Research On Video Analytics Based On Edge Computing And Positioning Application

Video analytics is the semantics of video data,from which to extract the content of interest.Today,a variety of emerging Internet of Things applications,such as virtual reality,augmented reality,autonomous driving,cognitive assistance,etc.,are supported by video analytics.The camera can be seen as a bionic to the human eye.Therefore,by analyzing the video data collected by the camera,various behaviors related to human vision can be reproduced,and the typical behavior is location information perception.With the vigorous development of computer vision and image processing disciplines,positioning technology based on video analytics has shown good development prospects.With the high-definition and large number of camera sensing devices,the amount of video data has increased by spurt.The traditional cloud computing method of uploading centralized processing brings huge traffic pressure to the core network,which can not meet the requirements of video analytics for real-time and network service quality.Edge Computing(EC)is a computing architecture that provides computing,network and storage resources to users on the edge of the network.It can effectively improve the user experience of delay-sensitive services and provide necessary support for video analytics.This paper uses the lightweight virtualization technology provided by the container to migrate the video analytics task to the edge of the network,and proposes a video analytics solution based on edge computing.Further,a positioning application system based on video analytics is designed and implemented.This article has mainly completed the following work:(1)Research on video analytics and edge computing was investigated.Through the research on the requirements of video analytics,this paper applies edge computing technology to video analytics,and proposes a video analytics scheme based on edge computing.And use the lightweight technology(Lightweight Virtualization(LV)provided by the container technology to flexibly offload the video analytics task to the edge computing node.In addition,considering that different video analytics tasks often have some identical processing procedures,this paper divides the video analytics tasks according to functions,and proposes a modular analytics model of video analytics.(2)A positioning scheme based on video analytics is proposed,including camera calibration and target feature point selection strategy.For camera calibration,a corner tile extraction method based on regular tiles is proposed to simplify the calibration process.Then,the proposed edge analytics based video analytics scheme is applied to indoor positioning,and a positioning application system is designed.Supported by the positioning scheme,the system combines video analytics techniques such as camera imaging,camera calibration,and face recognition.With the aid of the auxiliary information provided by the regular tiles laid in the indoor scene,the image is built without relying on third-party ranging tools.The one-to-one mapping of the real scenes completes the positioning.(3)Realizing a positioning system based on video analytics,including camera calibration,streaming service configuration,video analytics module construction and Web services.The system's web services provide video stream management,video stream viewing and location capabilities.Then the system's network service quality,extended migration performance and positioning accuracy were evaluated.Experiments show that the proposed positioning scheme has better positioning accuracy,and also proves the superiority of applying edge computing and lightweight virtualization technology to video analytics.