servers for processing and storage.With the surge in network data volume,the network becomes overloaded and there is an increase in data transmission latency.To address this,this thesis proposes a decentralized edge computing scheme,which partially offloads the computing power of the cloud center to the edge,thereby reducing data flow in the network.The thesis is focused on solving the following problems with existing monitoring systems driven by actual application needs:1.Designed video processing algorithms for monitoring,including relevant algorithms for video processing on the terminal and edge side.The cloud-edge collaborative method is used to perform pre-processing operations on the edge side,report data signals to the cloud,and perform feature matching in the cloud to implement edge video surveillance applications.2.Designed scheduling algorithms for edge video processing services.The edge requires a large amount of video processing services to process video data collected by cameras,but the edge’s computing resources are limited.Therefore,these microservices need to be scheduled to appropriate edge servers,using limited edge resources to run more video processing services where possible while minimizing latency.3.Designed and implemented an edge computing platform,addressing Kubernetes’ s clustering only at the cloud and edge device access points.Unified message formats were designed for communication between the cloud center and the edge,taking into account the special network conditions of the edge and cloud.The camera devices on the edge were abstracted into microservices,which can provide video data and corresponding metadata information to the outside world,and the cloud can control the terminal devices through this module.The edge was made capable of offline autonomy.For storage operations,a storage module was implemented to notify the edge to encode,wrap,and store video data when the processing results of the monitoring application match predefined scenarios.A push component for video services was implemented for real-time video streams from cameras and stored video data.Based on the Kubernetes framework,this thesis designs and implements a video monitoring system based on edge computing,combining the crucial demands of monitoring scenarios in edge computing.The system solves problems such as high service latency and great network bandwidth pressure in the central cloud computing model.The cloud is responsible for integrating data generated at the edge and long-term storage of video data,while the edge is responsible for processing short-term storage of video data with AI,encoding,and wrapping techniques from video capture devices. |
