Video Surveillance System Based On Edge Computing

Edge computing is a new computing system and technology that sinks computing power from the cloud to the edge of the network.Which could achieve real-time business,efficient data processing,application intelligence,security and privacy protection.Edge computing can make traditional video surveillance application scenarios no longer limited by network bandwidth transmission capabilities,enabling real-time judgment and analysis.The works of this article is as follows:(1)This article has independently built a video surveillance system under edge computing,integrated and proposed a complete video analysis processing framework under edge computing,including video stream acquisition,video frame filtering algorithm filter layer,video analysis layer arranged using Docker,and front-end technology rendering layer.The video analysis layer uses containers to make each functional module(face recognition)independent of the environment and independent of each other,which is convenient for monitoring and management.Experiments show that the algorithm proposed by the filter layer can significantly reduce the analysis processing time and average processing delay.(2)CNN face recognition model was built with Tensorflow,dlib,etc.Through continuous adjustment of CNN parameters,the experiment obtained the learning rate and single training sample capacity value that achieved the best balance of CNN performance such as recognition rate,total training time,and loss entropy.(3)An optimized model of CNN cooperative training is proposed.It consists of an edge server and multiple intelligent video terminals.The training set is first sent to the video terminal.The edge server performs part of the CNN.The rest of the CNN propagation is completed by the video terminal.The edge server updates the parameters after receiving the feedback and performs next spread again.The model adaptively divides the CNN between the video terminal and the edge server according to the available bandwidth.The experiment shows that the training performance is not disturbed by the edge server performance,and the total training time is saved by more than 10 times,which effectively alleviates a negative impact on training performance due to server performance and network bandwidth.(4)The collaborative computing optimization strategy is proposed,and the specific video data resource allocation scheme and matrix model are formulated.The detailed steps of collaborative computing optimization of CNN face recognition under edge computing are given.From a mathematical point of view shows that the optimization can improve the processing speed and the utilization efficiency of computing resources of many video terminals.(5)An optimization model for edge node scheduling is proposed,and a task control center is established that matches node priorities according to node status,assigns resources to specific task request nodes and uploads them first.A task scheduling optimization algorithm based on edge node scheduler is proposed.The comparison of data examples and experimental results shows that the algorithm can improve the accuracy of face recognition by 10%.In the case of poor network bandwidth,the face recognition rate of the algorithm is also maintained at more than 80%,which can meet the video data analysis and processing requirements of the video surveillance system under edge computing.