Performance Analysis of Large-scale Realtime Video Stream Configurations

Abstract

The rapid technology development has resulted in sensors able to deliver very high-quality output. The high quality calls for efficient compression algorithms to handle the vast amount of data produced. This study aims to find a compression algorithm that delivers video streams of the highest possible quality given the constraints real-time processing using the User Datagram Protocol (UDP). This paper describes the experimental approach created to find such compression algorithm. Machine learning in the form of Bayesian optimization was applied to evaluate and hence deduce the optimal encoder parameters for each encoder and resolution in scope. WebM Project’s VP9 implementation proved to be the most optimal encoder in scope for all resolutions evaluated in the experiment but the highest (QXGA - 2048x1536). For video streams in QXGA, VP9 hardware accelerated by Intel’s QuickSync was found to perform best.