Research On Video Service Model Deployment And Adaptive Configuration Algorithm For Edge Computing

With the emergence of intelligent application services that rely on real-time video processing technology,such as intelligent monitoring and virtual reality,and the rapid development of deep learning in the field of computer vision,intelligent video services have greatly enriched people’s daily life.However,limited by the hardware configuration of mobile terminals,the user devices are often unable to deploy and run the video service models for video processing.By using edge computing technology,the video service models are deployed on the edge cloud close to user devices to directly process the video streaming requests of user devices,which can effectively reduce the network delay and solve the problem of insufficient computing resources of user devices.Since the computing resources and bandwidth resources of the edge cloud are limited,how to use the limited resources to improve the video service quality of the edge cloud has become an urgent problem to be solved.First of all,in order to reasonably utilize the computing resources and bandwidth resources of the edge cloud,a video service deployment model based on the edge cloud is established by analyzing the video processing delay and video processing accuracy.Considering the heterogeneity and unpredictability of video service requests of user devices,a video service model deployment algorithm based on Multi-Armed Bandit is proposed.In this algorithm,the total time delay and accuracy of the system are weighted and summed according to the user’s demand of video service,which are taken as the measurement values and the optimization targets.The algorithm calculates the historical observation values of the video service models based on the measurement values,and selects multiple video service models with smaller historical observation values for deployment.The experimental results show that compared with the traditional algorithm,the learning regret value of the proposed algorithm is reduced by 36%,the proposed algorithm implements the more efficient video service model deployment strategies.Secondly,in order to reduce the delay of edge cloud video service and improve the average accuracy,a video service adaptive configuration system model for edge cloud is established for the configuration of frame sampling rate and resolution of video recorded by user devices.Based on deep deterministic policy gradient reinforcement learning,an adaptive video service configuration algorithm for edge cloud is proposed.The algorithm selects the appropriate video service models and gives the best video parameter configuration for user devices.In this algorithm,video service model,frame sampling rate and resolution are taken as the output actions of the agent,and the total delay and accuracy of the system are weighted according to the user’s video service requests,witch is taken as the reward value.The experimental results show that,compared with the existing algorithms,the average processing delay of the proposed algorithm is reduced by 87 ms,and the average accuracy of the proposed algorithm is increased by 6%,which effectively improves the quality of edge cloud video services.