Abstract
From a video coding perspective, there are two challenges when performing live video distribution over error-prone networks, such as wireless networks: random access and packet loss repair. There is a scarceness of solutions that do not impact steady-state usage and users with reliable connections. The proposed solution minimizes this impact by complementing a compression-efficient video stream with a companion stream solely consisting of keyframes. Although the core idea is not new, this paper is the first work to provide restrictions and modifications necessary to make this idea work using the High-Efficiency Video Coding (H.265/HEVC) compression standard. Additionally, through thorough quantification, insight is provided on how to provide low-latency fast channel switching capabilities and error recovery at low quality impact, i.e., less than 0.94 average Video Multimethod Assessment Fusion (VMAF) score decrease. Finally, worst-case drift artifacts are described and visualized such that the reader gets an overall picture of using the keyframe insertion technique.
Highlights
Low-latency video distribution is a challenging domain because requiring a minimal buffer in all the components of the distribution chain drastically restricts the technological solutions on offer
We present and evaluates the experimental results concerning the proposed method based on keyframe insertion for H.265/HEVC
A rough estimate based on these numbers tells us that end user devices with reliable connections can enjoy a bitrate reduction in the range of 69% compared to classical streams with a keyframe period of one second and no companion stream configuration
Summary
Low-latency video distribution is a challenging domain because requiring a minimal buffer in all the components of the distribution chain drastically restricts the technological solutions on offer. By predicting blocks of pixels from one frame to the other, dependencies between successive frames emerge These dependencies provide enormous compression efficiency gains, they make the video stream susceptible for packet loss and random-access restrictions. The main idea is to separate compression efficiency from random access and error resilience such that individual channel switching or low performance does not influence the larger group of high-performing end users. This separation is reflected in the generation of two video streams: a normal stream (NS) and a companion stream (CS).
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
