The Study Of Streaming Optimization Technology For Panoramic Videos

With the rapid development of network technology,the demands of consumers for the video industry are gradually increasing,especially for the virtual reality technology and panoramic video technology which are developed recently.The panoramic video contains all the surrounding scene contents within 360-degree angles,and the users can watch the whole videos by simply switching the viewing angle,which providing the users with an immersive experience.However,a whole panoramic video includes a huge amount of data,so the transmission of videos is limited by the current network bandwidth and videos need to be fragmented in both time and space for efficient transmission.Transmission of traditional videos mostly uses HTTP-based adaptive streaming protocol to segment the videos in time,especially MPEG-DASH protocol,which has become an international standard and is supported and applied by many major companies.So this paper uses the DASH protocol to encapsulate the panoramic videos and slices videos in time.At the same time,since the user cannot watch the entire 360-degree video in space at a certain moment,the video is divided into multiple tiles,which improves the network utilization greatly.Based on this scheme,this paper builds a panoramic video transmission system,including video capture and packaging,server construction,and web player implementation.The player can not only support the playback of multi-tile DASH panoramic videos,but also can play videos continuously when changing the angle of view,laying a foundation for subsequent research work.Based on the transmission system,this paper applies the HTTP/2.0 protocol to optimize the startup delay and transmission delay.HTTP/2.0 protocol has the new feature of pushing resources.During the video startup and transmission process,the server pushes requested segments and subsequent related resources to the player.Experiments have confirmed that the effect is significant in reducing the startup delay and transmission delay.Using the communication mechanism of WebSocket,the player transmits the predicted view message to the server,and the server pushes segments of the predicted view in advance.This scheme improves the transmission performance of the system finally.In order to maximize the user's experience including bitrate,rebuffer time and video quality smoothness,the player selects the videos in most suitable bitrate for requesting and downloading adaptively according to the current network bandwidth.In this paper,three traditional adaptive algorithms based on buffer,based on bitrate and based on target-buffer and bandwidth both are implemented,and deep reinforcement learning is applied to panoramic videos.The four algorithms are compared under the same environment and experiments show that the RL algorithm can make the users' experience better under various network conditions.