Low-Latency Delivery of Adaptive Video Streaming Services

Today, service providers such as Netflix and YouTube generate more than 75% of the data transferred over the Internet, a number which is expected to grow in the near future. To deliver video and other media efficiently, HTTP adaptive streaming (HAS) is generally used. In HAS, the video is temporally segmented and encoded at different quality representations. Based on the network characteristics and the user preferences, the client can decide upon the most appropriate quality of each of these segments. Because the quality can be adapted, the video can typically be played out without interruptions.

In this dissertation, multiple optimizations for HAS are presented, focusing on different use cases: traditional video, personalized news content, virtual reality and volumetric media. In all use cases, we attempt to limit the latency for video delivery and grant the user a pleasant user experience. In the context of virtual reality, solutions are proposed which adapt the video quality to the userU+2019s movement, and allow the user to walk around in virtual scenes which were captured by a large number of cameras.