360-Degree Video Dynamic Adaptive Request Algorithm Based On Deep Reinforcement Learning And Engineering Platform Implementation

With the rapid development of video streaming technology and the popularity of 5G and other low-latency network communication techniques,the requirement of video streaming media has gradually developed from ordinary two-dimensional video to virtual reality(VR)and augment reality(AR)and other immersive video fields.360-degree video is one of the main application formats for immersive video,which is popular because users can freely switch perspectives to watch panoramic content and get immersive experience.However,due to the high frame rate and resolution of 360-degree video,resulting in a huge amount of data,and the existing bandwidth resources are limited,so how to effectively transmit is a bottleneck restricting the further development of 360-degree.To provide high quality of experience(QoE)for users,this paper proposed a deep reinforcement learning based dynamic adaptive streaming system.The system is composed of four platform:video splicing platform,video live platform,video-on-demand platform,and deep reinforcement learning based 360-degree video request strategy.First of all,a video splicing platform based on FFmpeg is implemented.The tile-based transmission is adopted for both live and on-demand 360-degree videos,and the videos need to be spliced before playing.Second,an encrypted video live streaming platform based on Nginx-RTMP is realized.In this platform,the 3DES encryption algorithm is used to encrypt the naked video live streaming.After that,the player on-demand platform based on DASH is implemented.Based on the DASH transmission framework of 2D video,a 360-degree video transmission platform based on Tile was developed.The platform can call the video adaptive request algorithm developed by C++or Python,and add local 2D video and 360-degree video palyback,local video encryption and decryption,web version DASH adaptive request playback,and DASH adaptive request playback function embedded in the platform.Finally,according to the deep reinforcement learning based bitrate adaptive request algorithm we realized a bitrate adaptive request algorithm for 360-degree video,and applied it to 360-degree video on demand platform based on DASH.