Deep Reinforcement Learning Based VR Streaming

With the increasing proportion of video content in Internet traffic,it will be a hot topic for academic and industrial circles to improve the experience of users watching video.Users are affected by the network environment when watching video,so how to improve the viewing experience of users in complex environment is extremely important to ensure the quality of video content service.As a classical solution,video adaptive bitrate algorithm has been widely used and deployed in real scenes.With the commercial deployment of 5G networks,video media forms like VR(Virtual Reality)video will be widely promoted.Therefore,it is still a great challenge to ensure the quality of VR video transmission.The adaptive bitrate algorithm based on deep reinforcement learning algorithm has excellent performance.After training,it can make the agents who choose the bitrate have the wisdom to analyze different network environments,so as to make the best choice of video bitrate.In this paper,based on the deep reinforcement learning algorithm,from the two aspects of ordinary video and VR video,the following research work is carried out:Firstly,this paper studies the hybrid adaptive bitrate algorithm based on deep reinforcement learning.A hybrid QoE(Quality of Experience)evaluation index is proposed,and the corresponding hybrid adaptive bitrate algorithm model is designed and trained.In traditional QoE and hybrid QoE,simulation results show that the performance of hybrid bit rate adaptive algorithm is better.Then,the adaptive bitrate algorithm of VR based on deep reinforcement learning is studied.This paper proposes the QoE evaluation index suitable for the panoramic video transmission strategy based on FOV(Field of View),compiles the virtual VR playing environment,and realizes the VR bit rate adaptive algorithm based on deep reinforcement learning.It has better user experience and higher transmission efficiency than the traditional algorithm by combining bandwidth prediction with buffer control.