VHAS: Video Highlight based Adaptive Streaming
Hangyeol Hong, Hye-Rin Kim, Chong-kwon Kim · 2022 International Conference on Information Networking (ICOIN) · 2022
As video streaming has become one of the most important services on the Internet, exploration of efficient video transmission methods have attracted vast research attractions during last several years. Existing ABR (Adaptive Bit Rate) algorithms assign the same bitrate as much as possible, assuming that the user watches all video scenes with the same attention and interest. However, because users engage more in highlight scenes than other parts of videos, the playback quality of the highlights should be higher than that of non-highlights. In this paper, we propose a highlight-based bitrate adaptation scheme called VHAS, Video Highlight based Adaptive Streaming. VHAS first extracts a highlight score per frame using graph neural network and object detection techniques. VHAS employs a DRL (Deep Reinforcement Learning) technique to learn the optimal policy from an incomprehensibly complicated operating environment. VHAS incorporates the extracted highlight scores into the reward model and decides the bitrate of video chunks based on their highlight scores. Performance experiments on real video streaming service environments show that VHAS can adaptively provide bitrate according to the highlight scores of chunks. In our study, the average QoE of VHAS is about 56%- 150% higher than that of state-of-the-art baselines.